CA2167927A1 - Virtual emissions monitor for automobile - Google Patents
Virtual emissions monitor for automobileInfo
- Publication number
- CA2167927A1 CA2167927A1 CA002167927A CA2167927A CA2167927A1 CA 2167927 A1 CA2167927 A1 CA 2167927A1 CA 002167927 A CA002167927 A CA 002167927A CA 2167927 A CA2167927 A CA 2167927A CA 2167927 A1 CA2167927 A1 CA 2167927A1
- Authority
- CA
- Canada
- Prior art keywords
- emissions
- output
- sensor
- predicted
- values
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Abandoned
Links
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
- G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
- G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
- G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
- G05B13/048—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators using a predictor
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F02—COMBUSTION ENGINES; HOT-GAS OR COMBUSTION-PRODUCT ENGINE PLANTS
- F02D—CONTROLLING COMBUSTION ENGINES
- F02D41/00—Electrical control of supply of combustible mixture or its constituents
- F02D41/02—Circuit arrangements for generating control signals
- F02D41/14—Introducing closed-loop corrections
- F02D41/1401—Introducing closed-loop corrections characterised by the control or regulation method
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N33/00—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
- G01N33/0004—Gaseous mixtures, e.g. polluted air
- G01N33/0009—General constructional details of gas analysers, e.g. portable test equipment
- G01N33/0073—Control unit therefor
- G01N33/0075—Control unit therefor for multiple spatially distributed sensors, e.g. for environmental monitoring
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
- G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
- G05B13/0265—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric the criterion being a learning criterion
- G05B13/027—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric the criterion being a learning criterion using neural networks only
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B17/00—Systems involving the use of models or simulators of said systems
- G05B17/02—Systems involving the use of models or simulators of said systems electric
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B23/00—Testing or monitoring of control systems or parts thereof
- G05B23/02—Electric testing or monitoring
- G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
- G05B23/0218—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults
- G05B23/0243—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults model based detection method, e.g. first-principles knowledge model
- G05B23/0254—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults model based detection method, e.g. first-principles knowledge model based on a quantitative model, e.g. mathematical relationships between inputs and outputs; functions: observer, Kalman filter, residual calculation, Neural Networks
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B23/00—Testing or monitoring of control systems or parts thereof
- G05B23/02—Electric testing or monitoring
- G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
- G05B23/0259—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterized by the response to fault detection
- G05B23/0267—Fault communication, e.g. human machine interface [HMI]
- G05B23/027—Alarm generation, e.g. communication protocol; Forms of alarm
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F02—COMBUSTION ENGINES; HOT-GAS OR COMBUSTION-PRODUCT ENGINE PLANTS
- F02D—CONTROLLING COMBUSTION ENGINES
- F02D41/00—Electrical control of supply of combustible mixture or its constituents
- F02D41/02—Circuit arrangements for generating control signals
- F02D41/14—Introducing closed-loop corrections
- F02D41/1401—Introducing closed-loop corrections characterised by the control or regulation method
- F02D2041/1433—Introducing closed-loop corrections characterised by the control or regulation method using a model or simulation of the system
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N33/00—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
- G01N33/0004—Gaseous mixtures, e.g. polluted air
- G01N33/0009—General constructional details of gas analysers, e.g. portable test equipment
- G01N33/0027—General constructional details of gas analysers, e.g. portable test equipment concerning the detector
- G01N33/0031—General constructional details of gas analysers, e.g. portable test equipment concerning the detector comprising two or more sensors, e.g. a sensor array
- G01N33/0034—General constructional details of gas analysers, e.g. portable test equipment concerning the detector comprising two or more sensors, e.g. a sensor array comprising neural networks or related mathematical techniques
Abstract
An internal combustion engine (360) is provided with a plurality of sensors to monitor the operation thereof with respect to various temperature measurements, pressure measurement, etc. A predictive model processor (322) is provided that utilizes model parameters stored in the memory (324) to predict from the sensor inputs a predicted emissions output. The model is strained with inputs provided by the sensor and an actual emissions sensor output. During operation, this predicted output on line (326) can be utilized to provide an alarm or to be stored in a history database in a memory (328). Additionally, the internal combustion engine (260) can have the predicted emissions output thereof peiodically checked to determine the accuracy of the model. This is effected by connecting the output of the engine to an external emissions sensor (310) and taking the difference between the actual output and the predicted output to provide an error. This is compared to a threshold and, if the error exceeds the threshold, the predictive model processor (322) can be placed in a training operation wherein new model parameters are generated. Additionally, retraining could take place external to the predictive model processor (322). During runtime, a control network (350) can be utilized to predict control parameters for minimizing the emissions output.
Description
2 1 6 7 q 2 7 PCT/US94/08657 .
VIRTUAL EMISSIONS MONITOR FOR AUTOMOBILE
TECIINICAL ~IELD OF THE INVENTION
The present invention pertains in general to emissions monitoring systems, and more particularly, to a system that replaces the continuous emissions monitor on a reciprocating engine with a virtual sensor.
CROSS REFERENCE TO RELATED APPLICATION
This application is a continuation-in-part of U.S. Patent Application Serial No. 08/102,405, filed August 5, 1993, and entitled "VIRTUAL CONTINUOUS
E~ISSION MONITORlNG SYSTEM WITH SENSOR VALIDATION" (Atty. Dkt.
No. PAVI 21,874).
wo 95/04878 2 1 6 7 9 2 7 PCT/US94/08657 BACKGROUND OF T~E INVENTION
As public awareness increases with respect to the environment, industry is required to make ~i nific~nt ch~nges Although industry is somewhat responsive topublic opinion, govell--llt;llL regulatory bodies are typically brought in to ensure that 5 public needs are met. In order to do this, government sets up regulatory arms of already eYi~ting branches of entities such as the En~ri.o"".~ l Protection Agency. These arms are given the task of putting in place policies regarding toxic waste, emissions, etc., that may effect the environment. Further, these regulatory bodies are also given the task of el~l~;ing these regulations. One particular area that has received a great deal of 10 attention in recent years is that of monitoring emissions of noxious gases being placed into the atmosphere by m~mlf~ctnring f7~t~.iliti~s Typically, the teçhniqlle for ensuring that noxious gases are being co-.e.;Lly m~ ol ed has been to implement Continuous F.mi~ ns Monitoring systems (CEM).
These systems are utilized to monitor the amount of emissions such as Sulfur Dioxide 15 (S2)~ Nitrogen Oxides (NOx), Carbon Monoxide (CO), Total reduced Sulfur (TRS), opacity, Volatile Organic Carbon (VOC), and hazardous subsLances of all sorts. The rl~c~ic~l way of monitoring these emissions is to install a Continuous Fmi~ions Monitor (CEM) in the plant on each emissions point source. Re~ tory Agencies provide foreach plant ~ elines as to how the output is to be re~ te~l, i.e., define the acceptable 20 limit ofthe emissions. With respect to a reciprocating engine, the EPA has set ~lid~lines as to the p~;-~----ance of the engine. Once m~mlf~ctllred, the engine is expected to meet these ~lid~linçs over the life ofthe engine, a~s~lming that it is plopell~ d However, new re~ tic)ns are being implem~nte~l that require some type of c~ntimlQus moniLo.i.~ to be implem~nted with periodic l~.h~c~lps to ensure that the monitor is 25 operating co-le-;Lly.
The classic CEM is composed of either an in situ analyzer installed directly in the stack, or the exhaust pipe of the reciprocating engine, or an extractive system which extracts a gas sample and conveys it to an analyzer at grade level. However, these sensors are quite expensive, difficult to m~int~in, and difficult to keep properly SU~S~U.~ SI~EE~ (R~LE 26) WO 95/04878 2 1 6 7 ~ 2 7 PCT/US94/08657 .
calibrated. As such, the reg~ tions that deal with a CEM system require the sensors to be calibrated frequently, which calibration procedure can take a number of hours, due to the complexity thereo Re~ tions allow a m~imllm downtime of ten percent for calibration. If a unit remains in operation greater than ten percent of the time with the 5 CEM down, the emissions level is con~idered by the Regulators to be at ~ al potential level. This results in out-of-compliance operation. Most m~nllf~ctllres will shut down operation rather than face the high penalties of such occurrence. One of the reasons for this is that the operation of the plant relative to the monitoring of the NOx emissions must be "truly continuous" such that no leeway is provided for faulty sensors, 10 sensors that have fallen out of calibration, etc. One solution to this has been to utilize redlln~n1 sensors, which is a very expensive solution. Therefore, there exists a need to provide a system that does not require the presence of a sensor while still ensuring that the output of the plant is within tolerances relative to noxious emissions.
T~TUT~ S~EET ~ULE ~
;.. ~
SUMMARY OF l~ INVENTION
The present invention disclosed and claimed herein comprises a method for monitoring emissions in an internal combustion engine that emits noxious pollutants and has associated ~I.el~;wiL}l a plurality of sensors for measuring select parameters of the 5 engine operation, these measured parameters provided as sensor output values. A
predictive model is provided in which a stored representation of the engine combined with an external emissions monitor is reprç,s~nted The predictive model is operable to output a predicted emissions value corresponding to the output of the external emissions monitor when ~tt~chPd to the engine, with the inputs to the predictive model 10 corresponding to select ones of the sensor output values. The select ones of the sensor output values are then input to the predictive model and a prediction is provided of the noxious pollutants during operation of the engine without the requirement of the external emissions monitor being connected to the engine. This is achieved during runtime-without a 1 ullLh~le monitor for the emissions.
In another aspect of the present invention, the operation of the predictive model is verified by ~tt~.hing an external emissions monitor to the engine to measure the noxious pollutants output thereby. The measured output is then co",pa, ~d with the predicted output. If the actual and predicted values differ by more than a predeterrnined amount, the stored represçnt~tit n is adjusted. The stored representation is generated 20 initially by training the predictive model on a training set of data that comprises as input data the actual sensor output values and as target output data the actual measured emissions output of the external emissions monitor. In one method of adjusting the stored l~s~ ;nn~ the predictive model is retrained to complete a new stored rep,es~ l;t n on a new set oftraining data. In another method, the predicted emissions 25 value is merely offset on the output of the predictive model such that the average difference between the offset predicted emissions value and the actual measured emissions value is less than a predetermined difference.
In yet another aspect of the present invention, the predicted emissions value output by the predictive n~lwol~ is colllpaled with an intemal threshold value. When the SUBSTITUT~ SH~ET (RULE 26) wo 95/04878 2 1 6 7 9 2 7 PCT/USg4/08657 predictive emissions value exceeds the internal threshold value, an alarm is generated. In one embodiment, this alarm is a light. The threshold value can be selected from a plurality of stored threshold values, which threshold value is selected from the plurality of stored threshold values in accord;~ce with predeterrnined criteria.
S l[n a further aspect of the present invention, a control system is provided for modif~ring the opel~Lillg pa,~.,t;Le,~ ofthe engine to adjust the output noxious polll-t~n~.e An emissions control system is provided which operates in response to receiving the predicted emissions value output by the predictive model and co",pal;ng it with a desired output level. The difference between the two levels is l";.~ ed by the control system.
SU~STIT~ S~ Rll~ 26) WO 95/04878 2 1 6 7 9 2 7 PCT/USg4/08657 BRIEF DESCRIPTION OF THE DRAWINGS
For a more complete underst~ntling of the present invention and the advantages thereof, reference is now made to the following description taken in conjunction with the acco.l.pa,.ying Drawings in which:
FIGURE 1 illustrates an overall block diagram of the virtual sensor of the present invention;
FIGURE l a illustrates a diag~ . " - A I ic view of the sensor validation system;
FIGU~E 2 illustrates a block diagram of the relation of the virtual sensor and the control system;
FIGURE 3 illustrates an embodiment llti~ ing a single,control network;
FIGURE 4 illustrates a diagrammatic view of a conventional neural network;
FIGURE 5a illustrates a more detailed block diagram of the control network;
FIGllRE 5b illustrates a detail of the iterate operation of FIG~RE 5a;
FIGURE 6 illustrates a detail of a typical plant, a boiler for a steam generation 1 5 facility;
FIGU~E 7 illustrates a block diagram of the sensor validation network;
FIGURE 8 illustrates a diag-~l--,--dlic view of the auto associative predictive network utilized in the system of FIGURE 7;
FIGUREs 9a and 9b illustrate plots of predicted versus actual pollutant sensor values and the dirrelence therebetween;
FIGlJREs 10a and 10b illustrate the plots of FIGUREs 9a and 9b, respectively, wherein one of the sensors is faulty;
FIGU~E 11 illustrates a flo~ ~ L for operating the overall system;
FIGURE 12 illustrates a flowchart for the sensor validation operation;
FIGURE 13 illustrates the prere--ed embodiment ofthe present invention wherein the virtual sensor is utilized in conjunction with an internal combustion engine;
FIGURE 14 illustrates a block diagram of a system wherein an external emissions sensor is utilized for llaini,lg, FIGIlRE 15 illustrates an alternate method wherein the rel.ai,ling operation is done external to the system with a reLI~ g processor;
SUBSTI~UTE ~!~E~T ~RULE 26) wo 95/04878 2 1 6 1 9 2 7 PCT/US94/08657 lFIGURE 16 illustrates a block diagram depicting the runtime operation ofthe internal combustion engine;
FIGURE 17 illustrates an overall view of the communication system;
FIGURE 18 illustrates a plot ofthe actual emissions output measured by an 5 external emissions monitor and the predicted output from the virtual sensor;
FIGURE 19 illustrates a flowchart depicting the overall runtime control operation; and FIGURE 20 illustrates a flowchart depicting the training operation.
SUBSn~l~TE St~tET (~ULE 26~
DETAILED DESCRIPTION OF THE INVENTION
Referring now to FIGURE 1, there is illustrated an overall block diagram of the system of the present invention. A plant 10 is provided that, during the normal operation thereof, releases some emissions 12 con~ g some level of pollutants. The pollllt~nts 5 12 are moniLoled by a pollutant sensor 14 or by utilization of EPA established referenGe methods, which sensor 14 is illustrated in phantom, to provide continuous emissions l.lol ilo~ g. This is rere.-ed to as a CEM. As will be described hereinbelow, the present invention provides a virtual sensor operation wherein the pollutant sensor 14 is only required for initial training of virtual sensor network. The pollutant sensor 14 is utilized 10 to gather training data to be combined with the control values and sensor values that are available to a Distributed Control System (DCS) 16, generally refe.-ed to as the plant h~...aLion system. The DCS 16 provides control values associated with control inputs to the system and sensor values to a computer 15. The computer 15 is comprised of a virtual sensor network 18 that ~ss~nti~ly provides a non-linear represent~tiQn of the 15 plant 10, which non-linear representation is a "learned" leprese..Lalion. The virtual sensor network 18 is operable to receive run time inputs 20 from a sensor validation system 22. The sensor validation system 22 is operable to receive actual measured inputs 24 from the plant 10 through the DCS 16. These measured inputs leplesellLmeasured state variables of the plant in the form of sensor values and also control values 20 that are input to the plant to provide control therefor. As will be described hereinbelow, the various inputs 24 are provided as inputs to the virtual sensor network 18 through the DCS 16. However, some of these inputs may be faulty and the sensor validation system 22 is operable to generate an alarm when any of the ~tt~rh~d sensors fails and to replace failed sensor values with reconciled sensor values.
The virtual sensor network 18 is operable to receive the inputs 20 and predict plant controls and alarms. The virtual sensor network 18 can predict what the pollutant levels are that normally would be monitored by the pollutant sensor 14; hence, it provides a virtual sensor. The sensor network 18 is a network that can be trained with a training system 28. The training system 28 utilizes as a target the actual pollutant level SL~ST~T~ E~RU~
wo 95/04878 2 1 6 7 9 ~ 7 P~T/US94/08657 on a lme. 13 as measured by the pollutant sensor 14 when it is present, and also the inputs 24 from the plant 10. The difference between the predicted pollutant level on a line 17 and the actual pollutant level on line 13 generates an error on line 19 that is used by the training system to adjust the stored representation in the virtual sensor module, so as to 5 ~ illl;,e the error. In operation, as will be described in more detail hereinbelow, the pollutant sensor 14 is a Continuous Fmie,eions Monitor (CEM) that is operable to be temporarily connected to the plant 10 to monitor the level ofthe pollutants 12. This provides a target to the training system 28. The network 18 is then trained with both the measured plant sensor and control values, not including the CEM output, and the CEM
10 output when present. This information is utilized to generate a training dataset.
~ fter training, the pollutant sensor 14 is removed and then the system operates by pretlicting what the output of the CEM or pollutant sensor 14 would be. The virtual sensor network 18 then replaces the pollutant sensor 14 and then can be utilized in a control function to predict plant control/alarms to ",~;"~ the operation ofthe plant 10 15 within acceptable standards. Further, the virtual sensor nelwolk 18 can be used solely to provide an output in place of the pollutant sensor 14 that can be utilized by the operator of the sensor to ensure that all nec~.es~ry procedures are being followed to ensure that the level of pollutants is within acceptable ranges. For example, if the predicted output from the n~lwolk 18 Pxceeded one ofthe established guidelines or 20 thresholds, the operator would then follow certain prescribed procedures to correct the situation. This would be the case even if the pollutant sensor 14 were present. The advantage to this is that the relatively expensive and difficult to Ill~ pollutant sensor 14 would not have to be present. Further, a new pollutant sensor 14 or a portable pollutant sensor 14 is periodically utilized to check the operation of a virtual sensor 25 network 18 to ensure that it is opel~ling colle-illy and that no parameters ofthe plant have ~h~ed such that the prediction is now incorrect or the model no longer represents the plant. In this manner, the system would have to be l ~LI ~h~ed by using a new set of - training data that would be provided by the operation of the connecting the pollutant sensor 14 to the plant lO. This could be the situation wherein some measurement device 30 degraded or the plant itself had physically changed parameters due to capital improvements, age, etc.
SI~B~T~U~ S~ lJL~ 26) WO 95/04878 2 ~ 6 7 S 2 7 PCT/US94/08657 In another mode of operation, the pollutant sensor 14 may be in a situation where it might be removed from the plant 10 for calibration purposes. During this time, the virtual sensor network 18 is then utilized to replace the sensor 14 during the calibration procedure.
Referring now to Figure 1 a, there is illustrated a block diagram of the operation ofthe sensor validation system 22. A plurality of sensors 27, 29, 31, 33 and 35 are illustrated. Each of the sensors 27, 29, 31, 33 and 35 have an output that is connected to the input of the virtual sensor 18. Additionally, each of the outputs is connected to an evaluation system 37 to d~Le~ ine if the sensor is valid, as will be described hereinbelow.
When any one of the sensors 27, 29, 31, 33 and 35 is determined to be faulty, it is replaced by a substitute sensor 39, which is a predicted sensor value that predicts the output of the faulty sensor ~Itili7ing a stored representation of the faulty sensor, which stored representation is a function of the other sensors 27, 29, 31, 33 and 35.
The.erore, the substitute sensor 39 requires as inputs the outputs ofthe valid sensors and the predicted output of the substitute sensor. This is illustrated in Figure la with the sensor 29 being substitl~ted~ with the substitute sensor 39 receiving as inputs the outputs of the sensors 27, 31, 33 and 35 and, in place of the output of the sensor 29, the predicted output of the substitute sensor 39. Further, another sensor could be substituted for with the output of the substitute sensor 39 being an input for the new and additional sensor (not shown).
Referring now to FIGURE 2, there is illustrated a block diagram for the operation wherein a virtual sensor predictive network 32 is provided which is operable to receive measured plant sensor values s(t) from the plant 10 and also the control values ~(t) which are inputs to the plant 10. The virtual sensor predictive network 32 is operable to output a predicted virtual sensor value oP(t) for input to a multiplexer 34.
The sensor value from sensor 14 is input on the line 36 to the multiplexer 34. The m--ltipl~oxçr 34 is operable to select either the predicted output of the network 32 or the actual output ofthe sensor 14 for input to a control system 38. The control system 38 is operable to generate the input values x(t) to the plant 10. The multiplexer 34 replesen the operation wherein the output of the network 32 is utilized to replace that of the sensor 14.
SlJ~TITU~E S~T (RU~ ~ 26) WO 95/04878 2 ~ 6 7 9 2 7 PCT~Sg4/08657 .
Rere,lh~ now to FIGURE 3, there is illustrated one embodiment of the system wherein a dynamic control system is provided. In this system, a control network 40 is provided which receives as an input the control input values x(t) and the sensor values s(t), the sensor values s(t) comprise the measured plant variables such as flow meter 5 measurel.lenLs, temperature measule",~nls, etc. In addition, the control net 40 is operable ~o receive a desired output value as one of the inputs. The control net 40 contains a stored representation of the plant and is operable to output a set of control input values x(t+l). These are input to a Distributed Control System (DCS) 42, which is operable to generate the control values ~(t). The control network 40 is a conventional 10 control network that is trained on a given desired input, and which control network 40 is operable to receive the sensor values and control values and generate the updated control values x(t+l) that are necessary to provide the desired outputs. . The control network 40 is generally comprised of a neural network having associated therewith weights that define the l~les~ liQn that is stored in the neural network. In the embodiment of 15 FIGURE 3, these weights are frozen and were learned by training the control network 40 on a given desired output with a given set of training data for the control values x(t) and the sensor values s(t). A desired output is provided as one input for selecting between sets of weights. The general operation of control nets is described in W.T. Miller, III, R. S. Sutton and P.J. Werbos, '~eural Networks for Control", The MITPress, 1990,20 which reference is incorporated herein by reference.
Re:ferring now to FIG~JRE 4, there is illustrated a detailed diagram of a col.vellLional neural network comprised of input nodes 44, hidden nodes 46 and output nodes 48. The input nodes 44 are comprised of N nodes labelled xl, X2, ... XN, which are operable to receive an input vector x(t) comprised of a plurality of inputs, INP1(t~, 25 INP2(t), ... INPN(t). Similarly, the output nodes 48 are labelled o" 2~ OK, which are operable to generate an output vector o(t), which is comprised of the output OUTl (t), OUT2(t), ... OUTK(t). The input nodes 44 are interconnected with the hidden nodes 46, hidden nodes 46 being labelled al, a2, .. . an, through an intel col~le~;lion network where each input node 44 is interconnected with each of the hidden nodes 46. However, some 30 intelc(~ e~iLion schemes do not require full interconnection. Each ofthe intel~iomle.;
has a weight Wij'. Each of the hidden nodes 46 has an output oi with a function g, the output of each of the hidden nodes defined as follows:
Sl~S ~ ~TU~E S~EET (RULE ~6) -Wil; Xi + b~ ) (1) Similarly, the output of each of the hidden nodes 46 is interconnecte~l with subst~nti~lly all of the output nodes 48 through an interconnect network, each of the interconnects having a weight Wjk2 associated therewith. The output of each of the output nodes is defined as follows:
k = g(~; W~k aJ + b2) (2) 5 This neural network is then trained to learn an function f(x(t), P) as follows:
o ( t) = f (~ ( t), P) (3) where o(t) is an output vector and P is a vector or p~Lers ("weights") that are variable during the learning stage. The goal is to ~ .e the Total-Sum-Square-Error filnr,tion -( t) -o ( t) ) 2 (4 ) The Total-Sum-Square-Error function is ",;n;l.l;~ed by rh~nging the parameters P ofthe 10 function f. This is done by the back propagation or a gradient descent method in the p~c;rt;;lled embodiment on the parameters Wjk2, Wjjl,bl" b2k. This is described in numerous articles, and is well known. Ther~role, the neural neLwulk is ess~nti~lly a ,u~lleLer fitting scheme that can be viewed as a class of st~tictir~l algo,iLlulls for fitting probability distributions. Alternatively, the neural network can be viewed as a filnction~l 15 appl u~hllhLor that fits the input-output data with a high-~limP.n~iQnal surface. The neural n~Lw~lk utilizes a very simple, almost trivial fiunction (typically sigmoids), in a multi-layer nested structure The neural network described above is just one example. Other types of neural networks that may be utilized are those using multiple hidden layers, radial basis Sl IBST~TUT~ SHE~T ~RULE 7~) WO 95t04878 2 1 6 7 9 2 7 PCT/US94/086S7 functions, ~ n bars (as described in U. S. Patent No. 5,113,483, issued May 12, 1992, which is incorporated herein by reference), and any other type of general neural - nt;Lwo~k. In the plere.led embodiment, the neural network utilized is of the type referred to as a multi-layer pelc~Lloll.
Rt;rt;llh1g now to FIGURE 5a, there is illustrated a block diagram of a control system for opl;",;~l;on/control of a plant's operation. The plant 10 has an input for receiving the control values x(t) and an output for providing the actual output y(t) with the sensor values s(t) being associated therewith, these being the internal state variables.
A plant predictive model 74 is developed with a neural network to accurately model the plant in accordance with the function f(x(t),s(t)) to provide an output oP(t), which represell~s the predicted output of plant predictive model 74. The inputs to the plant model 74 are the control values x(t) and the sensor values s(t). For purposes ofo~ i ;Qn/control, the plant model 74 is deemed to be a relatively accurate model of the operation ofthe plant 72. In an opl;~ ;on/control procedure, an operator independently generates a desired output value od(t) for input to an error generation block 78 that also receives the predicted output oP(t). An error is generated between the desired and the predicted outputs and input to an inverse plant model 76 which is ntic~l to the neural network repreS~ntin~ the plant predictive model 74, with the exception that it is operated by back propa~tin~ the error through the original plant model with the weights of the predictive model frozen. This back prop~tion of the error through the network is similar to an inversion of the network with the output of the plant model 76 repres~ontin~ a ~x(t+l) utilized in a gradient descent operation illustrated by an iterate block 77. In operation, the value /~ ~(t+l) is added initially to the input value ~c(t) and this sum then processed through plant predictive model 74 to provide a new predicted output oP(t) and a new error. This iteration continues until the error is reduced below a predetermined value. The final value is then output as the new predicted control values x(t+1).
This new x(t+l) value comprises the control values that are required to achieve the desired actual output from the plant 72. This is input to a control system 73, wherein 30 a new value is presented to the system for input as the control values x(t). The control system ~3 is operable to receive a generalized control input which can be varied by the ~ U~ S~-ET ~ LE ~
distributed control system 73. The general terminology for the back propagation of error for control purposes is "Back Propagation-to-Activation" (BPA).
In the plt;r~;llcd embodiment, the method utilized to back propagate the error through the plant model 76 is to utilize a local gradient descent through the network 5 from the output to the input with the weights frozen. The first step is to apply the present inputs for both the control values x(t) and the sensor values s(t) into the plant model 74 to generate the predicted output oP(t). A local gradient descent is then pc~ru~ ed on the neural n~Lw~Jlh from the output to the input with the weights frozen by inputting the error beL~eell the desired output od(t) and the predicted output oP(t) in lO accordance with the following equation:
A~(t) = Z(t + 1) - ~(t) = I~ a(O-d~t) - oP(t))2 ~5) where 1l is an adjustable "step size" parameter. The output is then regenerated from the new x(t), and the gradient descent procedure is iterated.
Referring now to FIGURE 5b, there is illustrated a detailed block dia8ram of theiterate block 77. The iterate block 77 is comprised of a sll".",il-~ junction which is 15 operable to receive the ~x(t+l) input and the output of a multiplexor/ latch block 86.
The multiplexor/latch block 86 is operable to receive both the output of the summing junction 84 for feedb~rk as one of the inputs and the control variable x(t). The output of the sllmming block 84 is the sum of the previous value of x(t) plus the new iterative change value /~x(t). This will then be iteratively s -mmçd with the previous value to 20 gencl ~le a new iterative value until the error is at a predetermined level. At this point, the output of the sl~mming junction 84 will comprise the new control value x(t+1).
Another standard method of o~ A I ion involves a random search through the various control values to minimi7e the square of the difference between the predicted outputs and the desired outputs. This is often referred to as a monte-carlo search. This 25 search works by making random çh~n~çs to the control values and feeding thesemodified control values into the model to get the predicted output. The predicted output is then coll~ared to the desired output and the best set of control values is tracked over S~BSTITU~E S~ IL~ 2~) .
the entire random search. Given enough random trials, a set of control values will be obtained that produces a predicted output that closely m~t~hes the desired output. For reference on this technique and associated, more sophisticated random o~ i7AI;ontechniques, see the paper by S. Kirkpatrick, C.D. Gelatt, M.P. Vecchi, "O,ul;...;~;1lion by Simul~ted ~nne~lin~" Science, vol. 220, 671-780 (1983), which reference is incorporated herein by lt;fe.t;llce.
Referring now to FIGURE 6, there is illustrated a diagrammatic view of a typical plant that may exist at a m~n~f~ctl~ring facility. The plant typically comprises a boiler 92 which has a firebox 94 disposed at the lower end thereo The boiler 92 interfaces with a stack 96 through a preheat chamber 98. Many tubes of which tube 100 is typical thereof are operable to run through the chamber 98 and enter the boiler 92. The tube 100 then passes in a sel~c;llLille manner through the boiler 92 to an output pressure vessel 104, which is pressurized. The vessel 104 is operable to generate steam out of an outlet 106.
The other end ofthe tube 100 that enters the chamber 98 is connected to a source 108 of the deionized water. In operation, the water is passed through the tube 100 to the challlbel 98, which picks up heat therein and then into the main boiler 92, where it is heated further. This then passes through to the vessel 104. The firebox 94 has a heating ment 116 associated therewith that is operable to receive gas through a gas line 118 and air through an air line 120. The mixture of the gas and the air allows the heating el~m~nt 116 to generate heat in the firebox 94 and heat up the water in the tube 100 within the boiler 92.
The tube 100, when it exits the source 108 with the deionized water at the source, has the flow thereof measured by the flow meter 122. A valve 124 allows control ofthe flow of fluid from the source 108 into the challll)el 98. Two telllpel~LLIre sensors 126 and 128 are provided at di~lt;lll locations along the tube 100 within the ch~llbel 90 to provide temperature measurements therefor. Additionally, temperature - sensors 130, 132 and 134 are provided along the tube 100 at various locations within the main boiler 92. A temperature sensor 136 is provided for the firebox 94. The level of the fluid within the pressure vessel 104 is measured by a level meter 142 and the pres~ule therein is llleasllled by a pressure meter 146. A flow meter 150 is provided formeasuring the flow of steam out ofthe pressure vessel and a control valve 152 provides SUB~T~TE S~EET (~ULE 2~) WO 95/04878 2 ~ 6 7 9 2 7 PCT/US94/086~7 control of the steam exiting the pressure vessel 104. The heater element 116 is controlled with a valve 158 on the gas line, which has the flow thereof measured by a flow meter 160. The flow meter on the air line 120 is measured by a flow meter 162. A
damper 163 in the stack 96 is utilized to control air flow through the firebox 94.
It can be seen that the sensor values s(t) of the plant are provided by the various te~ el~LLIre and flow measurement devices. Further, the control values, in the form of the various valves and damper positions provide the control values to the plant.Thel ;;fol e, an operator can control the operation of the plant by controlling the various 10 flow meters and other control values, some of which are not illustrated. The l~ it~g inputs that are necess~ry in order to provide adequate control of the plant for the purpose of continuous emissions monitoring are the NOx levels. These are provided by the virtual sensor network 18 of FIGURE 1. However, as described above, periodically a portable unit 170, having disposed thereon a CEM 172, is conn~cted via a duct 174 to 15 the stack 96 to measure the amount of NOx in the output emissions to the air. The CEM
172 then generates a report as to the level of the NOx. If this level is within acceptable standards, then this is merely reported. However, if the level is outside of acceptable limits, this is reported to the plant operator and either ~ ges are made or the plant is shut down. Additionally, the information generated by the CEM 172 is generated on a 20 time base and this comprises training data. This training data, since it is on a common time base, can then be colllbilled or merged with data associated with the sensor values and the control values, which are also on a time base, to provide new training data for the virtual sensor network 18. This can be utilized by the training system 20 to retrain the virtual sensor network 18, if nPce~ y.
R~ft;llhlg now to FIGURE 7, there is illustrated a block diagram ofthe pr~fell~dembodiment for the sensor validation system 22. To ensure that the overall inputs ~(t) to the network 18 are "valid", it is nece,ss~ly to perforrn some type of conlpalison with expected or predicted values. If it is suspected that the generated values are not accurate, then an alarm is generated to advise the plant operator or the control system of the need to calibrate the sensor or to repair the sensor, and an estim~ted or predicted value for that sensor value is substituted for the actual measured value of the sensor.
SIJ~STITUTE SHEET ~RULE 2~
.
In a pr~rel,ed embodiment, an auto associative predictive neural network 180 is provided which is a network having an input layer for receiving select ones of the inputs s(t) on an input 182 Although not illustrated, only certain ones of the actual sensor values are nece~saly as inputs to the virtual sensor network 18 in order to provide an accurate prediction of the NOx levels that would generally be provided by the pollutant sensor 14. These are deLellllilled by p~lrolllling a sensitivity analysis. This is described in U.S. patent application Serial No. 056, 197, filed April 30, 1993 and entitled "Method and Apparatus for Detel n,il~ng the SensiLi~/iLy of Inputs to a Neural Network on Output Parameters" (Atty. Dkt. No. PAVI-21,761), which is ~ ned to the present Assignee.
By utili7:ing the sensiLiviLy analysis, the number of inputs to the neLwul k 18 can be significantly reduced and only the important inputs ~Itili7ed This ~i~3nifit~ntly reduces the size of the auto associative predictive network 180 and also the virtual sensor network 18.
The actual inputs X(t) are input to a multiplexer 186 which is operable to select between the predicted inputs ~CP(t) output by the nc;Lwol k 180, which is a predicted output, and the actual inputs ~C(t). In operation, a first cycle occurs when the multiplexer selects the actual inputs x(t). The predicted inputs xP(t) are then input to a subtraction circuit 188 to deLelll,il~e the difference between ~C(t) and ~P(t). This di~rei~ce is input to coll,pal~tor 190 for co",pa~ison with thresholds stored in a threshold memory 192. The one of the actual inputs to the network 180 having associated therewith the largest error as colll~)are-d to the acceptable threshold is then conn~cted to the associated predicted output of the network 180. The actual inputs X(t) with the substituted or lecol~l~ecLed input is then again cycled through the auto associative predictive network 180. On this next cycle, the difference between the actual and the predicted values are againdetermined, colll~a~ed with the thresholds, and the one ofthe actual inputs having the largest error is reconnected to the associated predicted input by the multiplexer 186.
This continues until all of the predicted inputs, with the determined faulty or un~cceFtable actual values replaced with the predicted values output by the network 180, are within a predetermined range. Once this has occurred, the predicted values from the n~Lwolk 180 are input to a multiplexer 196, and the multiplexer 196 selectin~ for output the, ~r, O"~ the actual values that were determined to be acceptable and the predicted values as a substitute for the actual values that were determined to be unacceptable. It SI~T~ TE SH~E~ (RlJLE 26) WO 9~/04878 2 1 6 7 9 2 7 PCT/US94/08657 should be noted that the predicted values are generated by running the network with the determined ~n~cceptable actual values replaced with the associated predicted values by the multiplexor 186. The output of the multiplexor 196 is then input to the virtual sensor network 18.
In another embodiment, the predicted input values output by the auto associativepredictive network 180 can be provided as the input to the virtual sensor network 18.
This would then not require the multiplexer 196 and, in fact, the auto associative predictive ntLwolk 180 can continually monitor and replace ones ofthe sensor inputs that are determined to be invalid.
Referring now to FIGI 1RE 8, there is illustrated a ~ g, ~ ;c view of the auto associative predictive network 180. The network is comprised of an input layer of nodes 198 and an output layer of nodes 200. There is one node in the layer 198 for each ofthe input vectors ~(t), illustrated as x,(t), x2(t) ... xn(t). Similarly, there is a single node for each of the predicted output variables xP(t) such that there are outputs xlP(t), x2P(t) xnP(t). The input layer of nodes 198 is mapped through to the output layer of nodes 200 through a hidden layer of nodes 202. The hidden layer of nodes 202 has a plurality of intelco~ ;o~.~ with each of the nodes in the input layer of nodes and each of the output layer of nodes 200. Each of these interconnections is w~ightetl Further, the number of nodes in the hidden layer of nodes 202 is less than the number of nodes in either the input layer 198 or the output layer 200. This is therefore lt;~lled to as a bowtie neLwolk. The nelwolh 180 can be trained via a back propagation ll~i~f~llgte~hni~ e. This is described in D.E. l~l-m~lh~rt, G.E. Hinton and R.J. Williams,"Learning Tnt~rn~l Reples~ ;t)ns by Propagations" in D.E. ~l~m~lh~rt and J.L.
McClelland, Parallel Disfribufive Processing, Vol. 1, 1986.
Referring now to FIGUREs 9a and 9b, there are illustrated two plots depicting operation of the sensor validation system 22. The actual inputs are represented by XA
and the predicted input is represented by Xp. It can be seen that the predicted input does not exactly follow the actual input, it being noted that the actual input is actually the input to the overall system. The difference between the actual and the predicted input values is illustrated in FIGURE 9b.
S~B~ITU~ ~HE~ UL~ 26~
~ ===
wo 95/04878 2 1 6 7 9 2 7 PCT/US94/08657 .
Referring now to FIGI~REs 10a and lOb, there is illustrated corresponding plots to those of FIGUREs 9a and 9b with the exception that the sensor genel~Li.lg the actual input fails. It can be seen that up to a point 204 on the curve X~, the predicted and actual sensor values track fairly well with minim~l error. However, at the point 204 the error 5 increases dramatically, indicating that the sensor no longer provides an value that corresponds to the predicted value. This is illustrated in FIGURE 10b, wherein the error increases. When the difference between XA and Xp is greater than a threshold, this inrlic~tes an invalid reading. However, as noted above, only the one of the sensors having the highest error above the threshold will be selected as repl~cement value by the multiplexer 86 for the next cycle. This is due to the fact that the network 180 is trained on all of the input variables and each of the input variables will affect the predicted values for the, ~ g ones. Therefore, if the actual input values associated with predicted output values having an error greater than the threshold were replaced, this would not be as accurate as iteratively reFI~cing one at a time.
R~;~.lil~ now to FIGURE 11, there is illustrated a flow~ a, L depicting the overall operation of the system. The fl~w~,hal L is initi~te~l at a start block 208 and then flows to a decision block 210. Decision block 210 determines whether the remote CEM
has been installed. If so, the program then flows to a function block 212 to measure the NOx levels with the remote CEM. The program then fiows to a decision block 214 to 20 determine whether the measured NOx values, measured in function block 212, are acceptable. If not, this in~ tes that the virtual sensor network 18 is out of spec and that the system has either chal1ged or the network no longer I ~presellLs the system. The program will then flow along an "N" path to a function block 216 to measure the system variables and then to a function block 218 to generate a training ~t~b~.~e. A training 25 d~t~baee ess~nti~lly utilizes the system variables that are measured along the same time base as the measured NOx levels. Typically, the remote CEM will be placed ~ c~nt to the m~ml~chlring facility and the pollllt~nt~ measured for a predetermined amount of time, which can be measured in hours, days or weeks. At the same time, the plant facility itself is measuring the plant variables. These are also placed on a time base and stored.
30 By n~ ging the two data sets, a training (J~t~b~e can be provided for training the virtual sensor network 18. This time merging operation is described in U.S. Patent Application S~IBSTITUTE SHEET (P~ULE 2~) wo 95/04878 2 1 6 7 9 2 7 PCT/US94/08657 Serial No. 980,664, filed,~ovember 24, 1993 and entitled "Method and Apparatus for Opel~til1g a Neural Network with Missing and/or Incomplete Data" (Atty. Dkt. No.PAVI-20,965).
Once the training database has been generated, the virtual sensor nc;twolk 18 is5 trained, as inflir~ted by a function block 220. This es~Pnti~lly generates weights, which can then be substituted for the neural network weights in the virtual sensor network 18.
The program then flows to a function block 222 to substitute new weights in the virtual sensor network 18. Thereafter, the program flows to a main ope~Lhlg portion ofthe program, which is initi~ted at a function block 224 to validate the sensors.
If the pollutant parameters measured in the function block 212 were acceptable, the program would flow from the decision block 218 along a "Y" path to the input of function block 224 to bypass the training step. Additionally, if the remote CEM is not present, the program would flow along an "N" path from the decision block 210 to the input of the sensor validation block 224.
The sensor validation block 224 validates the sensors and, if one is found invalid, it substitutes a predicted value for that invalid sensor. The program would then flow to a function block 226 to determine if certain sensors needed to be replaced by predicted values. If so, the program would flow along a "~' path to replace the invalid sensors with the predicted sensor value. The program would then flow to a function block 232 20 to predict the pollutant value and then to a function block 232 to control the plant. The program would then flow back to a decision block 210. If it were determined thatsensors did not need to be replaced by their predicted values, the program would flow along an "~' path from the decision block 226 to the input of function block 230.
Referring now to FIGURE 12, there is illustrated a function block depicting the 25 operation of the sensor validation. The program is initi~ted at a start block 240 and then flows to a function block 242 to input the various sensor re~ing~ The program then flows to a function block 244 to run the sensor validation model and then to a decision block 246 to compare the predicted input values with the thresholds and generate an S~B~lTl~rE ~IEE~ ULE ~6-~
wo gs/04878 2 1 6 7 9 2 7 PCTIUS94/08657 error signal when any of the predicted input values exceed the thresholds for that given variable, it being noted that there can be a threshold for each variable as to what con~titutes an error for that sensor value. When an error exists, the program flows to a function block 248 to replace the largest input error with the mean value for that input.
5 An alarm is generated at this point to warn of the failed sensor. The program will then flow back the input of a function block 244.
When the system has iteratively detel ll"ned that there are no longer any predictive outputs that exceed these thresholds, the program will flow from a decision block 246 to a function block 250 to replace all detected errors with predicted sensor lO values and then to a function block 252 to output reconciled sensor values. The program will then flow to a return block 254.
Referring now to FIGURE l3, there is illustrated an embodiment v~ heleill the virtual sensor is utilized in conjunction with an internal combustion engine 260. The int~rn~l comh--stion engine 260 receives air on an intake port 262. This is input to a l 5 butterfly valve 264 which is basically the throttle valve that is controlled by the foot pedal on an automobile. The butterfly valve then feeds the restricted airflow to an intake manifold 266, which is input to the internal combustion engine 260. A plurality of fuel injection valves, only one of which is illustrated, fuel injection valve 268. The filel injection valve 268 is operable to inject fuel into the intake manifold 266 in a re~ ted 20 amount, which is determined by a number of factors, this being conventional. The internal combustion engine 260 exhausts the combustion ingredients into an exhaust m~nifold 270. The exhaust manifold 270 is conn~cted to an exhaust pipe 272 which is connected to the input of a catalytic converter 274. The catalytic converter 274 then interfaces with the tail pipe 276 to output the combustion gases.
The exhaust manifold 270 is connected through a pipe 278 to an emissions gas recirc.... ....l~tion valve (EGR) 280. The EGR 280 is interfaced with the intake manifold 266 through a pipe 282. The EGR 280 is a conventional pollution control device that is operable to bleed offa small portion of the exhaust gases from the exhaust manifold 270 back into the intake manifold for recombustion thereof. EGR valves typically operate at S~85TITU~ E~T (~ 63 WO 95/04878 2 1 6 7 9 2 7 PCTrUS94/08657 higher RPMs of the engine, as the recirculation of the exhaust gases at low RPMs causes the engine to idle roughly.
In typical internal combustion ~.~in~.~, an oxygen sensor 286 is disposed in the exhaust manifold 270. The oxygen sensor 270 basically provides for measurement of the 5 exhaust gas ingredient concentration, typically oxygen, that exists in the exhaust manifold 270. These type of sensors are utilized for air-fuel ratio control systems. The sensor is input to a central processing unit (CPIJ) 288 which controls the operation of the fuel injectors through a line 290 and the fuel supply thereto. Again, these are collvelllional systems.
10In addition to the oxygen sensor 286, the CPU 288 is operable to monitor a large number of p~llelers lègal~ling the internal combustion engine, which parameters require some type of sensors. In the example illustrated in FIGURE 13, the operation of the ignition, i.e., spark advance, timing, etc., is provided to an ignition module 292. The manifold temperature is provided by a manifold temperature sensor (TM) 294 and the 15 cylinder telllpel~ e is provided by a cylinder telllpe~Lule sensor (TCYL) 296. The sensors 294 and 296 allow a telllpel~ re measurement to be made of each cylinder and the overall m~nifold The back pressure in the exhaust pipe 270 is provided by a pressure sensor (PEX) 298 and the pressure in the intake manifold is provided by a pressure sensor (PMAN) 300. Typically, the pressure sensor 300 measures a vacuum for 20 a conventional engine, which could be a positive pressure when the system is operated in conjunctiQn with a turbocharger. The intake manifold temperature is provided by a telllpel~ re sensor (TA) 302, which ls connected to the intake manifold 266. Theposition of the butterfiy valve 264 is provided by a position sensor (~th) 306. The tem~elal~lre of the catalytic collvel Ler is measured by a temperature sensor (TCAT) 308.
25 The CPU 288 incol~ol~Les the full sensor system as described hereinabove and is operable to predict the emissions emitted by the internal combustion engine which, in the plerelled embodiment, are primarily NOX, and either output this information in the form of a display, store it as a history or utilize it to control the operation of the engine.
30When controlling the operation ofthe engine, an emissions control system 308 is provided, which is operable to control certain parameters of the system. For example, S~ST~rUTE ~ ET ~ULE ~
WO 95/04878 2 ~ 6 7 9 2 7 PCT/US94/08657 .
one parameter that could be controlled is the internal threshold to the CPU 288 that clet~rmines the air-fuel ratio. Typically, the oxygen sensor 286 operates on a threshold such that when it is above the threshold, the air-fuel mixture is ch~rlged in one direction and, when the oxygen sensor falls below the threshold, the air-fuel ratio is ~h~nged in the 5 other direction. By ~h~nginc~ the threshold, the average air-fuel ratio will be changed, and Ihelerole, the actual emissions output can also be çh~n~e~ This will be described in more detail hereinbelow.
For systems that require emissions monitoring, there are a number of methods that have been employed in the past. One method that has been accepted by the 10 Environment~l Protection Agency (EPA) is to have a dashboard light that will illllmin~te after a predetermined number of miles, which predetermined number of miles is determined empirically. When the light is ill~lmin~tetl, the vehicle must be taken in for inspectiGn. With the emissions sensor of the present invention, the actual emissions can be predicted from the operation of the engine and the light illllmin~te(l, indicating that the 15 predicted emissions levels have fallen below a predetermined threshold. Thesethresholds can be ~h~nge~l, depending upon the type of vehicle, the area of the country, etc. For example, most commercial vehicles have less ~L~;"genL pollution standards and certain areas like Los Angeles, California have very stringent standards. In this manner, di~rellL thresholds can be loaded into the CPU 288, which thresholds can be sPIected in 20 accordance with predetermined criteria. Additionally, a history can be provided of the vehicle as to the emissions generated by the engine and this stored for downloading at a later time for the purpose of monitoring the operation of the vehicle and the associated engme.
Since the emissions prediction is achieved u~ ing a model of the system which 25 utilizes the sensor outputs as inputs to provide the predicted output, the model must be initially trained. This initial training operation can be done on a "generic" engine with a "standard" emissions monitor and the then the model parameters downloaded to a standard integrated circuit which is utilized for all vehicles. This ~c sllmes that the model trained on the generic system holds true with respect to all subsequent systems and the 30 m~mlf~ct~ring tolerances associated therewith. However, even if the generic model does provide a true representation of all engines m~mlf~chlred for a given type of engine in SIJBSTITUTE SHEET (RULE 26~
combination with a standard çmi~ion~ monitor, the parameters of the engine will change over time. It may thel~role be necess~y to update the training. This can be achieved in two ways. In the first method, the emissions can be measured with an emissions sensor that is external to the vehicle and not an integral part thereof and this compared to the predicted emissions output. If an error exists that is too large, the system then can be retrained. The . tiLl ~nil~g can either be a complete I t;L. ~inillg of the model or merely an update of the training weights. In any event, this requires an actual emissions sensor to be ~ltili7e(l In a second method, the model can have a "bias" applied thereto to provide a slight offset. This also e~ui.es actual emissions to be monitored. The actual values are necess~.y to know how to adjust the bias.
Referring now to FIGURE 14, there is illustrated a block diagram of a system wherein an external emissions sensor 310 is utilized for l-~i. ing. The internalcombustion engine is represented by a block 312, which receives inputs on a line 314 and provides measured outputs or state variables s(t) on lines 316, these comprising the inputs to the model However, these state variables are first processed by sensorvalidation module 318, which was described hereinabove, and which is operable tosubstitute a predicted sensor output in the event of a failure of one of the sensors. This is described hereinabove with .erelence to FIGI~REs 7 and 8. The v~lid~ted statevariables s(t)' are output on lines 320 to a predictive model processor 322. Thepredictive model processor 322 is operable to interface with the memory 324 for storing model parameters to process the pa.~-..e~ers and provide a predicted output e(t) on a line 326 ~lrlition~lly~ the predictive model processor 322 is operable to store a time history ofthe predicted output in a nlellloly 328 The model that is stored is a represent~tion of the combined engine and emissions sensor Therefore, the model will have associated therewith all aspects of both the engine and the emissions sensor. Even if the emissions sensor is inaccurate, the model is only as good as the sensor, but this inaccuracy must be incorporated into the model.
This is important in that regulatory bodies require that the output measurement comply to their standards, which standards are defined by their equipment. If, for example, the emissions sensor that complied with their standards were in fact inaccurate, it would be important to predict an output with these inaccuracies. To correct these inaccuracies UT~ S~ UL~- 2~
wo g~/~78 2 1 6 7 9 2 7 PCT/US94/086S7 .
would not be acceptable. Therefore, the model is determined utili7in~ as part ofthe system the actual hardw~ sensor, which is removed during operation ofthe eng-ine.
r.~tern~l to the internal combustion engine 312, the external emissions sensor 310 is connPcted to the output ofthe internal combustion engine, which comprises a line 330 5 labelled y(t). This represents the output of the system. This is merely the output of the exhaust pipe. The emissions sensor 310 is connected to the output to provide an actual output value ofthe emissions on a line 332. This is input to a difference device 334 to determine the difference between the output of the emissions sensor 310 and the predicted output on line 326. This generates an error E, which is input to a co",pa,~lor 336. The co.l.~Lor 336 COlllp~ht;S the error E with a predetermined threshold and then outputs a "Train" signal on a line 340. If the error exceeds the threshold, a training operation is initi~ted This is input to the predictive model processor 322. The predictive model processor 322 then enters into a Ll~il~ing mode utili7:ing the actual emissio~s sensor outputs on a line 344 and the state variable inputs 320 to retrain the 15 model. These p~..elers are then input to the memory 324. After L~ ~ing~ the system will again be v~ ted by col,lpaling the operation ofthe internal combustion engine and the emissions output thereby with the predicted emissions output. Once the error has been ...;..;...;,e~ i.e., reduced below the threshold, the system will be "v~ ted". The predictive model processor 320 can train the network by two methods. In the first 20 method, it can completely regen~l~Le model parameters from scratch ~ltili~ing a typical training algorithm. In the second method, it can merely update the model pal~lllc;Lel ~, i.e., provide a minor adj-letm~nt thereto in order to reduce the error.
l~erelling now to FIGURE 15, there is illustrated an alternate method whelt;in the l~Ll~lling operation is done external to the system with a leLI~nillg processor 346.
25 The l~ ining processor 346 is operable to receive on the input thereofthe output state variables from the internal combustion engine 312 on lines 348, the actual output - emissions sensor on the line 332 and the error output of the difference device 334. The inillg processor 346 then d~L~Illlilles whether ~ il~ing is necç~s~ry, and if so, the leLl~.fi.lg processor 346 will either update the model parameters or generate a new set of 30 model p~lleLers. During an update process, the old model parameters from the memory 324 are ~lplo~ded and adjusted and then downloaded back to the memo~ 324.~J~UT~ T ~ L~ 2û.~
.
In the complete training process, completely new model parameters are generated and then downloaded to the memory 324. Of course, after I~Ll~h~ g or any modification of the model p&~ lers 324, the system is again checked.
In the training of the network, one technique that can be utilized is 5 back~l-,pagation, as described in D.E. P~llm~lh~rt, G.E. Hinton, and R.J. Williams, "Learning Internal Represent~tiQns by Error Propagation" in D.E. l?llm~lh~rt & J.L.
McClelland, Parallel Distribu~ed Processing, Vol. 1, 1986. In this technique as applied to a neural nGlV~O1k~ training is achieved by ~ g the Least Mean Square Errors with back~ropagation. This utilizes the technique of steepest descçnts, wherein the 10 weights Wu f a neural network and the parameters associated with the activation function are varied to Ill;~ e the error function. This bacl~,opagation te~hniq~1e is ess~nti~lly a common, non-linear least squares algorithrn. It is a natural, non-linear extension of the linear nets commonly used in adaptive signal processing. Use of the chain rule in comruting derivatives ofthe error generated during the training procedure 15 provides useful interpolation to the ...;I~;.";~;on process and allows an easy generalization to multi-layers of non-linear units in a neural network. For one or more output units, the error output is i~,;,.;".;~.etl to:
1 ~ ( z (t) _ yf(t) )2 (6) where: y(t) = output of a neural net; and z(t) = specified target output for a given output pattern.
20 For a network that contains non-linear hidden units, the term y(t) contains contributions from the outputs of hidden units in a hidden layer. Because the hidden layer has a non-linear ~ srel function, the output of the hidden layer will be an output of the non-linear function of its input and the error E becomes a square of non-linear function weights since the hidden layer outputs are fed into the topmost output layer in a conventional 25 three layer neural network. The backpropagation algorithm is described in the literature and also described in U.S. Patent No. 5,113,483 issued May 12, 1992 and entitled SU~S~I~UTE ~ 6~
WO 95/04878 2 1 6 7 9 2 7 PCT/US94/086~7 "Network with Semi-Localized Non-Linear Mapping of the input Space". This patent is incorporated herein by lt;rt;l~nce.
l[n addition to bach~l opagation, other techniques for training a neural networkcan be ~1tili~-l, such as radial basis functions or C~ ssi~n bars. Further, it is not 5 necee~, y to utilize a neural nc;~wol k to provide a stored repres~nt~tion of the system. A
fuzzy system (which is very similar to a radial basis function network) can also be ili~ed Ref~ling now to FIGURE 16, there is illustrated a block diagram depicting the runtime operation of the internal combustion engine 312. The system of FIGURE 1610 utilizes a control network 350 which is operable to receive the control input ~(t) on line 314, the v~ ted sensor outputs s(t)' on lines 320 and output updated control inputs ~(t+1) on a line 352. The control network 350 is described above with refel~nce to FIGURE 5a and can actually incorporate the model that is implçm~nted by the predicted model processor. A desired or target emissions level is input thereto. The line 352 is 15 input to an engine control system 354 which is operable to effect the various controls on the internal combustion engine. Any one of the controls can be manipulated to control the emissions within predet~ lined ~lid~.lines, these controls associated with controlling he air-fuel ratio.
The overall system is controlled by a runtime operating system 354 which is 20 operable to receive the predicted output on the line 326 from the predictive model processor 322 and also receive the output of the sensor validation module 318 intlic~tin~
which, if any, of the sensors have been determined to be in error. This h~rullllalion~ in addition to the predicted emissions value output on line 326 is then utilized by the runtime opel~Lil1g system to either store it in the memory 328 associated with the 25 historical h~----aLion or make various decisions as to what should be done with respect to the predicted emissions information.
Runtime thresholds are prestored in a memory 356 and are utilized by the runtimeoperating system for com~alison with the predicted emissions. If the predicted emissions 30 exceed the selected one of the thresholds, some action must be taken. For example, ~IB~ITUr~ ~HE~T ~LE 26J
WO 95/04878 2 1 6 7 9 2 7 PCT~US94/08657 emissions may be acceptable at one threshold in one area of the country and ~n~cceptable in another area ofthe country. Further, the predicted emissions levels may also have an acceptability that is a functioh of other parameters, such as telllpe,~ re and humidity. The runtime thresholds could be selected as a function of atmospheric 5 con-lition~ or other criteria. However, in the pl~relled embodiment, it is anticipated that thresholds will be selected as a function of the locale that the engine is disposed in.
Further, they could even be s~lected as a function of the time of day.
A user input 358 iS utilized to select the thresholds or input thresholds via an input/output circuit 360. Further, the input/output circuit 360 is operable to interface with a display 360 and also with a comm.lnic~tion system 362. The display 360 can be, for example, a warning light. Further, it could be some type of display that actually outputs an analog value in the form a "gas gauge" for viewing by the driver. This would allow the driver to actually view the emissions levels as a function of his driving 15 conditions, etc. In one mode, the communications system 362 is provided such that industrial engines at remote sites can be controlled on a periodic basis to download the stored history h~lllla~ion in memory 328 to a central station.
Referring now to FIGURE 17, there is illustrated an overall view ofthe 20 communication system. A plurality of engines re~lled to as plants 370 are disposed at remote locations, each having a virtual emissions monitor 372 associated therewith and each having a communications device 374 associated therewith. In the described embodiment, each of the communication devices 374 is operable to ll~llslnil i~ ion over a wireless commllniç~tion path via an ~ntçnn~ 376, which ~ntPnn~ 376 can operate 25 in both a receive mode and a ll~l~nlil mode. The ant~nn~ 376 are operable to co""~ ;c~te with an ~ntçnn~ 378 on a colllllland station 380. The protocol utilized for the tr~n~mi.c~ion can be any type of conventional protocol. ~ltho~lgh a wireless system is illustrated, it should be understood that a fixed wire system could be lltili7e~
In addition to training the system as described above lltili7:ing techniques such as 30 back propagation, G~ n bars, radial basis functions, etc., the network could have the parameters thereof offset or a bias adjust applied thereto. After training a neural nc;lwulk in the normal manner, a situation could occur wherein the average of the plant Sl~S~T~TE SH~E~ ~Rl I~E 26~
wo 95/04878 2 1 6 7 9 2 7 PCT/US94/08657 output drifts by some amount. This situation is illustrated in FIGURE 18, wherein a solid curve is illustrated replesç~ the actual emissions output measured by an external emissions monitor and the dotted line represents the predicted output from the virtual sensor. It can be seen that there is an average error that occurs over time. This average 5 error is determined and utilized to determine an offset, the average error defined as follows:
( ~ ) = ( o" - op ) (7) where: o~ is the actual plant output; and op is the output value predicted by the model.
The averages are deterrnined over all of the examples in some past window of time.
10 Adding this average error to the model output results in the following:
o/ = op + ( e ) (8) such that the new average error of the model is zero, as follows:
( e ~ o~ ~ / ) = ( a ) ~ ( p ) ~ ( a ) + ( p ) = ( 9 ) This bias adju.stm~nt must be made periodically or in an ongoing manner, l~tili7:ing moving averages.
In another parameter adjl-stment system, a first principles model could be 15 utili7ed First principle models are well-known and rely upon the fact that there are only a small number of controls that can be tuned or adjusted on the internal combustion - engine. Typically, this tuning or control takes the form of adjusting parameters, i.e., co-efficients in the model to ...;..;...;~e the error that exists between the model output and the actual plant or engine output. This procedure is completely analogous to Ll~illing a 20 neural network model. The differences between the two training operations occur in the method that the p~ e.~ appear in the model and perhaps the way in which they are SU~S~lr~TE s~r (~l~LE ~6~
_ WO 95/04878 2 ~ 6 7 9 2 7 PCT/US94/08657 adjusted. First principle models are often simpler than neural network models in that they are often linear and have fewer adjl.~tment parameters. An example ofthe first prinr.iple models applied to NOx emissions from an internal combustion engine is as follows:
NOxppm = k ( Tc Tm ) ( 10 ) where: Tc is the cylinder te..lpe~ re;
Tm is the manifold telll~ L-Ire and the average of the difference therebetween is taken over all cylinders; and k is the adjustable parameter.
The pal ~llle~er k can be adjusted in the same iterative manner that one adjusts the lO pal~lllelel~ in a neural network model, i.e., by gradient descP.nt; that is, one minimi7es the overall error between the model and the plant as follows:
( a ~ ~p )2 = 2 ~ e2 (11) 2 e~ampl eS e~ampl es via iteratively ~h~nging k according to the gradient descent equation:
/~K = -n ~ (12) In the present case, this is simply:
~K = ne ( To - T,~) (13) In the above equation, a simple control example would be a situation wherein if the NOx l 5 output was high, a signal would be issued that would result in red~lcing the air-fuel ratio by reducin~ the air and increasing the fuel.
SU~S~TU~E SHEEr (~ULE ~6~
wo 95/04878 2 1 6 7 9 ~ 7 PCT/US94/08657 Rt;relling now to FIGURE l9, there is illustrated a flowchart depicting the overall runtime control operation. The program would be initi~ted at a start block 386 and the~ proceeds to a function block 388 to compare the expected emissions value to an internal runtime threshold, it being remembered that these thresholds can be selected by 5 the user for a particular cirCllm~t~n~e The program then flows to a decision block 390 to determine if the expected or predicted emissions exceed the threshold. If so, this inllic~tes that some control operation or failure mech~nicm must be performed. The program would flow to a decision block 392 to determine if a control operation is present. If so, the program would flow along a "Y" path to a function block 394 to l0 adjust the controls in accordance with the predetermined control operation which could utilize the first principles operation as described above, or a control network. The program would then flow to a decision block 396 to determine if it is possible to effect an a~prop~iale control. In order to determine this, maximum limits are set within the operating parameters of the engine by which the controls can be modified to reduce 15 çmi.e.cionc If these control limits are exceede-l, the engine operation will deteriorate, even though the emissions was applopliaLe. If the maximum control limit has beenexceeded, the program would flow along a "Y" path to a function block 398 to set the default settings for the emissions control parameters and then to a function block 400 to store the history. If the m~im-lm control limit had not been exceede(l the controls 20 would be accepted and the program would flow along an "N" path from decision block 396 to the function block 400. If no controls were available, the program would also flow to the function block 400 from the decision block 392 along the "N'' path thereof.
Once the history has been stored as to whether a control operation had been effected or a simple 1 ~;co~ dillg of the emissions history had been made, the program 25 would flow to a decision block 402 to determine if an alarm operation were present.
This could occur in the event that the co~pa-ison made at decision block 390 required an alarm to be set or if the presence of the default setting in the function block 398 required an alarm to be set. If an alarm is to be set, the program would flow along the "Y" path from decision block 402 to a function block 404 to set the alarm, this possibly 30 being a light on the dashboard or an audible alarm. Further, this could be the existence of an alarm communication to a central station requiring a m~int~n~nce check. The program would then flow to a return block 406. If the alarm operation had not been SVSSTITUTE Sl IEET (RULE 26) WO 9s/04878 2 i 6 7 q 2 7 PCT/[TS94/08657 present, the program would flow from the decision block 402 along the "N" path to the return block 406. Similarly, if the predicted emissions did not exceed the threshold, the program would flow to the return block 406 along the "N" path from the decision block 390.
Referring now to FIGURE 20, there is illustrated a flowchart depicting the training operation. The program is initi~ted at a start block 410 and then proceeds to a function block 412 to measure the actual emissions. The program then flows to a function block 414 to comp~e the measured actual emissions to the expected or predicted value output by the virtual sensor. The program then flows to a decision block 416 to determine if the difference between the measured actual emissions and thepredicted value exceed the threshold. If so, the program wouLd flow along a "Y" path to a decision block 418 to dt;t~ e if a training operation is to be performed. If not, the program would flow along an "N" path to a function block 420 to update a record ~at~ha~e and then to a return block 422. Similarly, if the error did not exceed the threshold, the program would flow from the decision block 416 along the '~' path to the function block 420.
If a training operation is to be pel ~I llled, the program would flow from the decision block 418 along a "Y" path to a decision block 424 to determine if a full training operation is to be performed. If yes, the program would flow along a "Y" path to a function block 426 to retrain a complete model and then to a function block 428 to download the model p~ll~Lers back to the memory associated with the model pal~~ . The program will then flow to a return block 430. If a full training operation were not to be p~l~lllled, i.e., a partial training operation only, the program would then flow along an '~' path from decision block 424 to a function block 432 to upload the model pal~llllGLt;l~ that were in the system and then adjust these model parameters as intlic~ted by a function block 434. After adjll~tm~nt, the program would flow to the input of the function block 428 to download the model parameters. It should be understood that the training operation can be an on-board operation utili7.in~ the model processor or an off-board operation lltili7ing an external processor.
su~sTIruTI~ s~ ULE 7B~
WO 9S104878 2 1 6 7 9 2 7 PCT/uS94/086s7 l[n summary, there has been provided a system for predicting emissions in order to obviate the need for an actual emissions sensor on a reciprocating engine. The system utilizes a predictive model that is trained on various sensor outputs from the reciprocating engine with a target output of the actual emissions taken during generation 5 of the training data set. Once trained, the system can operate without the actual emissions sensor, but represents the actual output of the emissions sensor. Periodically, the model can be checked to determine if the predicted value has deviated more than a predeLe~ ed amount from an actual measurement by pelrolll~ing an actual measurement with an external emissions sensor. If the measurement has deviated, the 10 system can be in-~ic~ted as out of spec or the network can be retrained.
Although the plerelled embodiment has been described in detail, it should be understood that various changes, substitutions and alterations can be made therein without departing from the spirit and scope of the invention as defined by the appended claims.
SUBSTITUTE SHEET (RULE 26)
VIRTUAL EMISSIONS MONITOR FOR AUTOMOBILE
TECIINICAL ~IELD OF THE INVENTION
The present invention pertains in general to emissions monitoring systems, and more particularly, to a system that replaces the continuous emissions monitor on a reciprocating engine with a virtual sensor.
CROSS REFERENCE TO RELATED APPLICATION
This application is a continuation-in-part of U.S. Patent Application Serial No. 08/102,405, filed August 5, 1993, and entitled "VIRTUAL CONTINUOUS
E~ISSION MONITORlNG SYSTEM WITH SENSOR VALIDATION" (Atty. Dkt.
No. PAVI 21,874).
wo 95/04878 2 1 6 7 9 2 7 PCT/US94/08657 BACKGROUND OF T~E INVENTION
As public awareness increases with respect to the environment, industry is required to make ~i nific~nt ch~nges Although industry is somewhat responsive topublic opinion, govell--llt;llL regulatory bodies are typically brought in to ensure that 5 public needs are met. In order to do this, government sets up regulatory arms of already eYi~ting branches of entities such as the En~ri.o"".~ l Protection Agency. These arms are given the task of putting in place policies regarding toxic waste, emissions, etc., that may effect the environment. Further, these regulatory bodies are also given the task of el~l~;ing these regulations. One particular area that has received a great deal of 10 attention in recent years is that of monitoring emissions of noxious gases being placed into the atmosphere by m~mlf~ctnring f7~t~.iliti~s Typically, the teçhniqlle for ensuring that noxious gases are being co-.e.;Lly m~ ol ed has been to implement Continuous F.mi~ ns Monitoring systems (CEM).
These systems are utilized to monitor the amount of emissions such as Sulfur Dioxide 15 (S2)~ Nitrogen Oxides (NOx), Carbon Monoxide (CO), Total reduced Sulfur (TRS), opacity, Volatile Organic Carbon (VOC), and hazardous subsLances of all sorts. The rl~c~ic~l way of monitoring these emissions is to install a Continuous Fmi~ions Monitor (CEM) in the plant on each emissions point source. Re~ tory Agencies provide foreach plant ~ elines as to how the output is to be re~ te~l, i.e., define the acceptable 20 limit ofthe emissions. With respect to a reciprocating engine, the EPA has set ~lid~lines as to the p~;-~----ance of the engine. Once m~mlf~ctllred, the engine is expected to meet these ~lid~linçs over the life ofthe engine, a~s~lming that it is plopell~ d However, new re~ tic)ns are being implem~nte~l that require some type of c~ntimlQus moniLo.i.~ to be implem~nted with periodic l~.h~c~lps to ensure that the monitor is 25 operating co-le-;Lly.
The classic CEM is composed of either an in situ analyzer installed directly in the stack, or the exhaust pipe of the reciprocating engine, or an extractive system which extracts a gas sample and conveys it to an analyzer at grade level. However, these sensors are quite expensive, difficult to m~int~in, and difficult to keep properly SU~S~U.~ SI~EE~ (R~LE 26) WO 95/04878 2 1 6 7 ~ 2 7 PCT/US94/08657 .
calibrated. As such, the reg~ tions that deal with a CEM system require the sensors to be calibrated frequently, which calibration procedure can take a number of hours, due to the complexity thereo Re~ tions allow a m~imllm downtime of ten percent for calibration. If a unit remains in operation greater than ten percent of the time with the 5 CEM down, the emissions level is con~idered by the Regulators to be at ~ al potential level. This results in out-of-compliance operation. Most m~nllf~ctllres will shut down operation rather than face the high penalties of such occurrence. One of the reasons for this is that the operation of the plant relative to the monitoring of the NOx emissions must be "truly continuous" such that no leeway is provided for faulty sensors, 10 sensors that have fallen out of calibration, etc. One solution to this has been to utilize redlln~n1 sensors, which is a very expensive solution. Therefore, there exists a need to provide a system that does not require the presence of a sensor while still ensuring that the output of the plant is within tolerances relative to noxious emissions.
T~TUT~ S~EET ~ULE ~
;.. ~
SUMMARY OF l~ INVENTION
The present invention disclosed and claimed herein comprises a method for monitoring emissions in an internal combustion engine that emits noxious pollutants and has associated ~I.el~;wiL}l a plurality of sensors for measuring select parameters of the 5 engine operation, these measured parameters provided as sensor output values. A
predictive model is provided in which a stored representation of the engine combined with an external emissions monitor is reprç,s~nted The predictive model is operable to output a predicted emissions value corresponding to the output of the external emissions monitor when ~tt~chPd to the engine, with the inputs to the predictive model 10 corresponding to select ones of the sensor output values. The select ones of the sensor output values are then input to the predictive model and a prediction is provided of the noxious pollutants during operation of the engine without the requirement of the external emissions monitor being connected to the engine. This is achieved during runtime-without a 1 ullLh~le monitor for the emissions.
In another aspect of the present invention, the operation of the predictive model is verified by ~tt~.hing an external emissions monitor to the engine to measure the noxious pollutants output thereby. The measured output is then co",pa, ~d with the predicted output. If the actual and predicted values differ by more than a predeterrnined amount, the stored represçnt~tit n is adjusted. The stored representation is generated 20 initially by training the predictive model on a training set of data that comprises as input data the actual sensor output values and as target output data the actual measured emissions output of the external emissions monitor. In one method of adjusting the stored l~s~ ;nn~ the predictive model is retrained to complete a new stored rep,es~ l;t n on a new set oftraining data. In another method, the predicted emissions 25 value is merely offset on the output of the predictive model such that the average difference between the offset predicted emissions value and the actual measured emissions value is less than a predetermined difference.
In yet another aspect of the present invention, the predicted emissions value output by the predictive n~lwol~ is colllpaled with an intemal threshold value. When the SUBSTITUT~ SH~ET (RULE 26) wo 95/04878 2 1 6 7 9 2 7 PCT/USg4/08657 predictive emissions value exceeds the internal threshold value, an alarm is generated. In one embodiment, this alarm is a light. The threshold value can be selected from a plurality of stored threshold values, which threshold value is selected from the plurality of stored threshold values in accord;~ce with predeterrnined criteria.
S l[n a further aspect of the present invention, a control system is provided for modif~ring the opel~Lillg pa,~.,t;Le,~ ofthe engine to adjust the output noxious polll-t~n~.e An emissions control system is provided which operates in response to receiving the predicted emissions value output by the predictive model and co",pal;ng it with a desired output level. The difference between the two levels is l";.~ ed by the control system.
SU~STIT~ S~ Rll~ 26) WO 95/04878 2 1 6 7 9 2 7 PCT/USg4/08657 BRIEF DESCRIPTION OF THE DRAWINGS
For a more complete underst~ntling of the present invention and the advantages thereof, reference is now made to the following description taken in conjunction with the acco.l.pa,.ying Drawings in which:
FIGURE 1 illustrates an overall block diagram of the virtual sensor of the present invention;
FIGURE l a illustrates a diag~ . " - A I ic view of the sensor validation system;
FIGU~E 2 illustrates a block diagram of the relation of the virtual sensor and the control system;
FIGURE 3 illustrates an embodiment llti~ ing a single,control network;
FIGURE 4 illustrates a diagrammatic view of a conventional neural network;
FIGURE 5a illustrates a more detailed block diagram of the control network;
FIGllRE 5b illustrates a detail of the iterate operation of FIG~RE 5a;
FIGURE 6 illustrates a detail of a typical plant, a boiler for a steam generation 1 5 facility;
FIGU~E 7 illustrates a block diagram of the sensor validation network;
FIGURE 8 illustrates a diag-~l--,--dlic view of the auto associative predictive network utilized in the system of FIGURE 7;
FIGUREs 9a and 9b illustrate plots of predicted versus actual pollutant sensor values and the dirrelence therebetween;
FIGlJREs 10a and 10b illustrate the plots of FIGUREs 9a and 9b, respectively, wherein one of the sensors is faulty;
FIGU~E 11 illustrates a flo~ ~ L for operating the overall system;
FIGURE 12 illustrates a flowchart for the sensor validation operation;
FIGURE 13 illustrates the prere--ed embodiment ofthe present invention wherein the virtual sensor is utilized in conjunction with an internal combustion engine;
FIGURE 14 illustrates a block diagram of a system wherein an external emissions sensor is utilized for llaini,lg, FIGIlRE 15 illustrates an alternate method wherein the rel.ai,ling operation is done external to the system with a reLI~ g processor;
SUBSTI~UTE ~!~E~T ~RULE 26) wo 95/04878 2 1 6 1 9 2 7 PCT/US94/08657 lFIGURE 16 illustrates a block diagram depicting the runtime operation ofthe internal combustion engine;
FIGURE 17 illustrates an overall view of the communication system;
FIGURE 18 illustrates a plot ofthe actual emissions output measured by an 5 external emissions monitor and the predicted output from the virtual sensor;
FIGURE 19 illustrates a flowchart depicting the overall runtime control operation; and FIGURE 20 illustrates a flowchart depicting the training operation.
SUBSn~l~TE St~tET (~ULE 26~
DETAILED DESCRIPTION OF THE INVENTION
Referring now to FIGURE 1, there is illustrated an overall block diagram of the system of the present invention. A plant 10 is provided that, during the normal operation thereof, releases some emissions 12 con~ g some level of pollutants. The pollllt~nts 5 12 are moniLoled by a pollutant sensor 14 or by utilization of EPA established referenGe methods, which sensor 14 is illustrated in phantom, to provide continuous emissions l.lol ilo~ g. This is rere.-ed to as a CEM. As will be described hereinbelow, the present invention provides a virtual sensor operation wherein the pollutant sensor 14 is only required for initial training of virtual sensor network. The pollutant sensor 14 is utilized 10 to gather training data to be combined with the control values and sensor values that are available to a Distributed Control System (DCS) 16, generally refe.-ed to as the plant h~...aLion system. The DCS 16 provides control values associated with control inputs to the system and sensor values to a computer 15. The computer 15 is comprised of a virtual sensor network 18 that ~ss~nti~ly provides a non-linear represent~tiQn of the 15 plant 10, which non-linear representation is a "learned" leprese..Lalion. The virtual sensor network 18 is operable to receive run time inputs 20 from a sensor validation system 22. The sensor validation system 22 is operable to receive actual measured inputs 24 from the plant 10 through the DCS 16. These measured inputs leplesellLmeasured state variables of the plant in the form of sensor values and also control values 20 that are input to the plant to provide control therefor. As will be described hereinbelow, the various inputs 24 are provided as inputs to the virtual sensor network 18 through the DCS 16. However, some of these inputs may be faulty and the sensor validation system 22 is operable to generate an alarm when any of the ~tt~rh~d sensors fails and to replace failed sensor values with reconciled sensor values.
The virtual sensor network 18 is operable to receive the inputs 20 and predict plant controls and alarms. The virtual sensor network 18 can predict what the pollutant levels are that normally would be monitored by the pollutant sensor 14; hence, it provides a virtual sensor. The sensor network 18 is a network that can be trained with a training system 28. The training system 28 utilizes as a target the actual pollutant level SL~ST~T~ E~RU~
wo 95/04878 2 1 6 7 9 ~ 7 P~T/US94/08657 on a lme. 13 as measured by the pollutant sensor 14 when it is present, and also the inputs 24 from the plant 10. The difference between the predicted pollutant level on a line 17 and the actual pollutant level on line 13 generates an error on line 19 that is used by the training system to adjust the stored representation in the virtual sensor module, so as to 5 ~ illl;,e the error. In operation, as will be described in more detail hereinbelow, the pollutant sensor 14 is a Continuous Fmie,eions Monitor (CEM) that is operable to be temporarily connected to the plant 10 to monitor the level ofthe pollutants 12. This provides a target to the training system 28. The network 18 is then trained with both the measured plant sensor and control values, not including the CEM output, and the CEM
10 output when present. This information is utilized to generate a training dataset.
~ fter training, the pollutant sensor 14 is removed and then the system operates by pretlicting what the output of the CEM or pollutant sensor 14 would be. The virtual sensor network 18 then replaces the pollutant sensor 14 and then can be utilized in a control function to predict plant control/alarms to ",~;"~ the operation ofthe plant 10 15 within acceptable standards. Further, the virtual sensor nelwolk 18 can be used solely to provide an output in place of the pollutant sensor 14 that can be utilized by the operator of the sensor to ensure that all nec~.es~ry procedures are being followed to ensure that the level of pollutants is within acceptable ranges. For example, if the predicted output from the n~lwolk 18 Pxceeded one ofthe established guidelines or 20 thresholds, the operator would then follow certain prescribed procedures to correct the situation. This would be the case even if the pollutant sensor 14 were present. The advantage to this is that the relatively expensive and difficult to Ill~ pollutant sensor 14 would not have to be present. Further, a new pollutant sensor 14 or a portable pollutant sensor 14 is periodically utilized to check the operation of a virtual sensor 25 network 18 to ensure that it is opel~ling colle-illy and that no parameters ofthe plant have ~h~ed such that the prediction is now incorrect or the model no longer represents the plant. In this manner, the system would have to be l ~LI ~h~ed by using a new set of - training data that would be provided by the operation of the connecting the pollutant sensor 14 to the plant lO. This could be the situation wherein some measurement device 30 degraded or the plant itself had physically changed parameters due to capital improvements, age, etc.
SI~B~T~U~ S~ lJL~ 26) WO 95/04878 2 ~ 6 7 S 2 7 PCT/US94/08657 In another mode of operation, the pollutant sensor 14 may be in a situation where it might be removed from the plant 10 for calibration purposes. During this time, the virtual sensor network 18 is then utilized to replace the sensor 14 during the calibration procedure.
Referring now to Figure 1 a, there is illustrated a block diagram of the operation ofthe sensor validation system 22. A plurality of sensors 27, 29, 31, 33 and 35 are illustrated. Each of the sensors 27, 29, 31, 33 and 35 have an output that is connected to the input of the virtual sensor 18. Additionally, each of the outputs is connected to an evaluation system 37 to d~Le~ ine if the sensor is valid, as will be described hereinbelow.
When any one of the sensors 27, 29, 31, 33 and 35 is determined to be faulty, it is replaced by a substitute sensor 39, which is a predicted sensor value that predicts the output of the faulty sensor ~Itili7ing a stored representation of the faulty sensor, which stored representation is a function of the other sensors 27, 29, 31, 33 and 35.
The.erore, the substitute sensor 39 requires as inputs the outputs ofthe valid sensors and the predicted output of the substitute sensor. This is illustrated in Figure la with the sensor 29 being substitl~ted~ with the substitute sensor 39 receiving as inputs the outputs of the sensors 27, 31, 33 and 35 and, in place of the output of the sensor 29, the predicted output of the substitute sensor 39. Further, another sensor could be substituted for with the output of the substitute sensor 39 being an input for the new and additional sensor (not shown).
Referring now to FIGURE 2, there is illustrated a block diagram for the operation wherein a virtual sensor predictive network 32 is provided which is operable to receive measured plant sensor values s(t) from the plant 10 and also the control values ~(t) which are inputs to the plant 10. The virtual sensor predictive network 32 is operable to output a predicted virtual sensor value oP(t) for input to a multiplexer 34.
The sensor value from sensor 14 is input on the line 36 to the multiplexer 34. The m--ltipl~oxçr 34 is operable to select either the predicted output of the network 32 or the actual output ofthe sensor 14 for input to a control system 38. The control system 38 is operable to generate the input values x(t) to the plant 10. The multiplexer 34 replesen the operation wherein the output of the network 32 is utilized to replace that of the sensor 14.
SlJ~TITU~E S~T (RU~ ~ 26) WO 95/04878 2 ~ 6 7 9 2 7 PCT~Sg4/08657 .
Rere,lh~ now to FIGURE 3, there is illustrated one embodiment of the system wherein a dynamic control system is provided. In this system, a control network 40 is provided which receives as an input the control input values x(t) and the sensor values s(t), the sensor values s(t) comprise the measured plant variables such as flow meter 5 measurel.lenLs, temperature measule",~nls, etc. In addition, the control net 40 is operable ~o receive a desired output value as one of the inputs. The control net 40 contains a stored representation of the plant and is operable to output a set of control input values x(t+l). These are input to a Distributed Control System (DCS) 42, which is operable to generate the control values ~(t). The control network 40 is a conventional 10 control network that is trained on a given desired input, and which control network 40 is operable to receive the sensor values and control values and generate the updated control values x(t+l) that are necessary to provide the desired outputs. . The control network 40 is generally comprised of a neural network having associated therewith weights that define the l~les~ liQn that is stored in the neural network. In the embodiment of 15 FIGURE 3, these weights are frozen and were learned by training the control network 40 on a given desired output with a given set of training data for the control values x(t) and the sensor values s(t). A desired output is provided as one input for selecting between sets of weights. The general operation of control nets is described in W.T. Miller, III, R. S. Sutton and P.J. Werbos, '~eural Networks for Control", The MITPress, 1990,20 which reference is incorporated herein by reference.
Re:ferring now to FIG~JRE 4, there is illustrated a detailed diagram of a col.vellLional neural network comprised of input nodes 44, hidden nodes 46 and output nodes 48. The input nodes 44 are comprised of N nodes labelled xl, X2, ... XN, which are operable to receive an input vector x(t) comprised of a plurality of inputs, INP1(t~, 25 INP2(t), ... INPN(t). Similarly, the output nodes 48 are labelled o" 2~ OK, which are operable to generate an output vector o(t), which is comprised of the output OUTl (t), OUT2(t), ... OUTK(t). The input nodes 44 are interconnected with the hidden nodes 46, hidden nodes 46 being labelled al, a2, .. . an, through an intel col~le~;lion network where each input node 44 is interconnected with each of the hidden nodes 46. However, some 30 intelc(~ e~iLion schemes do not require full interconnection. Each ofthe intel~iomle.;
has a weight Wij'. Each of the hidden nodes 46 has an output oi with a function g, the output of each of the hidden nodes defined as follows:
Sl~S ~ ~TU~E S~EET (RULE ~6) -Wil; Xi + b~ ) (1) Similarly, the output of each of the hidden nodes 46 is interconnecte~l with subst~nti~lly all of the output nodes 48 through an interconnect network, each of the interconnects having a weight Wjk2 associated therewith. The output of each of the output nodes is defined as follows:
k = g(~; W~k aJ + b2) (2) 5 This neural network is then trained to learn an function f(x(t), P) as follows:
o ( t) = f (~ ( t), P) (3) where o(t) is an output vector and P is a vector or p~Lers ("weights") that are variable during the learning stage. The goal is to ~ .e the Total-Sum-Square-Error filnr,tion -( t) -o ( t) ) 2 (4 ) The Total-Sum-Square-Error function is ",;n;l.l;~ed by rh~nging the parameters P ofthe 10 function f. This is done by the back propagation or a gradient descent method in the p~c;rt;;lled embodiment on the parameters Wjk2, Wjjl,bl" b2k. This is described in numerous articles, and is well known. Ther~role, the neural neLwulk is ess~nti~lly a ,u~lleLer fitting scheme that can be viewed as a class of st~tictir~l algo,iLlulls for fitting probability distributions. Alternatively, the neural network can be viewed as a filnction~l 15 appl u~hllhLor that fits the input-output data with a high-~limP.n~iQnal surface. The neural n~Lw~lk utilizes a very simple, almost trivial fiunction (typically sigmoids), in a multi-layer nested structure The neural network described above is just one example. Other types of neural networks that may be utilized are those using multiple hidden layers, radial basis Sl IBST~TUT~ SHE~T ~RULE 7~) WO 95t04878 2 1 6 7 9 2 7 PCT/US94/086S7 functions, ~ n bars (as described in U. S. Patent No. 5,113,483, issued May 12, 1992, which is incorporated herein by reference), and any other type of general neural - nt;Lwo~k. In the plere.led embodiment, the neural network utilized is of the type referred to as a multi-layer pelc~Lloll.
Rt;rt;llh1g now to FIGURE 5a, there is illustrated a block diagram of a control system for opl;",;~l;on/control of a plant's operation. The plant 10 has an input for receiving the control values x(t) and an output for providing the actual output y(t) with the sensor values s(t) being associated therewith, these being the internal state variables.
A plant predictive model 74 is developed with a neural network to accurately model the plant in accordance with the function f(x(t),s(t)) to provide an output oP(t), which represell~s the predicted output of plant predictive model 74. The inputs to the plant model 74 are the control values x(t) and the sensor values s(t). For purposes ofo~ i ;Qn/control, the plant model 74 is deemed to be a relatively accurate model of the operation ofthe plant 72. In an opl;~ ;on/control procedure, an operator independently generates a desired output value od(t) for input to an error generation block 78 that also receives the predicted output oP(t). An error is generated between the desired and the predicted outputs and input to an inverse plant model 76 which is ntic~l to the neural network repreS~ntin~ the plant predictive model 74, with the exception that it is operated by back propa~tin~ the error through the original plant model with the weights of the predictive model frozen. This back prop~tion of the error through the network is similar to an inversion of the network with the output of the plant model 76 repres~ontin~ a ~x(t+l) utilized in a gradient descent operation illustrated by an iterate block 77. In operation, the value /~ ~(t+l) is added initially to the input value ~c(t) and this sum then processed through plant predictive model 74 to provide a new predicted output oP(t) and a new error. This iteration continues until the error is reduced below a predetermined value. The final value is then output as the new predicted control values x(t+1).
This new x(t+l) value comprises the control values that are required to achieve the desired actual output from the plant 72. This is input to a control system 73, wherein 30 a new value is presented to the system for input as the control values x(t). The control system ~3 is operable to receive a generalized control input which can be varied by the ~ U~ S~-ET ~ LE ~
distributed control system 73. The general terminology for the back propagation of error for control purposes is "Back Propagation-to-Activation" (BPA).
In the plt;r~;llcd embodiment, the method utilized to back propagate the error through the plant model 76 is to utilize a local gradient descent through the network 5 from the output to the input with the weights frozen. The first step is to apply the present inputs for both the control values x(t) and the sensor values s(t) into the plant model 74 to generate the predicted output oP(t). A local gradient descent is then pc~ru~ ed on the neural n~Lw~Jlh from the output to the input with the weights frozen by inputting the error beL~eell the desired output od(t) and the predicted output oP(t) in lO accordance with the following equation:
A~(t) = Z(t + 1) - ~(t) = I~ a(O-d~t) - oP(t))2 ~5) where 1l is an adjustable "step size" parameter. The output is then regenerated from the new x(t), and the gradient descent procedure is iterated.
Referring now to FIGURE 5b, there is illustrated a detailed block dia8ram of theiterate block 77. The iterate block 77 is comprised of a sll".",il-~ junction which is 15 operable to receive the ~x(t+l) input and the output of a multiplexor/ latch block 86.
The multiplexor/latch block 86 is operable to receive both the output of the summing junction 84 for feedb~rk as one of the inputs and the control variable x(t). The output of the sllmming block 84 is the sum of the previous value of x(t) plus the new iterative change value /~x(t). This will then be iteratively s -mmçd with the previous value to 20 gencl ~le a new iterative value until the error is at a predetermined level. At this point, the output of the sl~mming junction 84 will comprise the new control value x(t+1).
Another standard method of o~ A I ion involves a random search through the various control values to minimi7e the square of the difference between the predicted outputs and the desired outputs. This is often referred to as a monte-carlo search. This 25 search works by making random çh~n~çs to the control values and feeding thesemodified control values into the model to get the predicted output. The predicted output is then coll~ared to the desired output and the best set of control values is tracked over S~BSTITU~E S~ IL~ 2~) .
the entire random search. Given enough random trials, a set of control values will be obtained that produces a predicted output that closely m~t~hes the desired output. For reference on this technique and associated, more sophisticated random o~ i7AI;ontechniques, see the paper by S. Kirkpatrick, C.D. Gelatt, M.P. Vecchi, "O,ul;...;~;1lion by Simul~ted ~nne~lin~" Science, vol. 220, 671-780 (1983), which reference is incorporated herein by lt;fe.t;llce.
Referring now to FIGURE 6, there is illustrated a diagrammatic view of a typical plant that may exist at a m~n~f~ctl~ring facility. The plant typically comprises a boiler 92 which has a firebox 94 disposed at the lower end thereo The boiler 92 interfaces with a stack 96 through a preheat chamber 98. Many tubes of which tube 100 is typical thereof are operable to run through the chamber 98 and enter the boiler 92. The tube 100 then passes in a sel~c;llLille manner through the boiler 92 to an output pressure vessel 104, which is pressurized. The vessel 104 is operable to generate steam out of an outlet 106.
The other end ofthe tube 100 that enters the chamber 98 is connected to a source 108 of the deionized water. In operation, the water is passed through the tube 100 to the challlbel 98, which picks up heat therein and then into the main boiler 92, where it is heated further. This then passes through to the vessel 104. The firebox 94 has a heating ment 116 associated therewith that is operable to receive gas through a gas line 118 and air through an air line 120. The mixture of the gas and the air allows the heating el~m~nt 116 to generate heat in the firebox 94 and heat up the water in the tube 100 within the boiler 92.
The tube 100, when it exits the source 108 with the deionized water at the source, has the flow thereof measured by the flow meter 122. A valve 124 allows control ofthe flow of fluid from the source 108 into the challll)el 98. Two telllpel~LLIre sensors 126 and 128 are provided at di~lt;lll locations along the tube 100 within the ch~llbel 90 to provide temperature measurements therefor. Additionally, temperature - sensors 130, 132 and 134 are provided along the tube 100 at various locations within the main boiler 92. A temperature sensor 136 is provided for the firebox 94. The level of the fluid within the pressure vessel 104 is measured by a level meter 142 and the pres~ule therein is llleasllled by a pressure meter 146. A flow meter 150 is provided formeasuring the flow of steam out ofthe pressure vessel and a control valve 152 provides SUB~T~TE S~EET (~ULE 2~) WO 95/04878 2 ~ 6 7 9 2 7 PCT/US94/086~7 control of the steam exiting the pressure vessel 104. The heater element 116 is controlled with a valve 158 on the gas line, which has the flow thereof measured by a flow meter 160. The flow meter on the air line 120 is measured by a flow meter 162. A
damper 163 in the stack 96 is utilized to control air flow through the firebox 94.
It can be seen that the sensor values s(t) of the plant are provided by the various te~ el~LLIre and flow measurement devices. Further, the control values, in the form of the various valves and damper positions provide the control values to the plant.Thel ;;fol e, an operator can control the operation of the plant by controlling the various 10 flow meters and other control values, some of which are not illustrated. The l~ it~g inputs that are necess~ry in order to provide adequate control of the plant for the purpose of continuous emissions monitoring are the NOx levels. These are provided by the virtual sensor network 18 of FIGURE 1. However, as described above, periodically a portable unit 170, having disposed thereon a CEM 172, is conn~cted via a duct 174 to 15 the stack 96 to measure the amount of NOx in the output emissions to the air. The CEM
172 then generates a report as to the level of the NOx. If this level is within acceptable standards, then this is merely reported. However, if the level is outside of acceptable limits, this is reported to the plant operator and either ~ ges are made or the plant is shut down. Additionally, the information generated by the CEM 172 is generated on a 20 time base and this comprises training data. This training data, since it is on a common time base, can then be colllbilled or merged with data associated with the sensor values and the control values, which are also on a time base, to provide new training data for the virtual sensor network 18. This can be utilized by the training system 20 to retrain the virtual sensor network 18, if nPce~ y.
R~ft;llhlg now to FIGURE 7, there is illustrated a block diagram ofthe pr~fell~dembodiment for the sensor validation system 22. To ensure that the overall inputs ~(t) to the network 18 are "valid", it is nece,ss~ly to perforrn some type of conlpalison with expected or predicted values. If it is suspected that the generated values are not accurate, then an alarm is generated to advise the plant operator or the control system of the need to calibrate the sensor or to repair the sensor, and an estim~ted or predicted value for that sensor value is substituted for the actual measured value of the sensor.
SIJ~STITUTE SHEET ~RULE 2~
.
In a pr~rel,ed embodiment, an auto associative predictive neural network 180 is provided which is a network having an input layer for receiving select ones of the inputs s(t) on an input 182 Although not illustrated, only certain ones of the actual sensor values are nece~saly as inputs to the virtual sensor network 18 in order to provide an accurate prediction of the NOx levels that would generally be provided by the pollutant sensor 14. These are deLellllilled by p~lrolllling a sensitivity analysis. This is described in U.S. patent application Serial No. 056, 197, filed April 30, 1993 and entitled "Method and Apparatus for Detel n,il~ng the SensiLi~/iLy of Inputs to a Neural Network on Output Parameters" (Atty. Dkt. No. PAVI-21,761), which is ~ ned to the present Assignee.
By utili7:ing the sensiLiviLy analysis, the number of inputs to the neLwul k 18 can be significantly reduced and only the important inputs ~Itili7ed This ~i~3nifit~ntly reduces the size of the auto associative predictive network 180 and also the virtual sensor network 18.
The actual inputs X(t) are input to a multiplexer 186 which is operable to select between the predicted inputs ~CP(t) output by the nc;Lwol k 180, which is a predicted output, and the actual inputs ~C(t). In operation, a first cycle occurs when the multiplexer selects the actual inputs x(t). The predicted inputs xP(t) are then input to a subtraction circuit 188 to deLelll,il~e the difference between ~C(t) and ~P(t). This di~rei~ce is input to coll,pal~tor 190 for co",pa~ison with thresholds stored in a threshold memory 192. The one of the actual inputs to the network 180 having associated therewith the largest error as colll~)are-d to the acceptable threshold is then conn~cted to the associated predicted output of the network 180. The actual inputs X(t) with the substituted or lecol~l~ecLed input is then again cycled through the auto associative predictive network 180. On this next cycle, the difference between the actual and the predicted values are againdetermined, colll~a~ed with the thresholds, and the one ofthe actual inputs having the largest error is reconnected to the associated predicted input by the multiplexer 186.
This continues until all of the predicted inputs, with the determined faulty or un~cceFtable actual values replaced with the predicted values output by the network 180, are within a predetermined range. Once this has occurred, the predicted values from the n~Lwolk 180 are input to a multiplexer 196, and the multiplexer 196 selectin~ for output the, ~r, O"~ the actual values that were determined to be acceptable and the predicted values as a substitute for the actual values that were determined to be unacceptable. It SI~T~ TE SH~E~ (RlJLE 26) WO 9~/04878 2 1 6 7 9 2 7 PCT/US94/08657 should be noted that the predicted values are generated by running the network with the determined ~n~cceptable actual values replaced with the associated predicted values by the multiplexor 186. The output of the multiplexor 196 is then input to the virtual sensor network 18.
In another embodiment, the predicted input values output by the auto associativepredictive network 180 can be provided as the input to the virtual sensor network 18.
This would then not require the multiplexer 196 and, in fact, the auto associative predictive ntLwolk 180 can continually monitor and replace ones ofthe sensor inputs that are determined to be invalid.
Referring now to FIGI 1RE 8, there is illustrated a ~ g, ~ ;c view of the auto associative predictive network 180. The network is comprised of an input layer of nodes 198 and an output layer of nodes 200. There is one node in the layer 198 for each ofthe input vectors ~(t), illustrated as x,(t), x2(t) ... xn(t). Similarly, there is a single node for each of the predicted output variables xP(t) such that there are outputs xlP(t), x2P(t) xnP(t). The input layer of nodes 198 is mapped through to the output layer of nodes 200 through a hidden layer of nodes 202. The hidden layer of nodes 202 has a plurality of intelco~ ;o~.~ with each of the nodes in the input layer of nodes and each of the output layer of nodes 200. Each of these interconnections is w~ightetl Further, the number of nodes in the hidden layer of nodes 202 is less than the number of nodes in either the input layer 198 or the output layer 200. This is therefore lt;~lled to as a bowtie neLwolk. The nelwolh 180 can be trained via a back propagation ll~i~f~llgte~hni~ e. This is described in D.E. l~l-m~lh~rt, G.E. Hinton and R.J. Williams,"Learning Tnt~rn~l Reples~ ;t)ns by Propagations" in D.E. ~l~m~lh~rt and J.L.
McClelland, Parallel Disfribufive Processing, Vol. 1, 1986.
Referring now to FIGUREs 9a and 9b, there are illustrated two plots depicting operation of the sensor validation system 22. The actual inputs are represented by XA
and the predicted input is represented by Xp. It can be seen that the predicted input does not exactly follow the actual input, it being noted that the actual input is actually the input to the overall system. The difference between the actual and the predicted input values is illustrated in FIGURE 9b.
S~B~ITU~ ~HE~ UL~ 26~
~ ===
wo 95/04878 2 1 6 7 9 2 7 PCT/US94/08657 .
Referring now to FIGI~REs 10a and lOb, there is illustrated corresponding plots to those of FIGUREs 9a and 9b with the exception that the sensor genel~Li.lg the actual input fails. It can be seen that up to a point 204 on the curve X~, the predicted and actual sensor values track fairly well with minim~l error. However, at the point 204 the error 5 increases dramatically, indicating that the sensor no longer provides an value that corresponds to the predicted value. This is illustrated in FIGURE 10b, wherein the error increases. When the difference between XA and Xp is greater than a threshold, this inrlic~tes an invalid reading. However, as noted above, only the one of the sensors having the highest error above the threshold will be selected as repl~cement value by the multiplexer 86 for the next cycle. This is due to the fact that the network 180 is trained on all of the input variables and each of the input variables will affect the predicted values for the, ~ g ones. Therefore, if the actual input values associated with predicted output values having an error greater than the threshold were replaced, this would not be as accurate as iteratively reFI~cing one at a time.
R~;~.lil~ now to FIGURE 11, there is illustrated a flow~ a, L depicting the overall operation of the system. The fl~w~,hal L is initi~te~l at a start block 208 and then flows to a decision block 210. Decision block 210 determines whether the remote CEM
has been installed. If so, the program then flows to a function block 212 to measure the NOx levels with the remote CEM. The program then fiows to a decision block 214 to 20 determine whether the measured NOx values, measured in function block 212, are acceptable. If not, this in~ tes that the virtual sensor network 18 is out of spec and that the system has either chal1ged or the network no longer I ~presellLs the system. The program will then flow along an "N" path to a function block 216 to measure the system variables and then to a function block 218 to generate a training ~t~b~.~e. A training 25 d~t~baee ess~nti~lly utilizes the system variables that are measured along the same time base as the measured NOx levels. Typically, the remote CEM will be placed ~ c~nt to the m~ml~chlring facility and the pollllt~nt~ measured for a predetermined amount of time, which can be measured in hours, days or weeks. At the same time, the plant facility itself is measuring the plant variables. These are also placed on a time base and stored.
30 By n~ ging the two data sets, a training (J~t~b~e can be provided for training the virtual sensor network 18. This time merging operation is described in U.S. Patent Application S~IBSTITUTE SHEET (P~ULE 2~) wo 95/04878 2 1 6 7 9 2 7 PCT/US94/08657 Serial No. 980,664, filed,~ovember 24, 1993 and entitled "Method and Apparatus for Opel~til1g a Neural Network with Missing and/or Incomplete Data" (Atty. Dkt. No.PAVI-20,965).
Once the training database has been generated, the virtual sensor nc;twolk 18 is5 trained, as inflir~ted by a function block 220. This es~Pnti~lly generates weights, which can then be substituted for the neural network weights in the virtual sensor network 18.
The program then flows to a function block 222 to substitute new weights in the virtual sensor network 18. Thereafter, the program flows to a main ope~Lhlg portion ofthe program, which is initi~ted at a function block 224 to validate the sensors.
If the pollutant parameters measured in the function block 212 were acceptable, the program would flow from the decision block 218 along a "Y" path to the input of function block 224 to bypass the training step. Additionally, if the remote CEM is not present, the program would flow along an "N" path from the decision block 210 to the input of the sensor validation block 224.
The sensor validation block 224 validates the sensors and, if one is found invalid, it substitutes a predicted value for that invalid sensor. The program would then flow to a function block 226 to determine if certain sensors needed to be replaced by predicted values. If so, the program would flow along a "~' path to replace the invalid sensors with the predicted sensor value. The program would then flow to a function block 232 20 to predict the pollutant value and then to a function block 232 to control the plant. The program would then flow back to a decision block 210. If it were determined thatsensors did not need to be replaced by their predicted values, the program would flow along an "~' path from the decision block 226 to the input of function block 230.
Referring now to FIGURE 12, there is illustrated a function block depicting the 25 operation of the sensor validation. The program is initi~ted at a start block 240 and then flows to a function block 242 to input the various sensor re~ing~ The program then flows to a function block 244 to run the sensor validation model and then to a decision block 246 to compare the predicted input values with the thresholds and generate an S~B~lTl~rE ~IEE~ ULE ~6-~
wo gs/04878 2 1 6 7 9 2 7 PCTIUS94/08657 error signal when any of the predicted input values exceed the thresholds for that given variable, it being noted that there can be a threshold for each variable as to what con~titutes an error for that sensor value. When an error exists, the program flows to a function block 248 to replace the largest input error with the mean value for that input.
5 An alarm is generated at this point to warn of the failed sensor. The program will then flow back the input of a function block 244.
When the system has iteratively detel ll"ned that there are no longer any predictive outputs that exceed these thresholds, the program will flow from a decision block 246 to a function block 250 to replace all detected errors with predicted sensor lO values and then to a function block 252 to output reconciled sensor values. The program will then flow to a return block 254.
Referring now to FIGURE l3, there is illustrated an embodiment v~ heleill the virtual sensor is utilized in conjunction with an internal combustion engine 260. The int~rn~l comh--stion engine 260 receives air on an intake port 262. This is input to a l 5 butterfly valve 264 which is basically the throttle valve that is controlled by the foot pedal on an automobile. The butterfly valve then feeds the restricted airflow to an intake manifold 266, which is input to the internal combustion engine 260. A plurality of fuel injection valves, only one of which is illustrated, fuel injection valve 268. The filel injection valve 268 is operable to inject fuel into the intake manifold 266 in a re~ ted 20 amount, which is determined by a number of factors, this being conventional. The internal combustion engine 260 exhausts the combustion ingredients into an exhaust m~nifold 270. The exhaust manifold 270 is conn~cted to an exhaust pipe 272 which is connected to the input of a catalytic converter 274. The catalytic converter 274 then interfaces with the tail pipe 276 to output the combustion gases.
The exhaust manifold 270 is connected through a pipe 278 to an emissions gas recirc.... ....l~tion valve (EGR) 280. The EGR 280 is interfaced with the intake manifold 266 through a pipe 282. The EGR 280 is a conventional pollution control device that is operable to bleed offa small portion of the exhaust gases from the exhaust manifold 270 back into the intake manifold for recombustion thereof. EGR valves typically operate at S~85TITU~ E~T (~ 63 WO 95/04878 2 1 6 7 9 2 7 PCTrUS94/08657 higher RPMs of the engine, as the recirculation of the exhaust gases at low RPMs causes the engine to idle roughly.
In typical internal combustion ~.~in~.~, an oxygen sensor 286 is disposed in the exhaust manifold 270. The oxygen sensor 270 basically provides for measurement of the 5 exhaust gas ingredient concentration, typically oxygen, that exists in the exhaust manifold 270. These type of sensors are utilized for air-fuel ratio control systems. The sensor is input to a central processing unit (CPIJ) 288 which controls the operation of the fuel injectors through a line 290 and the fuel supply thereto. Again, these are collvelllional systems.
10In addition to the oxygen sensor 286, the CPU 288 is operable to monitor a large number of p~llelers lègal~ling the internal combustion engine, which parameters require some type of sensors. In the example illustrated in FIGURE 13, the operation of the ignition, i.e., spark advance, timing, etc., is provided to an ignition module 292. The manifold temperature is provided by a manifold temperature sensor (TM) 294 and the 15 cylinder telllpel~ e is provided by a cylinder telllpe~Lule sensor (TCYL) 296. The sensors 294 and 296 allow a telllpel~ re measurement to be made of each cylinder and the overall m~nifold The back pressure in the exhaust pipe 270 is provided by a pressure sensor (PEX) 298 and the pressure in the intake manifold is provided by a pressure sensor (PMAN) 300. Typically, the pressure sensor 300 measures a vacuum for 20 a conventional engine, which could be a positive pressure when the system is operated in conjunctiQn with a turbocharger. The intake manifold temperature is provided by a telllpel~ re sensor (TA) 302, which ls connected to the intake manifold 266. Theposition of the butterfiy valve 264 is provided by a position sensor (~th) 306. The tem~elal~lre of the catalytic collvel Ler is measured by a temperature sensor (TCAT) 308.
25 The CPU 288 incol~ol~Les the full sensor system as described hereinabove and is operable to predict the emissions emitted by the internal combustion engine which, in the plerelled embodiment, are primarily NOX, and either output this information in the form of a display, store it as a history or utilize it to control the operation of the engine.
30When controlling the operation ofthe engine, an emissions control system 308 is provided, which is operable to control certain parameters of the system. For example, S~ST~rUTE ~ ET ~ULE ~
WO 95/04878 2 ~ 6 7 9 2 7 PCT/US94/08657 .
one parameter that could be controlled is the internal threshold to the CPU 288 that clet~rmines the air-fuel ratio. Typically, the oxygen sensor 286 operates on a threshold such that when it is above the threshold, the air-fuel mixture is ch~rlged in one direction and, when the oxygen sensor falls below the threshold, the air-fuel ratio is ~h~nged in the 5 other direction. By ~h~nginc~ the threshold, the average air-fuel ratio will be changed, and Ihelerole, the actual emissions output can also be çh~n~e~ This will be described in more detail hereinbelow.
For systems that require emissions monitoring, there are a number of methods that have been employed in the past. One method that has been accepted by the 10 Environment~l Protection Agency (EPA) is to have a dashboard light that will illllmin~te after a predetermined number of miles, which predetermined number of miles is determined empirically. When the light is ill~lmin~tetl, the vehicle must be taken in for inspectiGn. With the emissions sensor of the present invention, the actual emissions can be predicted from the operation of the engine and the light illllmin~te(l, indicating that the 15 predicted emissions levels have fallen below a predetermined threshold. Thesethresholds can be ~h~nge~l, depending upon the type of vehicle, the area of the country, etc. For example, most commercial vehicles have less ~L~;"genL pollution standards and certain areas like Los Angeles, California have very stringent standards. In this manner, di~rellL thresholds can be loaded into the CPU 288, which thresholds can be sPIected in 20 accordance with predetermined criteria. Additionally, a history can be provided of the vehicle as to the emissions generated by the engine and this stored for downloading at a later time for the purpose of monitoring the operation of the vehicle and the associated engme.
Since the emissions prediction is achieved u~ ing a model of the system which 25 utilizes the sensor outputs as inputs to provide the predicted output, the model must be initially trained. This initial training operation can be done on a "generic" engine with a "standard" emissions monitor and the then the model parameters downloaded to a standard integrated circuit which is utilized for all vehicles. This ~c sllmes that the model trained on the generic system holds true with respect to all subsequent systems and the 30 m~mlf~ct~ring tolerances associated therewith. However, even if the generic model does provide a true representation of all engines m~mlf~chlred for a given type of engine in SIJBSTITUTE SHEET (RULE 26~
combination with a standard çmi~ion~ monitor, the parameters of the engine will change over time. It may thel~role be necess~y to update the training. This can be achieved in two ways. In the first method, the emissions can be measured with an emissions sensor that is external to the vehicle and not an integral part thereof and this compared to the predicted emissions output. If an error exists that is too large, the system then can be retrained. The . tiLl ~nil~g can either be a complete I t;L. ~inillg of the model or merely an update of the training weights. In any event, this requires an actual emissions sensor to be ~ltili7e(l In a second method, the model can have a "bias" applied thereto to provide a slight offset. This also e~ui.es actual emissions to be monitored. The actual values are necess~.y to know how to adjust the bias.
Referring now to FIGURE 14, there is illustrated a block diagram of a system wherein an external emissions sensor 310 is utilized for l-~i. ing. The internalcombustion engine is represented by a block 312, which receives inputs on a line 314 and provides measured outputs or state variables s(t) on lines 316, these comprising the inputs to the model However, these state variables are first processed by sensorvalidation module 318, which was described hereinabove, and which is operable tosubstitute a predicted sensor output in the event of a failure of one of the sensors. This is described hereinabove with .erelence to FIGI~REs 7 and 8. The v~lid~ted statevariables s(t)' are output on lines 320 to a predictive model processor 322. Thepredictive model processor 322 is operable to interface with the memory 324 for storing model parameters to process the pa.~-..e~ers and provide a predicted output e(t) on a line 326 ~lrlition~lly~ the predictive model processor 322 is operable to store a time history ofthe predicted output in a nlellloly 328 The model that is stored is a represent~tion of the combined engine and emissions sensor Therefore, the model will have associated therewith all aspects of both the engine and the emissions sensor. Even if the emissions sensor is inaccurate, the model is only as good as the sensor, but this inaccuracy must be incorporated into the model.
This is important in that regulatory bodies require that the output measurement comply to their standards, which standards are defined by their equipment. If, for example, the emissions sensor that complied with their standards were in fact inaccurate, it would be important to predict an output with these inaccuracies. To correct these inaccuracies UT~ S~ UL~- 2~
wo g~/~78 2 1 6 7 9 2 7 PCT/US94/086S7 .
would not be acceptable. Therefore, the model is determined utili7in~ as part ofthe system the actual hardw~ sensor, which is removed during operation ofthe eng-ine.
r.~tern~l to the internal combustion engine 312, the external emissions sensor 310 is connPcted to the output ofthe internal combustion engine, which comprises a line 330 5 labelled y(t). This represents the output of the system. This is merely the output of the exhaust pipe. The emissions sensor 310 is connected to the output to provide an actual output value ofthe emissions on a line 332. This is input to a difference device 334 to determine the difference between the output of the emissions sensor 310 and the predicted output on line 326. This generates an error E, which is input to a co",pa,~lor 336. The co.l.~Lor 336 COlllp~ht;S the error E with a predetermined threshold and then outputs a "Train" signal on a line 340. If the error exceeds the threshold, a training operation is initi~ted This is input to the predictive model processor 322. The predictive model processor 322 then enters into a Ll~il~ing mode utili7:ing the actual emissio~s sensor outputs on a line 344 and the state variable inputs 320 to retrain the 15 model. These p~..elers are then input to the memory 324. After L~ ~ing~ the system will again be v~ ted by col,lpaling the operation ofthe internal combustion engine and the emissions output thereby with the predicted emissions output. Once the error has been ...;..;...;,e~ i.e., reduced below the threshold, the system will be "v~ ted". The predictive model processor 320 can train the network by two methods. In the first 20 method, it can completely regen~l~Le model parameters from scratch ~ltili~ing a typical training algorithm. In the second method, it can merely update the model pal~lllc;Lel ~, i.e., provide a minor adj-letm~nt thereto in order to reduce the error.
l~erelling now to FIGURE 15, there is illustrated an alternate method whelt;in the l~Ll~lling operation is done external to the system with a leLI~nillg processor 346.
25 The l~ ining processor 346 is operable to receive on the input thereofthe output state variables from the internal combustion engine 312 on lines 348, the actual output - emissions sensor on the line 332 and the error output of the difference device 334. The inillg processor 346 then d~L~Illlilles whether ~ il~ing is necç~s~ry, and if so, the leLl~.fi.lg processor 346 will either update the model parameters or generate a new set of 30 model p~lleLers. During an update process, the old model parameters from the memory 324 are ~lplo~ded and adjusted and then downloaded back to the memo~ 324.~J~UT~ T ~ L~ 2û.~
.
In the complete training process, completely new model parameters are generated and then downloaded to the memory 324. Of course, after I~Ll~h~ g or any modification of the model p&~ lers 324, the system is again checked.
In the training of the network, one technique that can be utilized is 5 back~l-,pagation, as described in D.E. P~llm~lh~rt, G.E. Hinton, and R.J. Williams, "Learning Internal Represent~tiQns by Error Propagation" in D.E. l?llm~lh~rt & J.L.
McClelland, Parallel Distribu~ed Processing, Vol. 1, 1986. In this technique as applied to a neural nGlV~O1k~ training is achieved by ~ g the Least Mean Square Errors with back~ropagation. This utilizes the technique of steepest descçnts, wherein the 10 weights Wu f a neural network and the parameters associated with the activation function are varied to Ill;~ e the error function. This bacl~,opagation te~hniq~1e is ess~nti~lly a common, non-linear least squares algorithrn. It is a natural, non-linear extension of the linear nets commonly used in adaptive signal processing. Use of the chain rule in comruting derivatives ofthe error generated during the training procedure 15 provides useful interpolation to the ...;I~;.";~;on process and allows an easy generalization to multi-layers of non-linear units in a neural network. For one or more output units, the error output is i~,;,.;".;~.etl to:
1 ~ ( z (t) _ yf(t) )2 (6) where: y(t) = output of a neural net; and z(t) = specified target output for a given output pattern.
20 For a network that contains non-linear hidden units, the term y(t) contains contributions from the outputs of hidden units in a hidden layer. Because the hidden layer has a non-linear ~ srel function, the output of the hidden layer will be an output of the non-linear function of its input and the error E becomes a square of non-linear function weights since the hidden layer outputs are fed into the topmost output layer in a conventional 25 three layer neural network. The backpropagation algorithm is described in the literature and also described in U.S. Patent No. 5,113,483 issued May 12, 1992 and entitled SU~S~I~UTE ~ 6~
WO 95/04878 2 1 6 7 9 2 7 PCT/US94/086~7 "Network with Semi-Localized Non-Linear Mapping of the input Space". This patent is incorporated herein by lt;rt;l~nce.
l[n addition to bach~l opagation, other techniques for training a neural networkcan be ~1tili~-l, such as radial basis functions or C~ ssi~n bars. Further, it is not 5 necee~, y to utilize a neural nc;~wol k to provide a stored repres~nt~tion of the system. A
fuzzy system (which is very similar to a radial basis function network) can also be ili~ed Ref~ling now to FIGURE 16, there is illustrated a block diagram depicting the runtime operation of the internal combustion engine 312. The system of FIGURE 1610 utilizes a control network 350 which is operable to receive the control input ~(t) on line 314, the v~ ted sensor outputs s(t)' on lines 320 and output updated control inputs ~(t+1) on a line 352. The control network 350 is described above with refel~nce to FIGURE 5a and can actually incorporate the model that is implçm~nted by the predicted model processor. A desired or target emissions level is input thereto. The line 352 is 15 input to an engine control system 354 which is operable to effect the various controls on the internal combustion engine. Any one of the controls can be manipulated to control the emissions within predet~ lined ~lid~.lines, these controls associated with controlling he air-fuel ratio.
The overall system is controlled by a runtime operating system 354 which is 20 operable to receive the predicted output on the line 326 from the predictive model processor 322 and also receive the output of the sensor validation module 318 intlic~tin~
which, if any, of the sensors have been determined to be in error. This h~rullllalion~ in addition to the predicted emissions value output on line 326 is then utilized by the runtime opel~Lil1g system to either store it in the memory 328 associated with the 25 historical h~----aLion or make various decisions as to what should be done with respect to the predicted emissions information.
Runtime thresholds are prestored in a memory 356 and are utilized by the runtimeoperating system for com~alison with the predicted emissions. If the predicted emissions 30 exceed the selected one of the thresholds, some action must be taken. For example, ~IB~ITUr~ ~HE~T ~LE 26J
WO 95/04878 2 1 6 7 9 2 7 PCT~US94/08657 emissions may be acceptable at one threshold in one area of the country and ~n~cceptable in another area ofthe country. Further, the predicted emissions levels may also have an acceptability that is a functioh of other parameters, such as telllpe,~ re and humidity. The runtime thresholds could be selected as a function of atmospheric 5 con-lition~ or other criteria. However, in the pl~relled embodiment, it is anticipated that thresholds will be selected as a function of the locale that the engine is disposed in.
Further, they could even be s~lected as a function of the time of day.
A user input 358 iS utilized to select the thresholds or input thresholds via an input/output circuit 360. Further, the input/output circuit 360 is operable to interface with a display 360 and also with a comm.lnic~tion system 362. The display 360 can be, for example, a warning light. Further, it could be some type of display that actually outputs an analog value in the form a "gas gauge" for viewing by the driver. This would allow the driver to actually view the emissions levels as a function of his driving 15 conditions, etc. In one mode, the communications system 362 is provided such that industrial engines at remote sites can be controlled on a periodic basis to download the stored history h~lllla~ion in memory 328 to a central station.
Referring now to FIGURE 17, there is illustrated an overall view ofthe 20 communication system. A plurality of engines re~lled to as plants 370 are disposed at remote locations, each having a virtual emissions monitor 372 associated therewith and each having a communications device 374 associated therewith. In the described embodiment, each of the communication devices 374 is operable to ll~llslnil i~ ion over a wireless commllniç~tion path via an ~ntçnn~ 376, which ~ntPnn~ 376 can operate 25 in both a receive mode and a ll~l~nlil mode. The ant~nn~ 376 are operable to co""~ ;c~te with an ~ntçnn~ 378 on a colllllland station 380. The protocol utilized for the tr~n~mi.c~ion can be any type of conventional protocol. ~ltho~lgh a wireless system is illustrated, it should be understood that a fixed wire system could be lltili7e~
In addition to training the system as described above lltili7:ing techniques such as 30 back propagation, G~ n bars, radial basis functions, etc., the network could have the parameters thereof offset or a bias adjust applied thereto. After training a neural nc;lwulk in the normal manner, a situation could occur wherein the average of the plant Sl~S~T~TE SH~E~ ~Rl I~E 26~
wo 95/04878 2 1 6 7 9 2 7 PCT/US94/08657 output drifts by some amount. This situation is illustrated in FIGURE 18, wherein a solid curve is illustrated replesç~ the actual emissions output measured by an external emissions monitor and the dotted line represents the predicted output from the virtual sensor. It can be seen that there is an average error that occurs over time. This average 5 error is determined and utilized to determine an offset, the average error defined as follows:
( ~ ) = ( o" - op ) (7) where: o~ is the actual plant output; and op is the output value predicted by the model.
The averages are deterrnined over all of the examples in some past window of time.
10 Adding this average error to the model output results in the following:
o/ = op + ( e ) (8) such that the new average error of the model is zero, as follows:
( e ~ o~ ~ / ) = ( a ) ~ ( p ) ~ ( a ) + ( p ) = ( 9 ) This bias adju.stm~nt must be made periodically or in an ongoing manner, l~tili7:ing moving averages.
In another parameter adjl-stment system, a first principles model could be 15 utili7ed First principle models are well-known and rely upon the fact that there are only a small number of controls that can be tuned or adjusted on the internal combustion - engine. Typically, this tuning or control takes the form of adjusting parameters, i.e., co-efficients in the model to ...;..;...;~e the error that exists between the model output and the actual plant or engine output. This procedure is completely analogous to Ll~illing a 20 neural network model. The differences between the two training operations occur in the method that the p~ e.~ appear in the model and perhaps the way in which they are SU~S~lr~TE s~r (~l~LE ~6~
_ WO 95/04878 2 ~ 6 7 9 2 7 PCT/US94/08657 adjusted. First principle models are often simpler than neural network models in that they are often linear and have fewer adjl.~tment parameters. An example ofthe first prinr.iple models applied to NOx emissions from an internal combustion engine is as follows:
NOxppm = k ( Tc Tm ) ( 10 ) where: Tc is the cylinder te..lpe~ re;
Tm is the manifold telll~ L-Ire and the average of the difference therebetween is taken over all cylinders; and k is the adjustable parameter.
The pal ~llle~er k can be adjusted in the same iterative manner that one adjusts the lO pal~lllelel~ in a neural network model, i.e., by gradient descP.nt; that is, one minimi7es the overall error between the model and the plant as follows:
( a ~ ~p )2 = 2 ~ e2 (11) 2 e~ampl eS e~ampl es via iteratively ~h~nging k according to the gradient descent equation:
/~K = -n ~ (12) In the present case, this is simply:
~K = ne ( To - T,~) (13) In the above equation, a simple control example would be a situation wherein if the NOx l 5 output was high, a signal would be issued that would result in red~lcing the air-fuel ratio by reducin~ the air and increasing the fuel.
SU~S~TU~E SHEEr (~ULE ~6~
wo 95/04878 2 1 6 7 9 ~ 7 PCT/US94/08657 Rt;relling now to FIGURE l9, there is illustrated a flowchart depicting the overall runtime control operation. The program would be initi~ted at a start block 386 and the~ proceeds to a function block 388 to compare the expected emissions value to an internal runtime threshold, it being remembered that these thresholds can be selected by 5 the user for a particular cirCllm~t~n~e The program then flows to a decision block 390 to determine if the expected or predicted emissions exceed the threshold. If so, this inllic~tes that some control operation or failure mech~nicm must be performed. The program would flow to a decision block 392 to determine if a control operation is present. If so, the program would flow along a "Y" path to a function block 394 to l0 adjust the controls in accordance with the predetermined control operation which could utilize the first principles operation as described above, or a control network. The program would then flow to a decision block 396 to determine if it is possible to effect an a~prop~iale control. In order to determine this, maximum limits are set within the operating parameters of the engine by which the controls can be modified to reduce 15 çmi.e.cionc If these control limits are exceede-l, the engine operation will deteriorate, even though the emissions was applopliaLe. If the maximum control limit has beenexceeded, the program would flow along a "Y" path to a function block 398 to set the default settings for the emissions control parameters and then to a function block 400 to store the history. If the m~im-lm control limit had not been exceede(l the controls 20 would be accepted and the program would flow along an "N" path from decision block 396 to the function block 400. If no controls were available, the program would also flow to the function block 400 from the decision block 392 along the "N'' path thereof.
Once the history has been stored as to whether a control operation had been effected or a simple 1 ~;co~ dillg of the emissions history had been made, the program 25 would flow to a decision block 402 to determine if an alarm operation were present.
This could occur in the event that the co~pa-ison made at decision block 390 required an alarm to be set or if the presence of the default setting in the function block 398 required an alarm to be set. If an alarm is to be set, the program would flow along the "Y" path from decision block 402 to a function block 404 to set the alarm, this possibly 30 being a light on the dashboard or an audible alarm. Further, this could be the existence of an alarm communication to a central station requiring a m~int~n~nce check. The program would then flow to a return block 406. If the alarm operation had not been SVSSTITUTE Sl IEET (RULE 26) WO 9s/04878 2 i 6 7 q 2 7 PCT/[TS94/08657 present, the program would flow from the decision block 402 along the "N" path to the return block 406. Similarly, if the predicted emissions did not exceed the threshold, the program would flow to the return block 406 along the "N" path from the decision block 390.
Referring now to FIGURE 20, there is illustrated a flowchart depicting the training operation. The program is initi~ted at a start block 410 and then proceeds to a function block 412 to measure the actual emissions. The program then flows to a function block 414 to comp~e the measured actual emissions to the expected or predicted value output by the virtual sensor. The program then flows to a decision block 416 to determine if the difference between the measured actual emissions and thepredicted value exceed the threshold. If so, the program wouLd flow along a "Y" path to a decision block 418 to dt;t~ e if a training operation is to be performed. If not, the program would flow along an "N" path to a function block 420 to update a record ~at~ha~e and then to a return block 422. Similarly, if the error did not exceed the threshold, the program would flow from the decision block 416 along the '~' path to the function block 420.
If a training operation is to be pel ~I llled, the program would flow from the decision block 418 along a "Y" path to a decision block 424 to determine if a full training operation is to be performed. If yes, the program would flow along a "Y" path to a function block 426 to retrain a complete model and then to a function block 428 to download the model p~ll~Lers back to the memory associated with the model pal~~ . The program will then flow to a return block 430. If a full training operation were not to be p~l~lllled, i.e., a partial training operation only, the program would then flow along an '~' path from decision block 424 to a function block 432 to upload the model pal~llllGLt;l~ that were in the system and then adjust these model parameters as intlic~ted by a function block 434. After adjll~tm~nt, the program would flow to the input of the function block 428 to download the model parameters. It should be understood that the training operation can be an on-board operation utili7.in~ the model processor or an off-board operation lltili7ing an external processor.
su~sTIruTI~ s~ ULE 7B~
WO 9S104878 2 1 6 7 9 2 7 PCT/uS94/086s7 l[n summary, there has been provided a system for predicting emissions in order to obviate the need for an actual emissions sensor on a reciprocating engine. The system utilizes a predictive model that is trained on various sensor outputs from the reciprocating engine with a target output of the actual emissions taken during generation 5 of the training data set. Once trained, the system can operate without the actual emissions sensor, but represents the actual output of the emissions sensor. Periodically, the model can be checked to determine if the predicted value has deviated more than a predeLe~ ed amount from an actual measurement by pelrolll~ing an actual measurement with an external emissions sensor. If the measurement has deviated, the 10 system can be in-~ic~ted as out of spec or the network can be retrained.
Although the plerelled embodiment has been described in detail, it should be understood that various changes, substitutions and alterations can be made therein without departing from the spirit and scope of the invention as defined by the appended claims.
SUBSTITUTE SHEET (RULE 26)
Claims (15)
1. A method for monitoring emissions in an internal combustion engine that emits noxious pollutant and provides a plurality of sensors for measuring selectparameters of the engine operation as sensor output values, comprising the steps of:
storing in a predictive model a representation of the combined engine and a standard external emissions monitor; the standard external emissions monitor operable to measure the noxious pollutants output by the engine, the predictive model having an output that provides a prediction of the output of the emissions monitor as a predicted emissions value and operable to receive as inputs select ones of the sensor output values;
inputting the select ones of the sensor output values to the predictive model; and predicting the output of the combined engine and external emissions monitor to provide an indication of the noxious pollutants output by the emissions monitor without requiring the emissions monitor to be present on the engine.
storing in a predictive model a representation of the combined engine and a standard external emissions monitor; the standard external emissions monitor operable to measure the noxious pollutants output by the engine, the predictive model having an output that provides a prediction of the output of the emissions monitor as a predicted emissions value and operable to receive as inputs select ones of the sensor output values;
inputting the select ones of the sensor output values to the predictive model; and predicting the output of the combined engine and external emissions monitor to provide an indication of the noxious pollutants output by the emissions monitor without requiring the emissions monitor to be present on the engine.
2. The method of Claim 1 and further comprising:
attaching an external emissions monitor substantially similar to the standard external emissions monitor to the engine;
comparing the output of the external emissions monitor to the predicted emissions value of the predictive model; and adjusting the stored representation in the predictive model whenever the predicted emissions value exceeds the actual output of the external emissions monitor by a predetermined amount.
attaching an external emissions monitor substantially similar to the standard external emissions monitor to the engine;
comparing the output of the external emissions monitor to the predicted emissions value of the predictive model; and adjusting the stored representation in the predictive model whenever the predicted emissions value exceeds the actual output of the external emissions monitor by a predetermined amount.
3. The method of Claim 2, wherein the stored representation in the predictive model is generated by training the predictive model on a training set of data comprised of actual sensor data received from the output of the sensors associated with the select ones of the sensor output values, and actual measured emissions output data from the external emissions monitor when connected to the engine.
4. The method of Claim 3, wherein the step of adjusting comprises retraining the predictive model on a new set of training data utilizing the external emissions monitor to provide the actual output data.
5. The method of Claim 3, wherein the step of adjusting comprises offsetting the predicted emissions value of the predictive model by a predetermined amount to minimize the error below a predefined level.
6. The method of Claim 1, and further comprising:
comparing the predicted emissions value with an internal threshold value;
and generating an alarm when the predicted emissions value exceeds the internal threshold value.
comparing the predicted emissions value with an internal threshold value;
and generating an alarm when the predicted emissions value exceeds the internal threshold value.
7. The method of Claim 6, wherein the alarm is a light on a display.
8. The method of Claim 6, and further comprising, storing a plurality of storage threshold values in a memory and selecting only one of the plurality of threshold values for the threshold value to utilize in the step of comparing, the step of selecting performed in accordance with predetermined criteria.
9. The method of Claim 1, and further comprising, storing a history of the predicted emissions values output by the predictive model as a function of time.
10. The method of Claim 8, and further comprising, downloading the stored history on a periodic basis.
11. The method of Claim 1, and further comprising:
determining if the select measured parameters are outside of acceptable limits in accordance with predetermined criteria; and substituting a known value for the sensor output values when it is determined that they are outside of the acceptable limits.
determining if the select measured parameters are outside of acceptable limits in accordance with predetermined criteria; and substituting a known value for the sensor output values when it is determined that they are outside of the acceptable limits.
12. The method of Claim 11, wherein the step of substituting comprises predicting from a past history of the measured sensor output values to be substituted, a predicted value, the predicted value comprising the known values.
13 . The method of Claim 12, wherein the past history of the function of the other measured sensor output values.
14. The method of Claim 11, wherein the step of determining comprises:
providing a sensor validation predictive network having as an input the actual sensor values of the engine, the sensor validation predictive network having associated therewith a stored representation of each of the actual sensor output values as a function of the other of the actual sensor output values to provide on the output thereof a predicted sensor output value for each of the actual sensor values input thereto;
predicting with the sensor validation predictive network the predicted sensor output value; and comparing each of the differences of the input actual sensor output values in the predictive sensor output values with predetermined limits for those differences.
providing a sensor validation predictive network having as an input the actual sensor values of the engine, the sensor validation predictive network having associated therewith a stored representation of each of the actual sensor output values as a function of the other of the actual sensor output values to provide on the output thereof a predicted sensor output value for each of the actual sensor values input thereto;
predicting with the sensor validation predictive network the predicted sensor output value; and comparing each of the differences of the input actual sensor output values in the predictive sensor output values with predetermined limits for those differences.
15. The method of Claim 1, and further comprising:
comparing the predicted emissions value output by the predictive model with a desired emissions value;
providing an emissions control system for modifying the operation of the engine; and controlling the emissions control system to operate in such a manner as to reduce any error between the predicted emissions value and the desired emissions value.
comparing the predicted emissions value output by the predictive model with a desired emissions value;
providing an emissions control system for modifying the operation of the engine; and controlling the emissions control system to operate in such a manner as to reduce any error between the predicted emissions value and the desired emissions value.
Applications Claiming Priority (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US08/102,405 US5386373A (en) | 1993-08-05 | 1993-08-05 | Virtual continuous emission monitoring system with sensor validation |
US08/102,405 | 1993-08-05 | ||
US08/149,216 | 1993-11-05 | ||
US08/149,216 US5539638A (en) | 1993-08-05 | 1993-11-05 | Virtual emissions monitor for automobile |
Publications (1)
Publication Number | Publication Date |
---|---|
CA2167927A1 true CA2167927A1 (en) | 1995-02-16 |
Family
ID=26799346
Family Applications (2)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CA002167588A Abandoned CA2167588A1 (en) | 1993-08-05 | 1994-07-27 | Virtual continuous emission monitoring system with sensor validation |
CA002167927A Abandoned CA2167927A1 (en) | 1993-08-05 | 1994-07-27 | Virtual emissions monitor for automobile |
Family Applications Before (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CA002167588A Abandoned CA2167588A1 (en) | 1993-08-05 | 1994-07-27 | Virtual continuous emission monitoring system with sensor validation |
Country Status (8)
Country | Link |
---|---|
US (2) | US5539638A (en) |
EP (2) | EP0712463B1 (en) |
JP (2) | JPH09504346A (en) |
KR (2) | KR960704258A (en) |
AU (2) | AU688353B2 (en) |
CA (2) | CA2167588A1 (en) |
DE (2) | DE69423895T2 (en) |
WO (2) | WO1995004878A1 (en) |
Families Citing this family (281)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5539638A (en) * | 1993-08-05 | 1996-07-23 | Pavilion Technologies, Inc. | Virtual emissions monitor for automobile |
US7082359B2 (en) * | 1995-06-07 | 2006-07-25 | Automotive Technologies International, Inc. | Vehicular information and monitoring system and methods |
US5703777A (en) * | 1994-10-20 | 1997-12-30 | Anr Pipeline Company | Parametric emissions monitoring system having operating condition deviation feedback |
US5704011A (en) * | 1994-11-01 | 1997-12-30 | The Foxboro Company | Method and apparatus for providing multivariable nonlinear control |
DE4441101B4 (en) * | 1994-11-18 | 2005-01-27 | Robert Bosch Gmbh | Method and device for determining diagnostic threshold values for a specific type of motor vehicle in the field |
DE19508560A1 (en) * | 1995-03-10 | 1996-09-12 | Bosch Gmbh Robert | Prepn. circuit for signals from probe in combustion engine exhaust pipe |
US8036788B2 (en) * | 1995-06-07 | 2011-10-11 | Automotive Technologies International, Inc. | Vehicle diagnostic or prognostic message transmission systems and methods |
US8024084B2 (en) * | 1995-06-07 | 2011-09-20 | Automotive Technologies International, Inc. | Vehicle diagnostic techniques |
US8157047B2 (en) | 1995-06-07 | 2012-04-17 | Automotive Technologies International, Inc. | Occupant protection systems control techniques |
US20080147280A1 (en) * | 1995-06-07 | 2008-06-19 | Automotive Technologies International, Inc. | Method and apparatus for sensing a rollover |
US20080046149A1 (en) * | 1995-06-07 | 2008-02-21 | Automotive Technologies International, Inc. | Vehicle Component Control Methods and Systems Based on Vehicle Stability |
US8229624B2 (en) * | 1995-06-07 | 2012-07-24 | American Vehicular Sciences Llc | Vehicle diagnostic information generating and transmission systems and methods |
US8060282B2 (en) * | 1995-06-07 | 2011-11-15 | Automotive Technologies International, Inc. | Vehicle component control methods and systems based on vehicle stability |
US8019501B2 (en) * | 1995-06-07 | 2011-09-13 | Automotive Technologies International, Inc. | Vehicle diagnostic and prognostic methods and systems |
US20080161989A1 (en) * | 1995-06-07 | 2008-07-03 | Automotive Technologies International, Inc. | Vehicle Diagnostic or Prognostic Message Transmission Systems and Methods |
US5970426A (en) * | 1995-09-22 | 1999-10-19 | Rosemount Analytical Inc. | Emission monitoring system |
DE19536571C2 (en) * | 1995-09-29 | 1998-09-03 | Siemens Ag | Method and device for metering the input of a reducing agent into the exhaust gas or exhaust air stream of an incineration plant |
DE19607101A1 (en) * | 1996-02-24 | 1997-08-28 | Hella Kg Hueck & Co | Electronic device and device for data transmission between two identical electronic devices |
US6017143A (en) | 1996-03-28 | 2000-01-25 | Rosemount Inc. | Device in a process system for detecting events |
US7949495B2 (en) | 1996-03-28 | 2011-05-24 | Rosemount, Inc. | Process variable transmitter with diagnostics |
US6539267B1 (en) | 1996-03-28 | 2003-03-25 | Rosemount Inc. | Device in a process system for determining statistical parameter |
US8290721B2 (en) | 1996-03-28 | 2012-10-16 | Rosemount Inc. | Flow measurement diagnostics |
US6654697B1 (en) | 1996-03-28 | 2003-11-25 | Rosemount Inc. | Flow measurement with diagnostics |
US5750886A (en) * | 1996-06-27 | 1998-05-12 | General Motors Corporation | Engine emissions analyzer with diagnostic |
EP0825506B1 (en) | 1996-08-20 | 2013-03-06 | Invensys Systems, Inc. | Methods and apparatus for remote process control |
US6519546B1 (en) | 1996-11-07 | 2003-02-11 | Rosemount Inc. | Auto correcting temperature transmitter with resistance based sensor |
US6601005B1 (en) | 1996-11-07 | 2003-07-29 | Rosemount Inc. | Process device diagnostics using process variable sensor signal |
US6754601B1 (en) | 1996-11-07 | 2004-06-22 | Rosemount Inc. | Diagnostics for resistive elements of process devices |
DE69714606T9 (en) * | 1996-12-31 | 2004-09-09 | Rosemount Inc., Eden Prairie | DEVICE FOR CHECKING A CONTROL SIGNAL COMING FROM A PLANT IN A PROCESS CONTROL |
US6236908B1 (en) | 1997-05-07 | 2001-05-22 | Ford Global Technologies, Inc. | Virtual vehicle sensors based on neural networks trained using data generated by simulation models |
DE69729222T2 (en) * | 1997-07-12 | 2005-05-04 | Idt Technology Ltd., Hunghom Kowloon | DEVICE FOR MONITORING THE AIR QUALITY |
US5941918A (en) * | 1997-07-30 | 1999-08-24 | Engelhard Corporation | Automotive on-board monitoring system for catalytic converter evaluation |
JPH1185719A (en) * | 1997-09-03 | 1999-03-30 | Matsushita Electric Ind Co Ltd | Parameter estimation device |
JP3331159B2 (en) * | 1997-09-16 | 2002-10-07 | 本田技研工業株式会社 | Plant control equipment |
JP3354088B2 (en) * | 1997-09-16 | 2002-12-09 | 本田技研工業株式会社 | Air-fuel ratio control system for exhaust system of internal combustion engine |
JP3592519B2 (en) * | 1997-09-16 | 2004-11-24 | 本田技研工業株式会社 | Air-fuel ratio control device for exhaust system of internal combustion engine and control device for plant |
US6405122B1 (en) * | 1997-10-14 | 2002-06-11 | Yamaha Hatsudoki Kabushiki Kaisha | Method and apparatus for estimating data for engine control |
AU1282199A (en) | 1997-11-14 | 1999-06-07 | Engelhard Corporation | Vehicular on-board monitoring system sensing combined hc/nox emissions for catalytic converter evaluation |
US6071476A (en) * | 1997-11-14 | 2000-06-06 | Motorola, Inc. | Exhaust gas sensor |
JP3805098B2 (en) * | 1998-03-26 | 2006-08-02 | 株式会社日立製作所 | Engine exhaust gas purification control device |
JP3484074B2 (en) * | 1998-05-13 | 2004-01-06 | 本田技研工業株式会社 | Plant control equipment |
DE19843960A1 (en) * | 1998-09-24 | 2000-03-30 | Siemens Ag | Method and device for the catalytic reduction of the pollutant content in the exhaust gas of an incineration plant |
US6216083B1 (en) | 1998-10-22 | 2001-04-10 | Yamaha Motor Co., Ltd. | System for intelligent control of an engine based on soft computing |
US6611775B1 (en) | 1998-12-10 | 2003-08-26 | Rosemount Inc. | Electrode leakage diagnostics in a magnetic flow meter |
US6615149B1 (en) | 1998-12-10 | 2003-09-02 | Rosemount Inc. | Spectral diagnostics in a magnetic flow meter |
JP3484088B2 (en) * | 1998-12-17 | 2004-01-06 | 本田技研工業株式会社 | Plant control equipment |
US6240343B1 (en) * | 1998-12-28 | 2001-05-29 | Caterpillar Inc. | Apparatus and method for diagnosing an engine using computer based models in combination with a neural network |
US6298454B1 (en) | 1999-02-22 | 2001-10-02 | Fisher-Rosemount Systems, Inc. | Diagnostics in a process control system |
US6151547A (en) * | 1999-02-24 | 2000-11-21 | Engelhard Corporation | Air/fuel ratio manipulation code for optimizing dynamic emissions |
US6353804B1 (en) * | 1999-04-09 | 2002-03-05 | General Electric Company | Method for statistically predicting equipment performance |
JP4265704B2 (en) * | 1999-04-14 | 2009-05-20 | 本田技研工業株式会社 | Air-fuel ratio control apparatus for internal combustion engine and control apparatus for plant |
US6336084B1 (en) | 1999-05-11 | 2002-01-01 | Ford Global Technologies, Inc. | Three-way catalyst model for an engine air-to-fuel ratio control system |
WO2000070531A2 (en) | 1999-05-17 | 2000-11-23 | The Foxboro Company | Methods and apparatus for control configuration |
US7089530B1 (en) | 1999-05-17 | 2006-08-08 | Invensys Systems, Inc. | Process control configuration system with connection validation and configuration |
US6788980B1 (en) | 1999-06-11 | 2004-09-07 | Invensys Systems, Inc. | Methods and apparatus for control using control devices that provide a virtual machine environment and that communicate via an IP network |
US6505517B1 (en) | 1999-07-23 | 2003-01-14 | Rosemount Inc. | High accuracy signal processing for magnetic flowmeter |
US6701274B1 (en) * | 1999-08-27 | 2004-03-02 | Rosemount Inc. | Prediction of error magnitude in a pressure transmitter |
US6519552B1 (en) * | 1999-09-15 | 2003-02-11 | Xerox Corporation | Systems and methods for a hybrid diagnostic approach of real time diagnosis of electronic systems |
SE515570C2 (en) | 1999-10-05 | 2001-09-03 | Abb Ab | A computer-based process and system for regulating an industrial process |
JP3688533B2 (en) * | 1999-11-12 | 2005-08-31 | 本田技研工業株式会社 | Degradation state evaluation method of exhaust gas purification catalyst device |
US6611735B1 (en) | 1999-11-17 | 2003-08-26 | Ethyl Corporation | Method of predicting and optimizing production |
US7136860B2 (en) * | 2000-02-14 | 2006-11-14 | Overture Services, Inc. | System and method to determine the validity of an interaction on a network |
DE10010681A1 (en) * | 2000-03-04 | 2001-09-06 | Heinz J Theuerkauf | Simulating signal from electronic sensor in motor vehicle using virtual sensor in vehicle control device, based on neural network model |
US6298718B1 (en) | 2000-03-08 | 2001-10-09 | Cummins Engine Company, Inc. | Turbocharger compressor diagnostic system |
US7739096B2 (en) | 2000-03-09 | 2010-06-15 | Smartsignal Corporation | System for extraction of representative data for training of adaptive process monitoring equipment |
WO2001067262A1 (en) | 2000-03-09 | 2001-09-13 | Smartsignal Corporation | Generalized lensing angular similarity operator |
US6957172B2 (en) | 2000-03-09 | 2005-10-18 | Smartsignal Corporation | Complex signal decomposition and modeling |
US6952662B2 (en) * | 2000-03-30 | 2005-10-04 | Smartsignal Corporation | Signal differentiation system using improved non-linear operator |
US6760716B1 (en) * | 2000-06-08 | 2004-07-06 | Fisher-Rosemount Systems, Inc. | Adaptive predictive model in a process control system |
US6898554B2 (en) * | 2000-06-12 | 2005-05-24 | Scientific Monitoring, Inc. | Fault detection in a physical system |
US6721609B1 (en) | 2000-06-14 | 2004-04-13 | Fisher-Rosemount Systems, Inc. | Integrated optimal model predictive control in a process control system |
DE60126708T2 (en) * | 2000-06-30 | 2007-10-25 | Siemens Corporate Research, Inc. | Method and system for industrial operator guidance by means of predictive alarm detection |
US7092803B2 (en) * | 2000-08-18 | 2006-08-15 | Idsc Holdings, Llc | Remote monitoring, configuring, programming and diagnostic system and method for vehicles and vehicle components |
JP4196535B2 (en) * | 2000-11-02 | 2008-12-17 | トヨタ自動車株式会社 | VEHICLE CONTROL DEVICE AND RECORDING MEDIUM |
US6556939B1 (en) * | 2000-11-22 | 2003-04-29 | Smartsignal Corporation | Inferential signal generator for instrumented equipment and processes |
US7233886B2 (en) * | 2001-01-19 | 2007-06-19 | Smartsignal Corporation | Adaptive modeling of changed states in predictive condition monitoring |
US6508241B2 (en) | 2001-01-31 | 2003-01-21 | Cummins, Inc. | Equivalence ratio-based system for controlling transient fueling in an internal combustion engine |
AU2002244045A1 (en) * | 2001-02-19 | 2002-09-04 | Rosemount Analytical Inc. | Improved generator monitoring, control and efficiency |
DE10108181A1 (en) * | 2001-02-21 | 2002-08-29 | Bosch Gmbh Robert | Method and device for correcting a temperature signal |
US6866610B2 (en) * | 2001-03-30 | 2005-03-15 | Toyota Jidosha Kabushiki Kaisha | Control apparatus and method for vehicle having internal combustion engine and continuously variable transmission, and control apparatus and method for internal combustion engine |
US6535124B1 (en) * | 2001-04-03 | 2003-03-18 | Abb Automation Inc. | Method and apparatus for digital analysis and signal conditioning in a turbine generator silo combustor |
US7539597B2 (en) | 2001-04-10 | 2009-05-26 | Smartsignal Corporation | Diagnostic systems and methods for predictive condition monitoring |
US20020183971A1 (en) * | 2001-04-10 | 2002-12-05 | Wegerich Stephan W. | Diagnostic systems and methods for predictive condition monitoring |
US6629059B2 (en) | 2001-05-14 | 2003-09-30 | Fisher-Rosemount Systems, Inc. | Hand held diagnostic and communication device with automatic bus detection |
US6859755B2 (en) | 2001-05-14 | 2005-02-22 | Rosemount Inc. | Diagnostics for industrial process control and measurement systems |
DE10222703B4 (en) * | 2001-05-23 | 2015-06-18 | Denso Corporation | Control unit for an internal combustion engine |
US6975962B2 (en) * | 2001-06-11 | 2005-12-13 | Smartsignal Corporation | Residual signal alert generation for condition monitoring using approximated SPRT distribution |
US7483867B2 (en) * | 2001-06-26 | 2009-01-27 | Intuition Intelligence, Inc. | Processing device with intuitive learning capability |
US7155321B2 (en) | 2001-08-06 | 2006-12-26 | Idsc Holdings Llc | System, method and computer program product for remote vehicle diagnostics, monitoring, configuring and reprogramming |
US6772036B2 (en) | 2001-08-30 | 2004-08-03 | Fisher-Rosemount Systems, Inc. | Control system using process model |
US6609051B2 (en) * | 2001-09-10 | 2003-08-19 | Daimlerchrysler Ag | Method and system for condition monitoring of vehicles |
US6666200B2 (en) * | 2001-12-10 | 2003-12-23 | Ford Global Technologies, Llc | Method for canister purge compensation using internal model control |
JP3686383B2 (en) * | 2002-02-22 | 2005-08-24 | 本田技研工業株式会社 | Plant control device |
US6718251B2 (en) | 2002-01-29 | 2004-04-06 | Cummins, Inc. | System for controlling exhaust emissions produced by an internal combustion engine |
US6901300B2 (en) | 2002-02-07 | 2005-05-31 | Fisher-Rosemount Systems, Inc.. | Adaptation of advanced process control blocks in response to variable process delay |
US20030162523A1 (en) * | 2002-02-27 | 2003-08-28 | Michael Kapolka | Vehicle telemetry system and method |
FR2838185B1 (en) * | 2002-04-05 | 2004-08-06 | Commissariat Energie Atomique | DEVICE FOR CAPTURING ROTATIONAL MOVEMENTS OF A SOLID |
US6697729B2 (en) | 2002-04-08 | 2004-02-24 | Cummins, Inc. | System for estimating NOx content of exhaust gas produced by an internal combustion engine |
US6882929B2 (en) * | 2002-05-15 | 2005-04-19 | Caterpillar Inc | NOx emission-control system using a virtual sensor |
US7035834B2 (en) * | 2002-05-15 | 2006-04-25 | Caterpillar Inc. | Engine control system using a cascaded neural network |
JP3951967B2 (en) * | 2002-08-01 | 2007-08-01 | トヨタ自動車株式会社 | Automatic adapting device |
EP1416143A1 (en) * | 2002-10-29 | 2004-05-06 | STMicroelectronics S.r.l. | Virtual sensor for the exhaust emissions of an endothermic motor and corresponding injection control system |
JP4104425B2 (en) * | 2002-10-30 | 2008-06-18 | 本田技研工業株式会社 | Method and apparatus for predicting intake pipe pressure of internal combustion engine |
US6925376B2 (en) * | 2003-04-01 | 2005-08-02 | Cummins, Inc. | System for diagnosing operation of a cooling system for an internal combustion engine |
US7181334B2 (en) * | 2003-05-14 | 2007-02-20 | General Motors Corporation | Method and apparatus to diagnose intake airflow |
US7242989B2 (en) * | 2003-05-30 | 2007-07-10 | Fisher-Rosemount Systems, Inc. | Apparatus and method for batch property estimation |
WO2005013019A2 (en) * | 2003-07-25 | 2005-02-10 | Yamaha Motor Co., Ltd | Soft computing optimizer of intelligent control system structures |
US8768573B2 (en) | 2003-08-11 | 2014-07-01 | American Vehicular Sciences, LLC | Technique for ensuring safe travel of a vehicle or safety of an occupant therein |
US20050089811A1 (en) * | 2003-10-24 | 2005-04-28 | United Dominion Industries, Inc. | Exhaust recirculating method and apparatus for a hydrocarbon fired burner |
US7246002B2 (en) * | 2003-11-20 | 2007-07-17 | General Electric Company | Method for controlling fuel splits to gas turbine combustor |
US7761923B2 (en) | 2004-03-01 | 2010-07-20 | Invensys Systems, Inc. | Process control methods and apparatus for intrusion detection, protection and network hardening |
US7251638B2 (en) | 2004-03-03 | 2007-07-31 | Yamaha Hatsudoki Kabushiki Kaisha | Intelligent robust control system for motorcycle using soft computing optimizer |
DE102004011236A1 (en) * | 2004-03-04 | 2005-09-29 | Bayerische Motoren Werke Ag | Process control system |
GB0410135D0 (en) * | 2004-05-06 | 2004-06-09 | Ricardo Uk Ltd | Cylinder pressure sensor |
US7136779B2 (en) * | 2004-05-28 | 2006-11-14 | Daimlerchrysler Ag | Method for simplified real-time diagnoses using adaptive modeling |
DE102004030782A1 (en) * | 2004-06-25 | 2006-01-19 | Fev Motorentechnik Gmbh | Vehicle control unit with a neural network |
US7536232B2 (en) * | 2004-08-27 | 2009-05-19 | Alstom Technology Ltd | Model predictive control of air pollution control processes |
US7784272B2 (en) * | 2004-08-31 | 2010-08-31 | Cummins Inc. | Control system for an engine aftertreatment system |
US7716014B2 (en) * | 2004-09-30 | 2010-05-11 | Rockwell Automation Technologies, Inc. | Reuse of manufacturing process design models as part of a diagnostic system |
US7213553B2 (en) * | 2004-11-12 | 2007-05-08 | Detroit Diesel Corporation | Internal EGR for an internal combustion engine |
DE112005002682B4 (en) * | 2004-11-25 | 2018-05-30 | Avl List Gmbh | Method for determining the particle emissions in the exhaust gas stream of an internal combustion engine |
AT413887B (en) * | 2004-11-25 | 2006-07-15 | Avl List Gmbh | Method for confirming deposition of particulate load in particulate filter in exhaust pipe of internal combustion engine, involves considering correction factor when ideal particulate emissions are at discharge mode |
JP4379336B2 (en) * | 2005-01-07 | 2009-12-09 | トヨタ自動車株式会社 | Control system evaluation apparatus, verification apparatus used in the evaluation apparatus, control system evaluation method, and computer program used therefor |
EP1693558A1 (en) * | 2005-02-16 | 2006-08-23 | ABB Technology AG | Method of forecasting of pollutant emissions of combustion processes |
US8768664B2 (en) * | 2005-03-18 | 2014-07-01 | CMC Solutions, LLC. | Predictive emissions monitoring using a statistical hybrid model |
US7421348B2 (en) * | 2005-03-18 | 2008-09-02 | Swanson Brian G | Predictive emissions monitoring method |
US20060224547A1 (en) * | 2005-03-24 | 2006-10-05 | Ulyanov Sergey V | Efficient simulation system of quantum algorithm gates on classical computer based on fast algorithm |
US20060218108A1 (en) * | 2005-03-24 | 2006-09-28 | Sergey Panfilov | System for soft computing simulation |
US8209156B2 (en) | 2005-04-08 | 2012-06-26 | Caterpillar Inc. | Asymmetric random scatter process for probabilistic modeling system for product design |
US8364610B2 (en) | 2005-04-08 | 2013-01-29 | Caterpillar Inc. | Process modeling and optimization method and system |
US20060229854A1 (en) * | 2005-04-08 | 2006-10-12 | Caterpillar Inc. | Computer system architecture for probabilistic modeling |
US7499777B2 (en) * | 2005-04-08 | 2009-03-03 | Caterpillar Inc. | Diagnostic and prognostic method and system |
US20060229852A1 (en) * | 2005-04-08 | 2006-10-12 | Caterpillar Inc. | Zeta statistic process method and system |
US20060229753A1 (en) * | 2005-04-08 | 2006-10-12 | Caterpillar Inc. | Probabilistic modeling system for product design |
US20060230097A1 (en) * | 2005-04-08 | 2006-10-12 | Caterpillar Inc. | Process model monitoring method and system |
US7565333B2 (en) * | 2005-04-08 | 2009-07-21 | Caterpillar Inc. | Control system and method |
US7877239B2 (en) * | 2005-04-08 | 2011-01-25 | Caterpillar Inc | Symmetric random scatter process for probabilistic modeling system for product design |
US8215098B2 (en) * | 2005-05-02 | 2012-07-10 | Cummins Inc. | Method and apparatus for diagnosing exhaust gas aftertreatment component degradation |
US8112565B2 (en) | 2005-06-08 | 2012-02-07 | Fisher-Rosemount Systems, Inc. | Multi-protocol field device interface with automatic bus detection |
US7818131B2 (en) * | 2005-06-17 | 2010-10-19 | Venture Gain, L.L.C. | Non-parametric modeling apparatus and method for classification, especially of activity state |
FR2887632B1 (en) * | 2005-06-22 | 2007-10-05 | Valeo Systemes Thermiques | DEVICE AND METHOD FOR MONITORING AND CONTROLLING AIR QUALITY, FOR MOTOR VEHICLE |
US20060293817A1 (en) * | 2005-06-23 | 2006-12-28 | Takahide Hagiwara | Intelligent electronically-controlled suspension system based on soft computing optimizer |
US20070061144A1 (en) * | 2005-08-30 | 2007-03-15 | Caterpillar Inc. | Batch statistics process model method and system |
US7231291B2 (en) * | 2005-09-15 | 2007-06-12 | Cummins, Inc. | Apparatus, system, and method for providing combined sensor and estimated feedback |
US20070068225A1 (en) | 2005-09-29 | 2007-03-29 | Brown Gregory C | Leak detector for process valve |
US7451004B2 (en) * | 2005-09-30 | 2008-11-11 | Fisher-Rosemount Systems, Inc. | On-line adaptive model predictive control in a process control system |
US7487134B2 (en) * | 2005-10-25 | 2009-02-03 | Caterpillar Inc. | Medical risk stratifying method and system |
US20070118487A1 (en) * | 2005-11-18 | 2007-05-24 | Caterpillar Inc. | Product cost modeling method and system |
US7499842B2 (en) * | 2005-11-18 | 2009-03-03 | Caterpillar Inc. | Process model based virtual sensor and method |
CN101365378A (en) * | 2005-11-29 | 2009-02-11 | 风险获利有限公司 | Residual-based monitoring of human health |
US7415389B2 (en) | 2005-12-29 | 2008-08-19 | Honeywell International Inc. | Calibration of engine control systems |
US7505949B2 (en) * | 2006-01-31 | 2009-03-17 | Caterpillar Inc. | Process model error correction method and system |
US20070203810A1 (en) * | 2006-02-13 | 2007-08-30 | Caterpillar Inc. | Supply chain modeling method and system |
US20170046458A1 (en) | 2006-02-14 | 2017-02-16 | Power Analytics Corporation | Systems and methods for real-time dc microgrid power analytics for mission-critical power systems |
US9092593B2 (en) | 2007-09-25 | 2015-07-28 | Power Analytics Corporation | Systems and methods for intuitive modeling of complex networks in a digital environment |
US20160246905A1 (en) | 2006-02-14 | 2016-08-25 | Power Analytics Corporation | Method For Predicting Arc Flash Energy And PPE Category Within A Real-Time Monitoring System |
US9557723B2 (en) | 2006-07-19 | 2017-01-31 | Power Analytics Corporation | Real-time predictive systems for intelligent energy monitoring and management of electrical power networks |
WO2007095585A2 (en) * | 2006-02-14 | 2007-08-23 | Edsa Micro Corporation | Systems and methods for real-time system monitoring and predictive analysis |
WO2007109081A2 (en) * | 2006-03-16 | 2007-09-27 | Applied Materials, Inc. | Method and apparatus for improved operation of an abatement system |
US7860857B2 (en) | 2006-03-30 | 2010-12-28 | Invensys Systems, Inc. | Digital data processing apparatus and methods for improving plant performance |
US7693608B2 (en) * | 2006-04-12 | 2010-04-06 | Edsa Micro Corporation | Systems and methods for alarm filtering and management within a real-time data acquisition and monitoring environment |
US7765795B2 (en) * | 2006-04-28 | 2010-08-03 | Caterpillar Inc | NOx control using a neural network |
US8515614B2 (en) * | 2006-05-08 | 2013-08-20 | Lee Bernard | Emission monitoring device and method |
US8275577B2 (en) | 2006-09-19 | 2012-09-25 | Smartsignal Corporation | Kernel-based method for detecting boiler tube leaks |
US7953501B2 (en) | 2006-09-25 | 2011-05-31 | Fisher-Rosemount Systems, Inc. | Industrial process control loop monitor |
US8788070B2 (en) | 2006-09-26 | 2014-07-22 | Rosemount Inc. | Automatic field device service adviser |
JP2010505121A (en) | 2006-09-29 | 2010-02-18 | ローズマウント インコーポレイテッド | Magnetic flow meter with verification |
US8478506B2 (en) * | 2006-09-29 | 2013-07-02 | Caterpillar Inc. | Virtual sensor based engine control system and method |
US7844352B2 (en) * | 2006-10-20 | 2010-11-30 | Lehigh University | Iterative matrix processor based implementation of real-time model predictive control |
US8180622B2 (en) * | 2006-10-24 | 2012-05-15 | Power Analytics Corporation | Systems and methods for a real-time synchronized electrical power system simulator for “what-if” analysis and prediction over electrical power networks |
US20080098725A1 (en) * | 2006-10-31 | 2008-05-01 | Caterpillar Inc. | Exhaust system having mid-reducer NOx sensor |
US8311774B2 (en) | 2006-12-15 | 2012-11-13 | Smartsignal Corporation | Robust distance measures for on-line monitoring |
US20080154811A1 (en) * | 2006-12-21 | 2008-06-26 | Caterpillar Inc. | Method and system for verifying virtual sensors |
US7483774B2 (en) * | 2006-12-21 | 2009-01-27 | Caterpillar Inc. | Method and system for intelligent maintenance |
US8285513B2 (en) | 2007-02-27 | 2012-10-09 | Exxonmobil Research And Engineering Company | Method and system of using inferential measurements for abnormal event detection in continuous industrial processes |
US7810476B2 (en) * | 2007-03-06 | 2010-10-12 | Gm Global Technology Operations, Inc. | Method and apparatus for estimating exhaust temperature of an internal combustion engine |
US20080264398A1 (en) * | 2007-04-26 | 2008-10-30 | Ford Global Technologies, Llc | Auxiliary Power Compensation During Map Testing |
US7787969B2 (en) * | 2007-06-15 | 2010-08-31 | Caterpillar Inc | Virtual sensor system and method |
US7831416B2 (en) | 2007-07-17 | 2010-11-09 | Caterpillar Inc | Probabilistic modeling system for product design |
US7788070B2 (en) * | 2007-07-30 | 2010-08-31 | Caterpillar Inc. | Product design optimization method and system |
US8898036B2 (en) | 2007-08-06 | 2014-11-25 | Rosemount Inc. | Process variable transmitter with acceleration sensor |
US7542879B2 (en) * | 2007-08-31 | 2009-06-02 | Caterpillar Inc. | Virtual sensor based control system and method |
US7593804B2 (en) * | 2007-10-31 | 2009-09-22 | Caterpillar Inc. | Fixed-point virtual sensor control system and method |
US8036764B2 (en) * | 2007-11-02 | 2011-10-11 | Caterpillar Inc. | Virtual sensor network (VSN) system and method |
US8224468B2 (en) * | 2007-11-02 | 2012-07-17 | Caterpillar Inc. | Calibration certificate for virtual sensor network (VSN) |
DE102007057311B3 (en) * | 2007-11-28 | 2009-06-10 | Continental Automotive Gmbh | Method and device for fault detection in emission-relevant control devices in a vehicle |
US7658098B2 (en) * | 2007-12-31 | 2010-02-09 | Min Sun | Method for controlling vehicle emissions |
EP2080953B1 (en) * | 2008-01-15 | 2014-12-17 | STEAG Powitec GmbH | Control loop and method for generating a process model therefor |
US20090222280A1 (en) * | 2008-02-29 | 2009-09-03 | Shiva Mittal | System and method for certification data management |
US8086640B2 (en) * | 2008-05-30 | 2011-12-27 | Caterpillar Inc. | System and method for improving data coverage in modeling systems |
US20090293457A1 (en) * | 2008-05-30 | 2009-12-03 | Grichnik Anthony J | System and method for controlling NOx reactant supply |
CN102124432B (en) | 2008-06-20 | 2014-11-26 | 因文西斯系统公司 | Systems and methods for immersive interaction with actual and/or simulated facilities for process, environmental and industrial control |
US8478473B2 (en) * | 2008-07-28 | 2013-07-02 | General Electric Company | Method and systems for controlling gas turbine engine temperature |
US7917333B2 (en) * | 2008-08-20 | 2011-03-29 | Caterpillar Inc. | Virtual sensor network (VSN) based control system and method |
US8301356B2 (en) * | 2008-10-06 | 2012-10-30 | GM Global Technology Operations LLC | Engine out NOx virtual sensor using cylinder pressure sensor |
JP5149988B2 (en) * | 2009-03-27 | 2013-02-20 | 本田技研工業株式会社 | Plant control equipment |
US7921734B2 (en) | 2009-05-12 | 2011-04-12 | Rosemount Inc. | System to detect poor process ground connections |
US8463964B2 (en) | 2009-05-29 | 2013-06-11 | Invensys Systems, Inc. | Methods and apparatus for control configuration with enhanced change-tracking |
US8127060B2 (en) | 2009-05-29 | 2012-02-28 | Invensys Systems, Inc | Methods and apparatus for control configuration with control objects that are fieldbus protocol-aware |
US8682630B2 (en) * | 2009-06-15 | 2014-03-25 | International Business Machines Corporation | Managing component coupling in an object-centric process implementation |
US20110082597A1 (en) | 2009-10-01 | 2011-04-07 | Edsa Micro Corporation | Microgrid model based automated real time simulation for market based electric power system optimization |
US8244505B2 (en) * | 2009-11-05 | 2012-08-14 | General Electric Company | Predicting NOx emissions |
US8726723B2 (en) | 2010-02-23 | 2014-05-20 | Cummins Emission Solutions | Detection of aftertreatment catalyst degradation |
US8508590B2 (en) * | 2010-03-02 | 2013-08-13 | Crown Equipment Limited | Method and apparatus for simulating a physical environment to facilitate vehicle operation and task completion |
US8538577B2 (en) * | 2010-03-05 | 2013-09-17 | Crown Equipment Limited | Method and apparatus for sensing object load engagement, transportation and disengagement by automated vehicles |
US8650009B2 (en) * | 2010-08-31 | 2014-02-11 | Rockwell Automation Technologies, Inc. | Sensor validation and value replacement for continuous emissions monitoring |
US20120095808A1 (en) * | 2010-10-15 | 2012-04-19 | Invensys Systems Inc. | System and Method for Process Predictive Simulation |
JP5287912B2 (en) * | 2011-03-15 | 2013-09-11 | 株式会社デンソー | Engine control device |
US9207670B2 (en) | 2011-03-21 | 2015-12-08 | Rosemount Inc. | Degrading sensor detection implemented within a transmitter |
GB201105830D0 (en) | 2011-04-06 | 2011-05-18 | Lysanda Ltd | Mass estimation model |
EP3435189B1 (en) | 2011-04-11 | 2022-02-09 | Crown Equipment Corporation | Apparatus for efficient scheduling for multiple automated non-holonomic vehicles using a coordinated path planner |
US8655588B2 (en) | 2011-05-26 | 2014-02-18 | Crown Equipment Limited | Method and apparatus for providing accurate localization for an industrial vehicle |
US8548671B2 (en) * | 2011-06-06 | 2013-10-01 | Crown Equipment Limited | Method and apparatus for automatically calibrating vehicle parameters |
US8589012B2 (en) | 2011-06-14 | 2013-11-19 | Crown Equipment Limited | Method and apparatus for facilitating map data processing for industrial vehicle navigation |
US8594923B2 (en) | 2011-06-14 | 2013-11-26 | Crown Equipment Limited | Method and apparatus for sharing map data associated with automated industrial vehicles |
US8793004B2 (en) | 2011-06-15 | 2014-07-29 | Caterpillar Inc. | Virtual sensor system and method for generating output parameters |
US9256224B2 (en) | 2011-07-19 | 2016-02-09 | GE Intelligent Platforms, Inc | Method of sequential kernel regression modeling for forecasting and prognostics |
US9250625B2 (en) | 2011-07-19 | 2016-02-02 | Ge Intelligent Platforms, Inc. | System of sequential kernel regression modeling for forecasting and prognostics |
US20140058634A1 (en) | 2012-08-24 | 2014-02-27 | Crown Equipment Limited | Method and apparatus for using unique landmarks to locate industrial vehicles at start-up |
US9056754B2 (en) | 2011-09-07 | 2015-06-16 | Crown Equipment Limited | Method and apparatus for using pre-positioned objects to localize an industrial vehicle |
US9677493B2 (en) | 2011-09-19 | 2017-06-13 | Honeywell Spol, S.R.O. | Coordinated engine and emissions control system |
US20130111905A1 (en) | 2011-11-04 | 2013-05-09 | Honeywell Spol. S.R.O. | Integrated optimization and control of an engine and aftertreatment system |
US9650934B2 (en) | 2011-11-04 | 2017-05-16 | Honeywell spol.s.r.o. | Engine and aftertreatment optimization system |
US9181878B2 (en) | 2011-12-19 | 2015-11-10 | Honeywell International Inc. | Operations support systems and methods for calculating and evaluating engine emissions |
US8700546B2 (en) * | 2011-12-20 | 2014-04-15 | Honeywell International Inc. | Model based calibration of inferential sensing |
US8977373B2 (en) * | 2011-12-28 | 2015-03-10 | Caterpillar Inc. | Systems and methods for extending physical sensor range using virtual sensors |
US9052240B2 (en) | 2012-06-29 | 2015-06-09 | Rosemount Inc. | Industrial process temperature transmitter with sensor stress diagnostics |
US9602122B2 (en) | 2012-09-28 | 2017-03-21 | Rosemount Inc. | Process variable measurement noise diagnostic |
US9784198B2 (en) * | 2015-02-12 | 2017-10-10 | GM Global Technology Operations LLC | Model predictive control systems and methods for increasing computational efficiency |
US9605615B2 (en) * | 2015-02-12 | 2017-03-28 | GM Global Technology Operations LLC | Model Predictive control systems and methods for increasing computational efficiency |
US9920697B2 (en) | 2014-03-26 | 2018-03-20 | GM Global Technology Operations LLC | Engine control systems and methods for future torque request increases |
DE102013207720B4 (en) | 2013-04-26 | 2019-10-17 | Fraunhofer-Gesellschaft zur Förderung der angewandten Forschung e.V. | Process for monitoring incinerators |
US10408028B2 (en) | 2013-10-15 | 2019-09-10 | Halliburton Energy Services, Inc. | Optimization of engine emissions from equipment used in well site operations |
KR101542976B1 (en) * | 2013-12-23 | 2015-08-07 | 현대자동차 주식회사 | Method of determining correcting logic for reacting model of selective catalytic reduction catalyst, method of correcting parameters of reacting model of selective catalytic reduction catalyst and exhaust system using the same |
EP3012694A1 (en) * | 2014-10-22 | 2016-04-27 | Siemens Aktiengesellschaft | Method for determining an emission behaviour |
EP3051367B1 (en) | 2015-01-28 | 2020-11-25 | Honeywell spol s.r.o. | An approach and system for handling constraints for measured disturbances with uncertain preview |
EP3056706A1 (en) | 2015-02-16 | 2016-08-17 | Honeywell International Inc. | An approach for aftertreatment system modeling and model identification |
US9593631B2 (en) * | 2015-03-24 | 2017-03-14 | General Electric Company | System and method for locating an engine event |
WO2016175809A1 (en) * | 2015-04-30 | 2016-11-03 | Cummins Inc. | Online engine emission calibration and estimation |
EP3091212A1 (en) | 2015-05-06 | 2016-11-09 | Honeywell International Inc. | An identification approach for internal combustion engine mean value models |
EP3125052B1 (en) | 2015-07-31 | 2020-09-02 | Garrett Transportation I Inc. | Quadratic program solver for mpc using variable ordering |
US10272779B2 (en) | 2015-08-05 | 2019-04-30 | Garrett Transportation I Inc. | System and approach for dynamic vehicle speed optimization |
GB2541948B (en) * | 2015-09-07 | 2020-02-12 | Jaguar Land Rover Ltd | A verification module for verifying accuracy of a controller |
US10662765B2 (en) * | 2015-09-18 | 2020-05-26 | Schlumberger Technology Corporation | Wellsite emissions monitoring and control |
US9657670B2 (en) | 2015-10-02 | 2017-05-23 | GM Global Technology Operations LLC | Exhaust system temperature estimation systems and methods |
US9644548B2 (en) | 2015-10-02 | 2017-05-09 | GM Global Technology Operations LLC | Exhaust system pressure estimation systems and methods |
US10415492B2 (en) | 2016-01-29 | 2019-09-17 | Garrett Transportation I Inc. | Engine system with inferential sensor |
US10769517B2 (en) * | 2016-03-05 | 2020-09-08 | Fujitsu Limited | Neural network analysis |
JP6494009B1 (en) * | 2016-03-10 | 2019-04-03 | シグニファイ ホールディング ビー ヴィ | Pollution estimation system |
US10124750B2 (en) | 2016-04-26 | 2018-11-13 | Honeywell International Inc. | Vehicle security module system |
US10690344B2 (en) * | 2016-04-26 | 2020-06-23 | Cleaver-Brooks, Inc. | Boiler system and method of operating same |
US10036338B2 (en) | 2016-04-26 | 2018-07-31 | Honeywell International Inc. | Condition-based powertrain control system |
CN105727722B (en) * | 2016-05-03 | 2019-01-15 | 福建省恒创环保科技有限公司 | Flue gas desulfurization, denitrification process monitoring management system |
US9938908B2 (en) | 2016-06-14 | 2018-04-10 | GM Global Technology Operations LLC | System and method for predicting a pedal position based on driver behavior and controlling one or more engine actuators based on the predicted pedal position |
DE102016216951A1 (en) * | 2016-09-07 | 2018-03-08 | Robert Bosch Gmbh | Model calculation unit and controller for selectively calculating an RBF model, a Gaussian process model and an MLP model |
US20180137218A1 (en) * | 2016-11-11 | 2018-05-17 | General Electric Company | Systems and methods for similarity-based information augmentation |
WO2018101918A1 (en) | 2016-11-29 | 2018-06-07 | Honeywell International Inc. | An inferential flow sensor |
US20180189647A1 (en) * | 2016-12-29 | 2018-07-05 | Google, Inc. | Machine-learned virtual sensor model for multiple sensors |
US11057213B2 (en) | 2017-10-13 | 2021-07-06 | Garrett Transportation I, Inc. | Authentication system for electronic control unit on a bus |
JP6760317B2 (en) * | 2018-03-14 | 2020-09-23 | オムロン株式会社 | Learning support device |
EP3776105B1 (en) | 2018-03-29 | 2022-05-04 | Ammann Schweiz AG | Emission control of an asphalt mixing plant |
EP4290412A3 (en) | 2018-09-05 | 2024-01-03 | Sartorius Stedim Data Analytics AB | Computer-implemented method, computer program product and system for data analysis |
DE102018125909A1 (en) * | 2018-10-18 | 2020-04-23 | Endress+Hauser Flowtec Ag | Determination method for determining a process variable |
CN109633195B (en) * | 2018-12-29 | 2022-05-17 | 上海一谱仪器科技股份有限公司 | Automatic sample introduction system of infrared carbon and sulfur analyzer |
WO2020180424A1 (en) | 2019-03-04 | 2020-09-10 | Iocurrents, Inc. | Data compression and communication using machine learning |
CN110273738B (en) * | 2019-03-14 | 2023-06-02 | 吉林大学 | Heavy engine emission diagnosis test system |
US11714114B2 (en) * | 2019-05-20 | 2023-08-01 | Miami University | Non-invasive diagnostic systems and methods for using the same |
JP7231144B2 (en) * | 2019-07-17 | 2023-03-01 | 株式会社トランストロン | Engine control device and neural network program it has |
US10972261B1 (en) * | 2019-10-18 | 2021-04-06 | Via Science, Inc. | Secure data processing |
US11768917B2 (en) | 2019-11-14 | 2023-09-26 | International Business Machines Corporation | Systems and methods for alerting to model degradation based on distribution analysis |
US11256597B2 (en) | 2019-11-14 | 2022-02-22 | International Business Machines Corporation | Ensemble approach to alerting to model degradation |
US11455561B2 (en) | 2019-11-14 | 2022-09-27 | International Business Machines Corporation | Alerting to model degradation based on distribution analysis using risk tolerance ratings |
US11810013B2 (en) | 2019-11-14 | 2023-11-07 | International Business Machines Corporation | Systems and methods for alerting to model degradation based on survival analysis |
KR102202926B1 (en) * | 2020-01-10 | 2021-01-14 | 주식회사 현대케피코 | A real time engine control method for reducing NOx in use with deep-learning |
CN115136079A (en) * | 2020-02-20 | 2022-09-30 | 赢创运营有限公司 | Method and system for process control |
US11691637B2 (en) * | 2020-06-19 | 2023-07-04 | Ghost Autonomy Inc. | Handling input data errors in an autonomous vehicle |
JP6950911B1 (en) * | 2020-07-31 | 2021-10-13 | 株式会社アイデミ— | Demand forecast program, demand forecast device, demand forecast method, demand forecast notification program, demand forecast notification device and demand forecast notification method |
US11686650B2 (en) | 2020-12-31 | 2023-06-27 | Robert Bosch Gmbh | Dynamic spatiotemporal beamforming |
US20220205451A1 (en) * | 2020-12-31 | 2022-06-30 | Robert Bosch Gmbh | Sensing via signal to signal translation |
US20230065744A1 (en) * | 2021-08-26 | 2023-03-02 | Halliburton Energy Services, Inc. | Graphical user interface for abating emissions of gaseous byproducts from hydrocarbon assets |
DE102022201637A1 (en) | 2022-02-17 | 2023-08-17 | Zf Friedrichshafen Ag | Computer-implemented method and system for creating a virtual clone of a sensor |
Family Cites Families (41)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
DE2245029C3 (en) * | 1972-09-14 | 1981-08-20 | Robert Bosch Gmbh, 7000 Stuttgart | Method and device for exhaust gas decontamination from internal combustion engines |
DE2247656C3 (en) * | 1972-09-28 | 1981-12-17 | Robert Bosch Gmbh, 7000 Stuttgart | Device for regulating the ratio of the fuel and air components of the operating mixture of an internal combustion engine |
DE2251167C3 (en) * | 1972-10-19 | 1986-07-31 | Robert Bosch Gmbh, 7000 Stuttgart | Device for exhaust gas detoxification from internal combustion engines |
JPS4987914A (en) * | 1972-12-26 | 1974-08-22 | ||
DE2301353A1 (en) * | 1973-01-12 | 1974-07-25 | Bosch Gmbh Robert | CONTROL DEVICE FOR EXHAUST GAS DETOXIFICATION FROM COMBUSTION MACHINERY |
DE2301354C3 (en) * | 1973-01-12 | 1981-03-12 | Robert Bosch Gmbh, 7000 Stuttgart | Device for regulating the fuel-air ratio in internal combustion engines |
US4007589A (en) * | 1973-01-31 | 1977-02-15 | Robert Bosch G.M.B.H. | Internal combustion exhaust catalytic reactor monitoring system |
DE2304622A1 (en) * | 1973-01-31 | 1974-08-15 | Bosch Gmbh Robert | DEVICE FOR MONITORING CATALYTIC REACTORS IN EXHAUST GAS DETOXIFICATION SYSTEMS OF COMBUSTION MACHINES |
JPS6014183B2 (en) * | 1975-11-11 | 1985-04-11 | 株式会社日本自動車部品総合研究所 | Air flow adjustment device |
JPS52110333A (en) * | 1976-03-08 | 1977-09-16 | Nissan Motor Co Ltd | Fuel-air ratio control device |
US4167161A (en) * | 1976-07-09 | 1979-09-11 | Mitsubishi Jidosha Kogyo Kabushiki Kaisha | Directional auxiliary intake injection for internal combustion engine |
US4245314A (en) * | 1978-02-27 | 1981-01-13 | The Bendix Corporation | Oxygen sensor qualifier |
US4315243A (en) * | 1980-09-16 | 1982-02-09 | Calvert Sr Willard R | Unused fuel indicator for automotive engines employing catalytic converters |
US4403473A (en) * | 1981-06-22 | 1983-09-13 | Caterpillar Tractor Co. | Ammonia/fuel ratio control system for reducing nitrogen oxide emissions |
US4438497A (en) * | 1981-07-20 | 1984-03-20 | Ford Motor Company | Adaptive strategy to control internal combustion engine |
JPS59195012A (en) * | 1983-04-20 | 1984-11-06 | Hitachi Ltd | Combustion control method |
US4548185A (en) * | 1984-09-10 | 1985-10-22 | General Motors Corporation | Engine control method and apparatus |
JPS61169635A (en) * | 1985-01-23 | 1986-07-31 | Hitachi Ltd | Air-fuel ratio controlling method |
US4663703A (en) * | 1985-10-02 | 1987-05-05 | Westinghouse Electric Corp. | Predictive model reference adaptive controller |
US4789939A (en) * | 1986-11-04 | 1988-12-06 | Ford Motor Company | Adaptive air fuel control using hydrocarbon variability feedback |
AT388780B (en) * | 1987-09-09 | 1989-08-25 | Jenbacher Werke Ag | DEVICE FOR DETECTING IGNITION AND FLAME BREAKERS |
US5003950A (en) * | 1988-06-15 | 1991-04-02 | Toyota Jidosha Kabushiki Kaisha | Apparatus for control and intake air amount prediction in an internal combustion engine |
DE3911145C1 (en) * | 1989-04-06 | 1990-04-26 | Daimler-Benz Aktiengesellschaft, 7000 Stuttgart, De | |
JPH0711256B2 (en) * | 1989-09-06 | 1995-02-08 | 本田技研工業株式会社 | Control device for internal combustion engine |
DE3931952A1 (en) * | 1989-09-25 | 1991-04-04 | Ind Tech Res Inst | DEVICE FOR AUTOMATICALLY ADJUSTING THE IDLE EMISSION OF AUTOMOTIVE AND MOTORCYCLES |
JP2792633B2 (en) * | 1990-02-09 | 1998-09-03 | 株式会社日立製作所 | Control device |
DE4004085A1 (en) * | 1990-02-10 | 1991-08-14 | Bosch Gmbh Robert | METHOD AND DEVICE FOR ELECTRONIC CONTROL AND / OR REGULATION OF AN INTERNAL COMBUSTION ENGINE OF A MOTOR VEHICLE |
US5222471A (en) * | 1992-09-18 | 1993-06-29 | Kohler Co. | Emission control system for an internal combustion engine |
US5077970A (en) * | 1990-06-11 | 1992-01-07 | Ford Motor Company | Method of on-board detection of automotive catalyst degradation |
FR2667113B1 (en) * | 1990-09-26 | 1993-06-25 | Semt Pielstick | METHOD FOR MONITORING THE EMISSION OF NITROGEN OXIDES BY AN INTERNAL COMBUSTION ENGINE. |
US5273019A (en) * | 1990-11-26 | 1993-12-28 | General Motors Corporation | Apparatus with dynamic prediction of EGR in the intake manifold |
US5228335A (en) * | 1991-02-25 | 1993-07-20 | The United States Of America As Represented By The United States Environmental Protection Agency | Method and apparatus for detection of catalyst failure on-board a motor vehicle using a dual oxygen sensor and an algorithm |
JPH0781701B2 (en) * | 1991-04-05 | 1995-09-06 | 川崎重工業株式会社 | A device for estimating unburned content in ash of a coal combustion furnace |
US5177464A (en) * | 1991-09-04 | 1993-01-05 | Ford Motor Company | Catalyst monitoring using a hydrocarbon sensor |
US5138163A (en) * | 1991-09-09 | 1992-08-11 | Ford Motor Company | Direct sampling of engine emissions for instantaneous analysis |
US5163412A (en) * | 1991-11-08 | 1992-11-17 | Neutronics Enterprises, Inc. | Pollution control system for older vehicles |
US5349541A (en) * | 1992-01-23 | 1994-09-20 | Electric Power Research Institute, Inc. | Method and apparatus utilizing neural networks to predict a specified signal value within a multi-element system |
US5213080A (en) * | 1992-07-10 | 1993-05-25 | Gas Research Institute | Ignition timing control |
US5220905A (en) * | 1992-07-17 | 1993-06-22 | Brad Lundahl | Reducing emissions using transport delay to adjust biased air-fuel ratio |
US5271674A (en) * | 1992-12-21 | 1993-12-21 | Riley Storker Corporation | Apparatus and method for predicting ash deposition on heated surfaces of a fuel burning combustion vessel |
US5539638A (en) * | 1993-08-05 | 1996-07-23 | Pavilion Technologies, Inc. | Virtual emissions monitor for automobile |
-
1993
- 1993-11-05 US US08/149,216 patent/US5539638A/en not_active Expired - Lifetime
-
1994
- 1994-07-27 WO PCT/US1994/008657 patent/WO1995004878A1/en active IP Right Grant
- 1994-07-27 AU AU73758/94A patent/AU688353B2/en not_active Ceased
- 1994-07-27 KR KR1019960700595A patent/KR960704258A/en not_active Application Discontinuation
- 1994-07-27 EP EP94922772A patent/EP0712463B1/en not_active Expired - Lifetime
- 1994-07-27 AU AU74776/94A patent/AU677694B2/en not_active Ceased
- 1994-07-27 CA CA002167588A patent/CA2167588A1/en not_active Abandoned
- 1994-07-27 EP EP94924530A patent/EP0712509B1/en not_active Expired - Lifetime
- 1994-07-27 CA CA002167927A patent/CA2167927A1/en not_active Abandoned
- 1994-07-27 WO PCT/US1994/008628 patent/WO1995004957A1/en active IP Right Grant
- 1994-07-27 DE DE69423895T patent/DE69423895T2/en not_active Expired - Lifetime
- 1994-07-27 JP JP7506469A patent/JPH09504346A/en active Pending
- 1994-07-27 JP JP7506465A patent/JPH09501782A/en active Pending
- 1994-07-27 DE DE69418199T patent/DE69418199T2/en not_active Expired - Lifetime
-
1996
- 1996-02-05 KR KR1019960700596A patent/KR960704150A/en not_active Application Discontinuation
- 1996-07-23 US US08/685,072 patent/US5682317A/en not_active Expired - Lifetime
Also Published As
Publication number | Publication date |
---|---|
JPH09501782A (en) | 1997-02-18 |
US5539638A (en) | 1996-07-23 |
DE69418199D1 (en) | 1999-06-02 |
AU677694B2 (en) | 1997-05-01 |
DE69423895T2 (en) | 2000-12-07 |
KR960704150A (en) | 1996-08-31 |
DE69423895D1 (en) | 2000-05-11 |
US5682317A (en) | 1997-10-28 |
KR960704258A (en) | 1996-08-31 |
EP0712509A1 (en) | 1996-05-22 |
AU688353B2 (en) | 1998-03-12 |
DE69418199T2 (en) | 1999-12-30 |
WO1995004878A1 (en) | 1995-02-16 |
EP0712463B1 (en) | 1999-04-28 |
JPH09504346A (en) | 1997-04-28 |
EP0712509B1 (en) | 2000-04-05 |
CA2167588A1 (en) | 1995-02-16 |
EP0712463A1 (en) | 1996-05-22 |
AU7375894A (en) | 1995-02-28 |
WO1995004957A1 (en) | 1995-02-16 |
AU7477694A (en) | 1995-02-28 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CA2167927A1 (en) | Virtual emissions monitor for automobile | |
US5386373A (en) | Virtual continuous emission monitoring system with sensor validation | |
JP4319461B2 (en) | NOx emission control system using virtual sensor | |
Pellegrinetti et al. | Nonlinear control oriented boiler modeling-a benchmark problem for controller design | |
CA2601446C (en) | Predictive emissions monitoring system and method | |
US20040230368A1 (en) | Method and apparatus to diagnose intake airflow | |
JP2010537192A (en) | System and method for virtual sensing based on empirical ensemble of gas emissions | |
Manzie et al. | Gaussian networks for fuel injection control | |
CN101755115A (en) | Abnormality detection device for internal combustion engine and air/fuel ratio control apparatus for internal combustion engine | |
US20020103547A1 (en) | Combustion vibration estimation apparatus, plant and gas turbine plant | |
Dong et al. | Emission monitoring using multivariate soft sensors | |
Hanzevack et al. | Virtual sensors for spark ignition engines using neural networks | |
WO2016111581A1 (en) | Apparatus and method for controlling egr valve | |
Sediako et al. | Heavy duty diesel engine modeling with layered artificial neural network structures | |
Romessis et al. | Setting up of a probabilistic neural network for sensor fault detection including operation with component faults | |
Mandloi et al. | Objective Determination of Degradation of Lambda Sensor Using Recursive Least Square Method | |
KrishnaKumar et al. | Jet engine performance estimation using intelligent system technologies | |
Copp et al. | Fuzzy modelling techniques applied to an air/fuel ratio control system | |
Arsie et al. | Experimental validation of a recurrent neural network for air-fuel ratio dynamic simulation in SIIC engines | |
Ward et al. | Bayesian statistics in engine mapping | |
Reed et al. | Bayesian Belief Networks for Fault Identification in Aircraft Gas Turbines | |
Hamer et al. | Optimization of thrust algorithm calibration for Computing System (TCS) for Thrust the NASA Highly Maneuverable Aircraft Technology (HiMAT) vehicle's propulsion system | |
Swanson | A cost effective advanced emissions monitoring solution for gas turbines: statistical hybrid predictive system that accurately measures nitrogen oxides, carbon monoxide, sulfur dioxide, hydrocarbon and carbon dioxide mass emission rates | |
Long | Computing engine efficiencies by neural nets | |
Shayler et al. | Applications of neural networks in automotive engine management systems |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
EEER | Examination request | ||
FZDE | Discontinued | ||
FZDE | Discontinued |
Effective date: 20020426 |