US20030187624A1

US20030187624A1 - CNC control unit with learning ability for machining centers

Info

Publication number: US20030187624A1
Application number: US10/217,043
Authority: US
Inventors: Joze Balic
Original assignee: Individual
Current assignee: Individual
Priority date: 2002-03-27
Filing date: 2002-08-12
Publication date: 2003-10-02
Also published as: SI21200A; US20050251284A1; US7117056B2

Abstract

A CNC control unit 1 with learning ability solves the problem of automatic and intelligent generating of NC programs for CNC machining centers for milling, drilling and similar operations. The key module of the CNC control unit 1 is a neural network (NN) device 7, that learns to generate the NC control programs through an NN teaching module 4. Upon completion of learning process the NN device 7 can generate automatically, without any intervention of the operator, merely on the basis of the CAD 2D, 2,5D or 3D part models, taken from a conventional CAD/CAM system 29, various new NC control programs for different parts, which have not been in the machining process before. The CNC control unit 1 with learning ability is suitable especially for machining centers intended for milling, including face milling (rough), contour milling (rough), final milling following the contour and in Z-plane, final contour 3D milling, contour final milling, milling in Z-plane, final contour milling on the equidistant, and milling of pockets; drilling, including normal drilling, deep drilling, and center drilling; and reaming, sinking and threading.

Description

FIELD OF THE INVENTION

The invention relates to manufacturing technology, in particular to programming and numerical control of machining centers for milling, drilling and similar operations.

The conventional control units for the CNC (Computer Numerical Control) machining tools, especially the control units for machining centers intended for milling, drilling and similar operations, typically contain a microcomputer, consisting of a decoder, a position memory, a function memory, an interpolation program, and a functions program. The NC program is fed to the computer either through a punched tape reading device or in a DNC (Direct Numerical Control) mode through an interface. A manual input is also foreseen, but it is limited to smaller corrections of the NC control program or to individual changes of technological parameters. The NC functions program comprising the technological data is sent to the NC machine through an adaptable interface. The NC positions program is send through a comparison unit and an amplifier unit to a step motor of the NC machine. Either the support for the part or the cutting tools of the machine can be moved following the geometric data. A position meter perceives every movement and sends a regulated position value to a position-measuring module, which forwards the data to comparison unit, where the difference between the actual and the desired position is calculated. The geometric data is obtained from the NC control program for the part and is processed in the position loop.

A new NC control program must be supplied to the CNC control system for each part, as the control system does not remember the operations already performed and can not automatically change the program parameters, with the exception of some technological parameters, for example the cutting conditions, corrections of tool length, offset of reference or zero points.

The use of modern CAD/CAM systems does not solve this problem. These systems just enable that a new NC control program is performed faster and more reliably. Some systems allow saving of certain processing strategies, nevertheless, the intervention of skilled NC programmer is still necessary. The programs made in this way cannot be used directly for the CNC control of the machine tool; they must be adapted in a post-processing phase. The task of this phase is to modify a generally valid file of the tool path (CLDATA) for each machine tool, i.e. for each CNC control. Every NC control program and every change must go through such post-processing phase.

The technical problems indicated above can be solved by an NC control system with learning ability and the ability of automatic intelligent generation of NC control programs which follow the computer read engineering drawings and/or 3D CAD models of the parts to be processed.

In the patent DE4O1 1591 (JP19890098177) an NC control unit with integrated learning function is described. The NC control unit makes a teaching NC control program that is compared with the inserted NC control program to make the resulting NC control program. The actual NC control program can be changed or supplemented by the machine operator who chooses the “teaching” way of operation and then supplements the NC control program. The solution requires the intervention of a skilled operator or a programmer.

The patent application US2001/0000805 A1 describes a device for generating a tool path on NC machines and the pertinent NC control system. First, the device identifies the geometric feature characteristics of a CAD model, and then it chooses the most suitable tool path amongst the stored processing procedures (machining cycles, sub-programs). Only the machining procedures, which have been defined as typical processing procedures for particular sub-programs, are available for selection.

The patent U.S. Pat. No. 6,022,132 describes a method and a device for programming the CNC machine tools on the basis of a probe built-in into the main spindle of the machine tool. The probe is moved manually following the required profile (tool path). After receiving the data the computer generates an automatic NC control program, which gives the commands for the movement to the CNC control system. This method does not include any elements of artificial intelligence.

The patent US63 14412B1 describes an evolutional control of a driving machine in a vehicle with respect to chosen coefficients. A scheme of the control unit is constructed on the evolution principles. The system is adapted for building-in into a vehicle.

The patent EP0726509A1 describes an interactive programming system for CNC machine tools. It allows the operator to choose interactively between different control programs and procedures, which are then automatically composed into an NC control program. The solution requires the intervention of an operator or a programmer.

The patent JP2001034155 describes a learning method and a device made for this purpose. A special man-machine interface is built-in into the control unit of the machine to enable a conversation with the user and the learning process.

The patent JP 11242510 describes a device and a method for generating the NC control programs. A special device saves the data about the part, pertinent coordinates, junction's characteristics and the time necessary for assembling the individual electronic components. The solution enables a reduction of time needed for preparing the NC control programs and a reduction of mistakes arising at preparing the programs.

In all the solutions presented above manual intervention of a skilled operator or a programmer is necessary for preparing the NC control program for CNC machine tools. The systems cannot create the NC control programs for the parts, which are not saved in the database, and cannot choose and use the machining strategies automatically.

DETAILED DESCRIPTION OF THE INVENTION

The object of the invention is to provide an improved programming and numerical control for machining centers intended for milling, drilling and similar operations which has the learning ability and the ability of automatic intelligent generation of NC control programs. The said object of the invention is achieved by means of a neural network (NN), which learns to generate NC control programs through a teaching module. Consequently, the NC control programs can be generated automatically without any intervention of the operator, merely on the basis of the 2D, 2,5D or 3D computer models of the parts to be processed.

The objects, advantages and features of the invention will be presented in detail by means of drawings in the following figures: [0015]
FIG. 1—shows a block diagram of a CNC unit with learning ability for machining centers according to present invention, [0016]
FIG. 2—shows a schematic layout of a neural network device, FIG. 3—shows a flow chart of learning and generating the neural network, FIG. 4—shows a schematic layout of the neural network.[0017]
The learning process and the automatic intelligent generation of the [0018] NC control programs 28 take place in a neural network (NN), built-in in a special NN device 7, which receives the learning instructions from the NN teaching module 4. The NN teaching module 4 is not a constituent part of the CNC unit 1 and works independently. Upon completion of learning process the NN device 7 can generate automatically, merely on the basis of the CAD part model 5, coming from conventional CAD/CAM system 29, and without any intervention by the operator, various new NC control programs 28 for different parts, which have not been in the machining process before.
The [0019] NC control programs 28 are fed from the NN device 7 to a modified microcomputer 2, which includes internal interface 9 for transmission of NC control programs 28 to a position memory 11 and to a function memory 12. To the function memory 12, the manual commands 6 from the manual input module 8 can be fed as well, to wit through the decoding module 10. The commands 6 are mostly of technological nature, i.e. feed rate, revolution speed, switch on/off of cooling liquid etc.
The teaching data for the [0020] NN device 7 come from a special teaching module 4, which is not a constituent part of the CNC control unit 1. The task of the teaching module 4 is to teach the neural network in the NN device 7 the principles and the technology of NC programming for all the machining operations on CNC machining centers, above all for milling, drilling and similar operations.
In general, different neural network systems and different software products, also software developed for commercial purposes, can be applied. However, if special criteria have to be considered and met in machining processes, e.g. costs, time, quality of cutting, tool life, high speed cutting etc., the neural networks developed especially for specific purposes should be used. [0021]
The schematic diagram of the [0022] NN device 7 according to the invention is shown in FIG. 3. The NN device 7 consists of a module 25 designed to recognize geometric and technological features from CAD part model 5 and to generate the features based CAD part model 26. The CAD part model 26 is fed to the NN milling module 27, which has before that, namely in the learning phase, been instructed by the NN teaching module 4 to generate the specific NC control program 28 for specific machining operation, e.g. for milling or drilling or similar operation.
In the learning phase, the [0023] N7N device 7 is connected to the teaching module 4 designed for instructing the neural network (NN). The teaching module 4 takes the data from conventional, commercially available CAD/CAM system for programming the NC/CNC machine tools. By means of a conventional CAD/CAM system 29, the teaching NC programs 36 are prepared for different parts, defined in engineering drawings module 35, and are sent to the teaching module 4. In the decision module 38, subsequent to the testing module 37, the decision is taken on the success of teaching. In case that the decision is NO, the path 39 is active and the repetition of the teaching process takes place. If on the other hand the NN device 7 has learned enough, the path 40 is active and the generated neural network is sent to the NN device 7.
The functioning principle of the [0024] NN device 7 is shown on FIG. 4. The neural network built-in in the NN device 7 consists of three layers: the input layer 43, the hidden layer 44 and the output layer 45. On the input layer 43, the X-Y-Z sets 42 of coordinate points appear, representing the coordinate point values obtained from the modified CAD model 26 for individual machining operations types 41. The coordinate point values are determined according to special procedure. Through the intermediate hidden layer 44 the input coordinates are transposed into output layer 45 in a form of a set of coordinate points X₁, Y₁, Z ₁ 46, representing the position values of the tool path for individual machining operations.
By means of the neural network the following machining operations can be carried out: face milling (rough), contour milling (rough), final milling after the contour and in Z-plain, final contour 3D milling, contour final milling, milling on Z-plain, final contour milling on equidistant, milling of pockets, normal drilling, deep drilling, centering, reaming and threading. [0025]
The [0026] CNC control unit 1 according to the invention can function in either of the following two modes:
1. Programming mode, i.e. the mode of intelligent and completely automated processing of a CAD part model into a specific NC control program. [0027]
2. Learning mode, in which a learned NC programming system based on the principle of a neural network is entered through the [0028] teaching module 4 into the NN device 7.
The principle of generation of the NC control program is shown in FIG. 4. In the programming mode, the [0029] CNC control unit 1 gets the data package of the CAD part model 5 from the conventional, commercially available CAD/CAM system 29 intended for programming the CNC machines. The model is then transmitted to the NN device 7, which identifies and classifies the individual geometric and technological features 25 of the CAD part model. Based on these characteristic features a new CAD part model 26 is built, which is transmitted to the N7N milling module 27, where on the basis of learned intelligent procedures the most suitable machining operations and cutting parameters (cutting speed, feed-rate and the depth of cutting) with respect to chosen conditions (machining time, surface quality, machining costs) are defined.
The output of the [0030] NN milling module 27 is the NC control program 28 for the processed part, which includes the geometric data about the mode of cutting tool path (linear G01 or circular G02/G03 interpolation), the coordinates of the cutting tool path (e.g. milling cutter), the technological data (revolution speed, feed-rate, depth of cutting) and auxiliary data (coordinates of reference, zero and starting points, direction of rotation of the main spindle M02/M03, change of cutting tools M06, etc.).
The data is then transmitted to [0031] internal interface 9, which splits the data in the NC control program into tool path data (coordinates of movement in axis X, Y, Z and/or rotation A, B, C around coordinate axis X, Y, Z) saved in position memory 11 and into functions data (M, S, T) saved in function memory 12.
The NC functions [0032] program 14, which contains the technological data, is transmitted through adaptable interface 18 to the NC machine 3. The NC position program is then sent through the comparison unit 15 and the amplifier unit 17 to the step motors 19 of the NC machine 3. Either the machine tool slide 32 or the cutting tools 31 can be moved in accordance with geometric data 24. The position meter 20 perceives the movement and sends a regulated value 22 into the position-measuring module 16, which transmits the data to comparison unit 15, where the difference between the actual and the programmed position is calculated.
The geometric data are obtained from the [0033] NC control program 28 for each part and are treated in the position regulation circle 23.
In the learning mode, the learned NC programming system based on the principle of a neural network is fed to the [0034] NN device 7 through the teaching module 4, which conducts the teaching of the NN device 7. The functioning of the NN module is schematically shown in FIG. 3.
The origin for the teaching process is the engineering drawing [0035] 35 of a prismatic part, suitable for processing on machining centers, designed for milling, drilling and similar operations. First, the teaching NC program 36 is generated by the conventional CAD/CAM system 29 and sent to the NN teaching module 4. Then, testing 37 of the obtained NC program is performed. In the decision module 38, the decision is brought on whether the NC control program is suitable and whether the neural network in the NN teaching module 4 has learned enough. In the beginning the statement NO 39 is valid and the teaching process is repeated using the engineering drawing 35 of another part. In such a way the series of teaching cycles is performed until the testing 37 shows, that the decisional condition in the IF module 38 is fulfilled, i.e. that the state 40 is accomplished. Here, the teaching process of the NN module ends and the learned neural network is transmitted into the NN device 7.
The [0036] CNC control unit 1 can learn how to generate the NC control programs for the following machining procedures:
2.1—milling [0037]
2.1.1—face milling (rough) [0038]
2.1.2—contour milling (rough) [0039]
2.1.3—final milling following the contour in Z-plane [0040]
2.1.4—final contour 3D milling [0041]
2.1.5—contour final milling [0042]
2.1.6—milling on Z-plane [0043]
2.1.7—final contour milling on equidistant [0044]
2.1.8—milling of pockets [0045]
2.2—drilling [0046]
2.2.1—normal drilling [0047]
2.2.2—deep drilling [0048]
2.2.3—centering [0049]
2.3—reaming, [0050]
2.3—sinking [0051]
2.4—threading [0052]
The [0053] NN device 7 can be built-in into any CNC control unit for milling machines as shown in FIG. 1. The standard parallel data transmission is used. In case that it is not possible to reprogram the internal interface 9, the NN device must be connected to existing DNC interface, which is a constituent part of every CNC control. The NN teaching module 4 is connected to the NN device 7 by means of a standard serial interface. The CAD part model 6 is sent to the NN device 7 through a standard communication interface.
For teaching of the [0054] NN device 7 through the teaching module 4, different commercially available CAD/CAM programming systems 29 can be used, for example Unigraphics Solution, I-Deas, Catia, HyperMill etc.

Claims

1. A CNC control unit 1 for machining centers for milling, drilling and similar operations with learning ability and the ability of automatic intelligent generation of NC control programs 28, comprising a neural network (NN) device 7, a modified microcomputer 2, a comparison unit 15, a position measuring unit 16 and an amplifier 17, the improvement comprising the NN device 7 that takes instructions from the NN teaching module 4, which is not a constituent part of the CNC control unit 1 and operates independently taking as a basis for its operation the data package received from a conventional CAD&CAM system, in order to operate as intelligent programming module after the teaching process is completed and to automatically generate adequate NC control program 28 for a given part to be processed in the machining center; the said NC control program 28 being fed to the modified micro computer 2 including an internal interface 9 with its first output connected to the position memory 11 and its second output connected to the function memory 12, which accepts also the manual input commands 6 fed through the manual input module 8 and the decoding unit 10; the modified microcomputer 2 further includes a function program module 14, with its input connected to the function memory 12 and its output connected to the adaptable interface system 18, and an interpolation program module 13 with its input connected to the position memory 11 and its output connected to the comparison unit 15 which has its other input connected to the output of the position measuring module 16, to which the position data 22 are fed from a position meter 20, while the output of the comparison unit 15 is connected to an amplifier 17 which feds the geometric data 24 to the step motor 19.

2. The CNC control unit 1 for machining centers defined in claim 1, wherein in the mode of intelligent and automated processing of a CAD part model 5 into a specific NC control program 28 the data package of the CAD part model 5 is fed to the N7N device 7, which first identifies and classifies the individual geometric and technological features 25 of the CAD part model 5 and then builds a new CAD part model 26, which is transmitted to the NN milling module 27, supplying on its output the NC control program 28 for the processed part, the said NC control program 28 being fed to the internal interface 9, which splits the data in the NC control program 28 into a position package saved in the position memory 11 and into function data saved in the function memory 12, wherein the NC function program 14, containing the technological data, is transmitted through adaptable interface 18 to the NC machine 3, while the NC position program, generated in the interpolation program module 13, is sent over the comparison unit 15 and the amplifier unit 17 to the step motors 19 of the NC machine 3 resulting in a suitable movement either of the parts support 32 or of the tools 31 in accordance with geometric data 24, wherein the position meter 20 perceives the movement and sends a regulated size 22 into the position measuring module 16, which transmits the data to comparison unit 15, where the difference between the actual and the programmed position is calculated, while the geometric data are obtained from the NC control program for each part and are treated in the position regulation circle 23.

3. The CNC control unit 1 defined in claim 1, that can automatically generate the NC control programs by means of the instructed neural network contained in the NN module 4 using the engineering drawings 35 of parts, wherein in the learning phase the NN device 7 is connected to the teaching module 4 which takes the teaching NC programs 36 for different parts defined in engineering drawings module 35 from the conventional CAD/CAM system 29, and wherein the decision on the success of teaching is taken in the decision module 38, subsequent to the testing module 37, so that in case the decision is NO, the path 39 is active and the repetition of the teaching process takes place, while on the other hand the NN device 7 has learned enough, the path 40 is active and the generated neural network is sent to the NN device 7.

4. The NN device 7 defined in claims 1, 2, and 3, wherein the said NN device 7 is realized in a microprocessor technique and contains an identification module 25 that recognizes the geometrical and technological features of different CAD part models 5, a generating module 26 that produces new feature based CAD models of parts, an NN milling module 27, and an automatically generated NC control program 28 for a given part.

5. The CNC control unit 1 defined in claim 1 wherein the NN milling module 27 enables intelligent, automatic generating of NC control programs that enable the following machining operations to be executed on prismatic parts: milling, including face milling (rough), contour milling (rough), final milling following the contour and in Z-plane, final contour 3D milling, contour final milling, milling in Z-plane, final contour milling on the equidistant, and milling of pockets; drilling, including normal drilling, deep drilling, and center drilling; and reaming, sinking and threading.

6. The procedure of teaching in the CNC control unit 1 defined in claims 1, 2, 3 and 5, the improvement comprising a process of generation of NC control programs in a conventional CAD/CAM system 29 based on engineering drawings of parts, the said NC control programs serving as teaching NC programs 36 in the NN module 4 that instructs the NN device 7, wherein the decision on the success of teaching is taken in the decision module 38, subsequent to the testing module 37, so that in case the decision is NO, the path 39 is active and the repetition of the teaching process takes place, while in case that the NN device 7 has learned enough, the path 40 is active and the generated neural network is sent to the NN device 7