TWI775059B - Tool wear prediction system using evolutionary fuzzy neural network and method thereof - Google Patents

Abstract

A tool wear prediction system using an evolutionary fuzzy neural network and a system thereof are proposed. The tool wear prediction system is configured to observe a plurality of cutting chips as the basis for tool wear. Since the cutting chips make direct contact with a tool during processing, it can reflect the real situation during processing compared signals obtained by a camera. A color of a surface of the cutting chips is imaged, and the acquired image color information is converted to a plurality of standard CIExy chromaticity coordinate values. The standard CIExy chromaticity coordinate values, a plurality of cumulative cutting time and a plurality of tool wear values are used to establish a prediction model. Therefore, the present disclosure uses an interval type-II fuzzy neural network as the prediction model, and a cooperative whale optimization algorithm to optimize the model parameters.

Description

Tool wear prediction system and method using evolutionary fuzzy neural network

The present invention relates to a tool wear prediction system and method, in particular to a tool wear prediction system and method using evolutionary fuzzy neural network.

In today's machinery industry, Computer Numerical Control (CNC) processing machines have a very high influence. In the process of using tools to process, the tools will inevitably wear, and wear is related to the quality and cost of the overall processing. The timing of changing the tool, too early or too late can increase the cost. In order to avoid the problem of increased cost, the prediction of tool wear is important.

There are two conventional sensing methods for tool wear, one of which is direct sensing, which can be subdivided into: radioactive material sensor, distance sensor, optical sensor and so on. The other is indirect sensing, which can be subdivided into contact type and non-contact type according to the contact or not. Contact type sensors include: dynamometer, accelerometer, current sensor, etc., while non-contact type can be Use acoustic emission sensors. The data obtained by indirect measurement is difficult to directly reflect the actual cutting conditions, and the algorithm of the conventional tool wear prediction technology has the problem that the convergence speed is too fast and it is easy to fall into the regional optimal solution. It can be seen that there is currently a lack of a tool wear prediction system and method that can optimize the model parameters to obtain a more accurate prediction tool wear value. Therefore, relevant researchers are all looking for solutions.

Therefore, the purpose of the present invention is to provide a tool wear prediction system and method using an evolutionary fuzzy neural network, which firstly takes an image of the chip surface color, and then converts the obtained image color information into standard chromaticity parameters , and use standard chromaticity parameters, cutting parameters, cumulative cutting time and tool wear values to build a predictive model. Then, the interval second type fuzzy neural network is used as the prediction model, and the dynamic grouping cooperative differential evolution algorithm is used to optimize the model parameters, so as to solve the problem that the algorithm in the conventional tool wear prediction technology has a too fast convergence speed and is easy to fall into the area. The problem of the best solution.

According to an embodiment of the structural aspect of the present invention, a tool wear prediction system using an evolutionary fuzzy neural network is provided, which is used to predict the actual tool wear value generated by the tool cutting a workpiece. The tool wear prediction system using evolutionary fuzzy neural network includes a cutting machine, a camera and an arithmetic processor, wherein the cutting machine drives the tool to cut the workpiece to generate multiple chips. The camera captures images of each chip. The arithmetic processor signal is connected to the cutting machine and the camera, and the arithmetic processor includes a parameter planning module, a time recording module, a color conversion module and a predicted tool wear value generating module. The parametric planning module provides complex cutting factors based on the Itaguchi orthogonal table Complex cutting parameters are planned, and these cutting factors correspond to the tools. The time recording module records the complex cumulative cutting time of the tool cutting process. The color conversion module receives the image of the chips and converts the image into complex standard chromaticity parameters. The predicted tool wear value generation module signal is connected to the parameter planning module, the time recording module and the color conversion module. The predicted tool wear value generation module generates complex predicted tool wear values according to the cutting parameters, standard chromaticity parameters and accumulated cutting time according to the operation of an interval type 2 fuzzy neural network. The dynamic grouping synergistic differential evolution calculation unit is adjusted to generate a lateral evolution predicted tool wear value and a longitudinal evolution predicted tool wear value, and the predicted tool wear value generation module will predict the tool wear value, the lateral evolution predicted tool wear value and the longitudinal evolution The predicted tool wear value is compared and the one with the smallest error from the actual tool wear value is selected to be updated as an optimal predicted tool wear value.

Thereby, the tool wear prediction system using the evolutionary fuzzy neural network of the present invention uses the interval type II fuzzy neural network as the prediction model, and uses the dynamic grouping cooperative differential evolution calculation unit to optimize the model parameters to solve the problem. The algorithm in the conventional tool wear prediction technology has the problem that the convergence speed is too fast and it is easy to fall into the regional optimal solution.

An embodiment of a method aspect according to the present invention provides a tool wear prediction method using an evolutionary fuzzy neural network, which is used to predict an actual tool wear value generated by a tool cutting a workpiece. The tool wear prediction method using an evolutionary fuzzy neural network includes a parameter planning step, a cutting step, a data collection step, a color conversion step, and a predicted tool wear value generation step. The parametric planning step provides complex cutting factors Complex cutting parameters are planned according to a Taguchi orthogonal table, and these cutting factors correspond to tools. The cutting step is to drive the tool to cut the workpiece to generate multiple chips, and record the multiple accumulated cutting time of the tool during the cutting process. In the data collection step, a camera is driven to capture an image of each chip. The color conversion step converts the image of the chips into complex standard chromaticity parameters. In the step of generating the predicted tool wear value, the cutting parameter, the standard chromaticity parameter and the accumulated cutting time are calculated according to an interval type II fuzzy neural network model to generate a complex number of predicted tool wear value. The interval second type fuzzy neural network model is adjusted by a dynamic grouping collaborative differential evolution algorithm to generate a lateral evolution prediction tool wear value and a longitudinal evolution prediction tool wear value, and the predicted tool wear value generation step will predict the tool wear. value, the predicted tool wear value of lateral evolution and the predicted tool wear value of longitudinal evolution are compared to select the one with the smallest deviation from the actual tool wear value, and update it as an optimal predicted tool wear value.

Thereby, the tool wear prediction method using the evolutionary fuzzy neural network of the present invention uses the interval second type fuzzy neural network model as the prediction model, and uses the dynamic grouping collaborative differential evolution algorithm to optimize the model parameters, so as to To solve the problem that the algorithm in the conventional tool wear prediction technology has too fast convergence speed and is easy to fall into the regional optimal solution.

100: A Tool Wear Prediction System Using Evolutionary Fuzzy Neural Networks

200: Cutting machine

300: Camera

400: arithmetic processor

410: Parameter planning module

420: Time Recording Module

430: Color conversion module

431: Scope selected submodule

432: Color correction submodule

433: first color conversion submodule

434: Second color conversion submodule

435: The third color conversion submodule

440: Predict tool wear value generation module

442: Interval Type II Fuzzy Neural Network

444: Dynamic Grouping Cooperative Differential Evolution Algorithm Unit

4441: Initialize submodule

4442: Grouping submodules

4443: Leader Confirmation Submodule

4444: Mutant Submodule

4445: Swap submodules

4446: select submodule

4447: Leader adjustment submod

4448: Update submodules

4449: Iterations judgment submodule

Layer1: the first layer

Layer2: The second layer

Layer3: The third layer

Layer4: the fourth layer

Layer5: fifth layer

X ₁ , X _n : individual vectors

Y: Best predicted tool wear value

500: A Tool Wear Prediction Method Using Evolutionary Fuzzy Neural Networks

S02: Parameter planning step

S04: Cutting step

S06: Data collection steps

S08: Color conversion steps

S081: Range selection step

S082: Color Correction Steps

S083: The first color conversion step

S084: Second color conversion step

S085: The third color conversion step

S10: Prediction of tool wear value generation step

S102: Interval Type II Fuzzy Neural Network Model

S104: Dynamic Clustering Cooperative Differential Evolution Algorithm

S1041: initialization step

S1042: Grouping step

S1043: Leader Confirmation Step

S1044: Mutation step

S1045: Exchange step

S1046: Selection step

S1047: Leader Adjustment Steps

S1048: Update steps

S1049: Step of judging the number of iterations

FIG. 1 is a block diagram illustrating a tool wear prediction system using an evolutionary fuzzy neural network according to a first embodiment of the present invention;

Fig. 2 is a block diagram of the color conversion module of the tool wear prediction system using evolutionary fuzzy neural network in Fig. 1;

FIG. 3 is a schematic diagram of the interval type 2 fuzzy neural network of the tool wear prediction system using the evolutionary fuzzy neural network of FIG. 1;

Fig. 4 is a block schematic diagram of the dynamic grouping cooperative differential evolution calculation unit of the tool wear prediction system using the evolutionary fuzzy neural network of Fig. 1;

FIG. 5 is a schematic flowchart of a tool wear prediction method using an evolutionary fuzzy neural network according to a second embodiment of the present invention;

FIG. 6 is a schematic flowchart showing the color conversion steps of the tool wear prediction method using the evolutionary fuzzy neural network in FIG. 5; and

FIG. 7 is a schematic flowchart of the dynamic grouping collaborative differential evolution algorithm of the tool wear prediction method using the evolutionary fuzzy neural network in FIG. 5 .

Please refer to FIG. 1 to FIG. 4 together. FIG. 1 is a block diagram of a tool wear prediction system 100 using an evolutionary fuzzy neural network according to the first embodiment of the present invention; FIG. 2 is a schematic diagram of a block diagram. FIG. 1 is a block diagram of the color conversion module 430 of the tool wear prediction system 100 using the evolutionary fuzzy neural network; FIG. 3 shows the tool wear prediction using the evolutionary fuzzy neural network in FIG. 1 A schematic diagram of the interval-type second type fuzzy neural network 442 of the system 100; and FIG. 4 shows the tool wear prediction system 100 using the evolutionary fuzzy neural network of FIG. 1 A block diagram of the dynamic grouping cooperative differential evolution calculation unit 444 of . As shown in the figure, the tool wear prediction system 100 using the evolutionary fuzzy neural network is used to predict the actual tool wear value generated by the tool cutting the workpiece. The tool wear prediction system 100 using an evolutionary fuzzy neural network includes a cutting machine 200 , a camera 300 and an arithmetic processor 400 .

The cutting table 200 drives the tool to cut the workpiece to generate a plurality of chips. Specifically, the cutting machine table 200 may be a five-axis machining machine or other suitable computer numerical control (Computer Numerical Control; CNC) machine tools, but the invention is not limited thereto.

The camera 300 faces the chips generated by the cutting machine 200 and captures images of the chips.

The operation processor 400 is connected to the cutting machine 200 and the camera 300 by signals, and the operation processor 400 includes a parameter planning module 410 , a time recording module 420 , a color conversion module 430 and a predicted tool wear value generation module 440 .

The parameter planning module 410 provides complex cutting factors to plan complex cutting parameters according to a Taguchi orthogonal table, and these cutting factors correspond to tools. Specifically, the cutting parameters include cutting speed (m/min), feed per edge (mm/rev), and depth of cut (mm), while Taguchi orthogonal table is selected to use L ₉ (3 ⁴ ) orthogonal table.

The time recording module 420 records the multiple accumulated cutting times of the tool cutting process.

The color conversion module 430 receives the image of the chips and converts the image into complex standard chromaticity parameters. Specifically, the color conversion module 430 includes a range selection submodule 431 , a color correction submodule 432 , a first color conversion submodule 433 , a second color conversion submodule 434 and a third color conversion submodule 435 . The range selection sub-module 431 selects a central area of each image, for example, selects a range of 300×300 pixels (pixel) in the image for feature extraction. The color correction sub-module 432 is connected to the signal range selection sub-module 431, and the color correction sub-module 432 performs color correction according to a color correction model for the central area to generate a standard color information. The standard color information may be CIELAB color information. The signal of the first color conversion sub-module 433 is connected to the color correction sub-module 432, and the first color conversion sub-module 433 converts the standard color information into a plurality of first stimulus values according to a first standard light source, and these first stimulus values can be is the tristimulus value of XYZ _D50 . The signal of the second color conversion sub-module 434 is connected to the first color conversion sub-module 433. The second color conversion sub-module 434 converts the first stimulus value into a plurality of second stimulus values according to a second standard light source. Stimulus values may be XYZ _D65 tristimulus values. The signal of the third color conversion sub-module 435 is connected to the second color conversion sub-module 434. The third color conversion sub-module 435 converts the second stimulus value into a standard chromaticity parameter according to a standard chromaticity relationship, and the standard chromaticity parameter Can be CIExy color information.

The predicted tool wear value generation module 440 is connected to the parameter planning module 410 , the time recording module 420 and the color conversion module 430 by signals. The predicted tool wear value generating module 440 calculates the cutting parameters, the standard chromaticity parameters and the accumulated cutting time according to an interval 2 type fuzzy neural network 442 to generate a complex predicted tool wear value. The interval second type fuzzy neural network 442 is adjusted by a dynamic grouping cooperative differential evolution calculation unit 444 to generate a lateral evolution predicted tool wear value and a longitudinal evolution predicted tool wear value, and the predicted tool wear value generation module 440 will The predicted tool wear value, the lateral evolution predicted tool wear value and the longitudinal evolution predicted tool wear value are compared to select the one with the smallest error from the actual tool wear value to update as an optimal predicted tool wear value Y. In detail, the Interval Type-II Fuzzy Neural Network 442 (IT2FNN) includes the first layer Layer1, the second layer Layer2, the third layer Layer3, the fourth layer Layer4 and the fifth layer Layer5 . The first layer Layer1, the second layer Layer2, the third layer Layer3, the fourth layer Layer4, and the fifth layer Layer5 are sequentially connected to each other. Each node of the first layer Layer1 is an input node and transmits input values (individual vectors X ₁ to X _n ) to the second layer Layer2. The second layer, Layer2, performs fuzzification operations, and each node is defined as an interval type-2 fuzzy set. The second layer Layer2 includes an interval second-type blur translation amount and an average value and a standard deviation of a Gaussian function. The nodes of the third layer Layer3 are called rule nodes, and each node represents a fuzzy rule. The fourth layer, Layer4, first reduces the interval type-2 fuzzy set to a type-1 fuzzy set through the reduction operation, and then uses the center of gravity to de-fuzz to obtain a clear output. The fourth layer, Layer4, contains a post-authentication weight. The fifth layer, Layer5, deblurs the output of the fourth layer, Layer4, by calculating the average value, so as to obtain the best predicted tool wear value Y. In other words, the interval type-2 fuzzy translation amount, the average value, the standard deviation and the post-identification weight are adjusted by the dynamic grouping cooperative differential evolution calculation unit 444, so that the interval-type type-2 fuzzy neural network 442 outputs the best predicted tool wear value Y.

In addition, the predicted tool wear value generation module 440 may include a dynamic grouping cooperative differential evolution calculation unit 444, and the dynamic grouping cooperative differential evolution calculation unit 444 includes an initialization sub-module 4441, a grouping sub-module 4442, and a leader confirmation sub-module Group 4443, Mutation sub-module 4444, Swap sub-module 4445, Selection sub-module 4446, Leader adjustment sub-module 4447, Update sub-module 4448, and iteration count sub-module 4449.

The initialization sub-module 4441 regards the interval type 2 fuzzy translation amount, the average value, the standard deviation and the post-identification weight as an individual and encodes it, and the interval type 2 fuzzy neural network 442 further includes a plurality of individuals.

The grouping sub-module 4442 is connected to the initialization sub-module 4441 by a signal, and the grouping sub-module 4442 groups the individuals into plural groups according to a group threshold. Specifically, the grouping sub-module 4442 first sorts all individuals according to their fitness values from high to low, and sets the initial value of the individual's group number to 0. After sorting, the individual with the highest fitness value and group number 0 is set as the leader, and the group number is updated to k, and then the group threshold is calculated. The group threshold includes the distance threshold Dis(k) and the fitness threshold Fit( k), as shown in the following formula:

其中NP為個體總數，D為維度，Leader_k,j是第k群的領導者，Dis_Threshold(k)、Fit_Threshold(k)則是第k群的距離閾值與適應值閾值，NG為未被分群個體數。 然後，分群子模組4442計算群組編號為0的個體與領導者的距離差及適應值差，並利用群組閾值來決定個體是否屬於同一群組，如下式所示：

Among them, NP is the total number of individuals, D is the dimension, Leader _k,j is the leader of the kth group, Dis_Threshold(k) and Fit_Threshold(k) are the distance threshold and fitness threshold of the kth group, and NG is the ungrouped individual. number. Then, the grouping sub-module 4442 calculates the distance difference and fitness difference between the individual whose group number is 0 and the leader, and uses the group threshold to determine whether the individual belongs to the same group, as shown in the following formula:

Fit(i)=|Fit(Leader _k )-Fit(X _i )| (6)。若Dis(i)<Dis_Threshold(k)且Fit(i)<Fit_Threshold(k)，則表示此個體與領導者是相似的，故將個體分入此群組並將群組編號更新為k。

Fit ( i )=| Fit ( Leader _k )- Fit ( X _i )| (6). If Dis(i)<Dis_Threshold(k) and Fit(i)<Fit_Threshold(k), it means that this individual is similar to the leader, so the individual is divided into this group and the group number is updated to k.

The leader confirming sub-module 4443 signal is connected to the grouping sub-module 4442, the leader confirming sub-module 4443 confirms whether each individual of each group is a leader, and the leader represents the number of predicted tool wear values of each group of individuals. smallest. Specifically, if one of the individuals in each group is the leader, mutation submodule 4444, exchange submodule 4445, and selection submodule 4446 are executed to generate laterally evolved predicted tool wear values. Conversely, if one of the individuals in each group is not the leader, the leader adjustment sub-module 4447, mutation sub-module 4444, exchange sub-module 4445 and selection sub-module 4446 are executed to generate the longitudinal evolution prediction tool wear value. .

The signal of the mutation sub-module 4444 is connected to the leader confirmation sub-module 4443, and the mutation sub-module 4444 obtains a mutation vector according to the leader's execution of a mutation evolution relation. The mutation evolution relational formula includes a mutation vector V _i , a random leader X _rL , a mutation weight factor F, a first random individual X _r1 and a second random individual X _r2 . The mutation evolution relational formula conforms to the following formula: V _i =X _rL +F×(X _r1 -X _r2 ) (7).

The signal of the exchange sub-module 4445 is connected to the mutation sub-module 4444, and the exchange sub-module 4445 executes an exchange relation according to the mutation vector V _i to obtain a test vector U _i . Specifically, the exchange relation includes a mutation vector V _i , an individual vector X _i and a test vector U _i . The exchange relation is as follows:

其中rand _j 為每個維度對應的隨機值，其值為[0,1]的亂數；CR為交換率，其值為[0,1]。

Among them, rand _j is the random value corresponding to each dimension, and its value is a random number of [0,1]; CR is the exchange rate, and its value is [0,1].

The selection sub-module 4446 is connected to the switching sub-module 4445 by the signal, the selection sub-module 4446 executes a selection relation according to the test vector U _i to obtain the selected individual vector X _i , and the interval second type fuzzy neural network 442 determines the individual The vector X _i operation is used to generate the lateral evolution prediction tool wear value. Specifically, the selection relation includes the test vector U _i and the individual vector X _i , and the selection relation conforms to the following formula:

The leader adjustment submodule 4447 signal connects the leader confirmation submodule 4443 and the mutation submodule 4444. The leader adjustment submodule 4447 arranges the individuals into p×n individual vectors, and divides the p×n individual vectors into n groups p × 1 individual vectors, and then find n vertical leaders of n groups of p × 1 individual vectors. The interval second type fuzzy neural network 442 generates a longitudinal evolution prediction tool wear value according to n longitudinal leader operations. The number of individuals is p, each individual contains a plurality of individual vectors, and the number of individual vectors of each individual is n.

Updated Submodule 4448 Signal Connection Selection Submodule 4446, more The new sub-module 4448 compares the predicted tool wear value, the lateral evolution predicted tool wear value and the longitudinal evolution predicted tool wear value, and selects the one with the smallest error to update the best predicted tool wear value Y.

The number of iterations judging sub-module 4449 signals to connect the updating sub-module 4448 and the sub-module 4442, and the number of iterations judging sub-module 4449 judges whether the number of iterations executed by the sub-module 4442 reaches a preset number of times; if not, Then the grouping sub-module 4442 confirms with the leader that the sub-module 4443 is executed again; if so, the dynamic grouping cooperative differential evolution calculation unit 444 terminates the execution.

Thereby, the tool wear prediction system 100 using the evolutionary fuzzy neural network of the present invention uses the interval second type fuzzy neural network 442 as the prediction model, and uses the dynamic grouping cooperative differential evolution calculation unit 444 to optimize the model parameters , in order to solve the problem that the algorithm in the conventional tool wear prediction technology has too fast convergence speed and is easy to fall into the regional optimal solution.

Please refer to FIG. 1 to FIG. 7 together, wherein FIG. 5 is a schematic flowchart of a tool wear prediction method 500 using an evolutionary fuzzy neural network according to the second embodiment of the present invention; FIG. 6 is a diagram showing the FIG. 5 is a schematic flowchart of the color conversion step S08 of the tool wear prediction method 500 using the evolutionary fuzzy neural network; and FIG. 7 is the tool wear prediction using the evolutionary fuzzy neural network in FIG. 5. A schematic flowchart of the dynamic grouping cooperative differential evolution algorithm S104 of the method 500 . As shown in the figure, the tool wear prediction method 500 using the evolutionary fuzzy neural network is used to predict the actual tool wear value generated by the tool cutting the workpiece. The tool wear prediction method 500 using evolutionary fuzzy neural network includes: Parameter planning step S02, cutting step S04, data collection step S06, color conversion step S08, and predicted tool wear value generation step S10.

The parameter planning step S02 is to provide complex cutting factors to plan complex cutting parameters according to a Taguchi orthogonal table, and these cutting factors correspond to tools. The parameter planning step S02 is executed through the parameter planning module 410 .

The cutting step S04 is to drive the tool to cut the workpiece to generate multiple chips, and record the multiple accumulated cutting times of the tool during the cutting process. The cutting step S04 is performed by the cutting machine 200 and the time recording module 420 .

The data collection step S06 is to drive a camera 300 to capture an image of each chip. The data collection step S06 is performed through the camera 300 .

The color conversion step S08 is to convert the image of the chips into complex standard chromaticity parameters. The color conversion step S08 is performed by the color conversion module 430 . Specifically, the color conversion step S08 includes a range selection step S081, a color correction step S082, a first color conversion step S083, a second color conversion step S084, and a third color conversion step S085. The range selection step S081 is to select the center area of each image, and the range selection step S081 is performed through the range selection sub-module 431 . The color correction step S082 is to perform color correction on the central area according to a color correction model to generate a standard color information. The color correction step S082 is performed by the color correction sub-module 432 . The first color conversion step S083 is to convert the standard color information into a plurality of first stimulus values according to a first standard light source. The first color conversion step S083 is performed by the first color conversion sub-module 433 . The second color conversion step S084 is to convert the first stimulus value into a plurality of second stimulus values according to a second standard light source. The second color conversion step S084 passes through the The second color conversion sub-module 434 executes. The third color conversion step S085 is to convert the second stimulus value into the standard chromaticity parameters according to a standard chromaticity relation. The third color conversion step S085 is performed by the third color conversion sub-module 435 .

The predicted tool wear value generating step S10 is to generate complex predicted tool wear values by operating the cutting parameters, the standard chromaticity parameters and the accumulated cutting time according to the second type fuzzy neural network model S102 in an interval. The interval second type fuzzy neural network model S102 is adjusted by the dynamic grouping collaborative differential evolution algorithm S104 to generate a lateral evolution predicted tool wear value and a longitudinal evolution predicted tool wear value, and the predicted tool wear value generation step S10 will predict the tool wear value, the predicted tool wear value of the lateral evolution and the predicted tool wear value of the longitudinal evolution are compared to select the one with the smallest difference from the actual tool wear value to update as an optimal predicted tool wear value Y. The predicted tool wear value generating step S10 is performed by the predicted tool wear value generating module 440 . The interval second type fuzzy neural network model S102 is executed by the interval second type fuzzy neural network 442 . The dynamic grouping cooperative differential evolution algorithm S104 is executed by the dynamic grouping cooperative differential evolution algorithm unit 444 .

In addition, the dynamic grouping cooperative differential evolution algorithm S104 may include an initialization step S1041, a grouping step S1042, a leader confirmation step S1043, a mutation step S1044, an exchange step S1045, a selection step S1046, a leader adjustment step S1047, an update step S1048, and an iteration The number of times judgment step S1049.

The initialization step S1041 is to set the interval second type fuzzy translation amount, The average value, standard deviation and post-identification weight are regarded as one individual and encoded, and the interval type II fuzzy neural network model S102 includes plural individuals. The initialization step S1041 is performed by the initialization sub-module 4441 .

The grouping step S1042 is to group the individuals into plural groups according to a group threshold. The grouping step S1042 is performed by the grouping sub-module 4442 .

The leader confirming step S1043 is to confirm whether each individual of each group is a leader, and the leader represents the minimum of the plurality of predicted tool wear values of the individuals of each group. The leader confirmation step S1043 is executed through the leader confirmation sub-module 4443 .

The mutation step S1044 is to execute a mutation evolution relational expression on these individuals to obtain a mutation vector. The mutation step S1044 is performed by the mutation sub-module 4444 .

The exchange step S1045 is to execute an exchange relation according to the mutation vector to obtain a test vector. The exchange step S1045 is performed by the exchange sub-module 4445 .

The selection step S1046 is to execute a selection relational expression according to the test vector to obtain a selected body vector, and the interval second type fuzzy neural network model S102 generates a lateral evolution prediction tool wear value according to the calculation of the individual vector. The selection step S1046 is performed through the selection sub-module 4446 .

The leader adjustment step S1047 is to arrange these individuals into p×n individual vectors, and divide the p×n individual vectors into n groups of p×1 individual vectors, and then find n vertical leaders of the n groups of p×1 individual vectors . The interval second type fuzzy neural network model S102 is calculated according to n vertical leaders The longitudinal evolution is generated to predict the tool wear value. The leader adjustment step S1047 is executed through the leader adjustment sub-module 4447 .

The updating step S1048 compares the predicted tool wear value, the lateral evolution predicted tool wear value, and the longitudinal evolution predicted tool wear value, and selects the one with the smallest error to update the best predicted tool wear value Y. The update step S1048 is performed by the update sub-module 4448 .

The iteration count judgment step S1049 is to judge whether an iteration count of the grouping step S1042 reaches a preset number of times; if not, re-execute the grouping step S1042 and the leader confirmation step S1043; if so, end the dynamic grouping cooperative differential evolution algorithm S104 . The step of determining the number of iterations S1049 is performed by the sub-module 4449 for determining the number of iterations.

Thereby, the tool wear prediction method 500 using the evolutionary fuzzy neural network of the present invention uses the interval second type fuzzy neural network model S102 as the prediction model, and uses the dynamic grouping collaborative differential evolution algorithm S104 to optimize the model parameters, in order to solve the problem that the algorithm in the conventional tool wear prediction technology has a too fast convergence speed and is easy to fall into the regional optimal solution.

Although the present invention has been disclosed in the above embodiments, it is not intended to limit the present invention. Anyone skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, the protection of the present invention The scope shall be determined by the scope of the appended patent application.

100: A Tool Wear Prediction System Using Evolutionary Fuzzy Neural Networks

200: Cutting machine

300: Camera

400: arithmetic processor

410: Parameter planning module

420: Time Recording Module

430: Color conversion module

440: Predict tool wear value generation module

442: Interval Type II Fuzzy Neural Network

444: Dynamic Grouping Cooperative Differential Evolution Algorithm Unit

Claims (8)

A tool wear prediction system using an evolutionary fuzzy neural network is used to predict an actual tool wear value generated by a tool cutting a workpiece. The tool wear prediction system using an evolutionary fuzzy neural network includes: a cutting a machine, for driving the tool to cut the workpiece to generate a plurality of chips; a camera for capturing an image of each of the chips; and an arithmetic processor for signally connecting the cutting machine and the camera, the arithmetic processor comprising: a parameter Planning module, providing complex cutting factors to plan complex cutting parameters according to a Taguchi orthogonal table, these cutting factors correspond to the tool; a time recording module, recording the complex cumulative cutting time of the cutting process of the tool; a color conversion module, Receive the images of the chips and convert the images into complex standard chromaticity parameters; and a predicted tool wear value generating module, the signal is connected to the parameter planning module, the time recording module and the color conversion module , the predicted tool wear value generating module generates a complex predicted tool wear value based on the cutting parameters, the standard chromaticity parameters and the accumulated cutting time according to a second type of fuzzy neural network operation in an interval. The type II fuzzy neural network is adjusted by a dynamic grouping cooperative differential evolution calculation unit to generate a lateral evolution predicted tool wear value and a longitudinal evolution predicted tool wear value, and the predicted tool wear value generating module generates these predicted tool wear values. Wear value, the lateral evolution prediction tool wear value and the longitudinal evolution prediction tool The tool wear value is compared to select the one with the smallest error from the actual tool wear value, so as to update it as an optimal predicted tool wear value; wherein, the second type of the interval fuzzy neural network includes a plurality of layers, and these layers connected to each other in sequence, wherein one of the layers includes an interval type 2 fuzzy translation and a mean value and a standard deviation of a Gaussian function, the other layer includes a posterior identification weight, and the interval second type fuzzy translation , the average value, the standard deviation and the weight of the rear identification part are adjusted by the dynamic grouping cooperative differential evolution calculation unit, so that the second type fuzzy neural network in the interval outputs the best predicted tool wear value. The tool wear prediction system using an evolutionary fuzzy neural network as claimed in claim 1, wherein the color conversion module comprises: a range selection submodule for selecting a central area of each of the images; a color correction submodule a group, the signal is connected to the range selection sub-module, the color correction sub-module performs color correction according to a color correction model for the central area, so as to generate a standard color information; a first color conversion sub-module, the signal is connected to the color a correction submodule, the first color conversion submodule converts the standard color information into a plurality of first stimulus values according to a first standard light source; a second color conversion submodule, the signal is connected to the first color conversion submodule a group, the second color conversion sub-module converts the first stimulus values into a plurality of second stimulus values according to a second standard light source; and a third color conversion sub-module, the signal is connected to the second color conversion sub-module group, the third color conversion sub-module bases the second stimulus values on a standard The chromaticity relationship is converted into these standard chromaticity parameters. The tool wear prediction system using evolutionary fuzzy neural network according to claim 1, wherein the predicted tool wear value generating module comprises the dynamic grouping cooperative differential evolution calculation unit, the dynamic grouping cooperative differential evolution calculation unit Including: an initialization sub-module, which takes the interval second type fuzzy translation amount, the average value, the standard deviation and the post-identification weight as an entity and encodes it, the interval second type fuzzy neural network It also includes a plurality of the individuals; a grouping sub-module, the signal is connected to the initialization sub-module, the grouping sub-module groups the individuals into plural groups according to a group threshold; a leader confirms the sub-module, the signal is connected In the grouping sub-module, the leader confirming sub-module confirms whether each of the individuals in each of the groups is a leader, and the leader represents the smallest of the plurality of the predicted tool wear values of the individuals in each of the groups ; a mutation sub-module, the signal is connected to the leader confirmation sub-module, the mutation sub-module obtains a mutation vector according to the leader's execution of a mutation evolution relationship; a switch sub-module, the signal is connected to the mutation sub-module set, the exchange sub-module executes an exchange relationship according to the mutation vector to obtain a test vector; and a selector sub-module connects the exchange sub-module with a signal, and the selector sub-module executes a selection relationship according to the test vector A selected body vector is obtained by using the formula, and the second type fuzzy neural network of the interval generates the lateral evolution prediction tool wear value according to the calculation of the individual vector. The tool wear prediction system using evolutionary fuzzy neural network as described in claim 3, wherein the dynamic grouping cooperative differential evolution calculation unit further comprises: a leader adjustment sub-module, the signal is connected to the leader confirmation sub-module group and the mutant submodule, the leader adjusts the submodule to arrange these individuals into p×n individual vectors, and divides the p×n individual vectors into n groups of p×1 individual vectors, and then finds out n groups of p×1 individual vectors. n longitudinal leaders of an individual vector, the second type of fuzzy neural network in this interval generates the longitudinal evolution prediction tool wear value according to the operation of the n longitudinal leaders; an update sub-module, the signal is connected to the selection sub-module , the update sub-module compares the predicted tool wear values, the lateral evolution predicted tool wear values, and the longitudinal evolution predicted tool wear values, and selects the one with the smallest error to update the best predicted tool wear value; and an iteration The number of times judgment sub-module, the signal connects the update sub-module and the grouping sub-module, and the iteration-count judging sub-module judges whether the number of iterations executed by the grouping sub-module reaches a preset number of times; if not, then The grouping submodule and the leader confirm that the submodule is executed again; if so, the dynamic grouping cooperative differential evolution calculation unit terminates execution. A tool wear prediction method using an evolutionary fuzzy neural network is used to predict an actual tool wear value generated by a tool cutting a workpiece. The tool wear prediction method using an evolutionary fuzzy neural network comprises the following steps: A parametric planning step that provides complex cutting factors based on a Taguchi orthogonal The table plans complex cutting parameters, and these cutting factors correspond to the tool; a cutting step is to drive the tool to cut the workpiece to generate complex chips, and record the complex accumulated cutting time of the tool during the cutting process; a data collection step is to A camera is driven to capture an image of each of the chips; a color conversion step is to convert the images of the chips into complex standard chromaticity parameters; and a predicted tool wear value generation step is to convert the cutting parameters, The standard chromaticity parameters and the accumulated cutting times are calculated according to an interval type 2 fuzzy neural network model to generate complex predicted tool wear values, and the interval type 2 fuzzy neural network model uses a dynamic grouping collaborative formula The differential evolution algorithm is adjusted to generate a lateral evolution predicted tool wear value and a longitudinal evolution predicted tool wear value, and the predicted tool wear value generating step generates these predicted tool wear values, the lateral evolution predicted tool wear value and the longitudinal evolution The predicted tool wear value is compared to select the one with the smallest error from the actual tool wear value, so as to update it as an optimal predicted tool wear value; wherein, the second type of the interval fuzzy neural network model includes a plurality of layers, the The layers are connected to each other in sequence, one of the layers includes an interval type 2 blur translation and an average value and a standard deviation of a Gaussian function, the other layer includes a posterior identification weight, the interval type 2 blur The translation amount, the average value, the standard deviation and the post-identification weight are adjusted by the dynamic grouping cooperative differential evolution algorithm, so that the second type fuzzy neural network model of the interval outputs the best predicted tool wear value. The tool wear prediction method using evolutionary fuzzy neural network as claimed in claim 5, wherein the color conversion step comprises: a range selection step, which is to select a central area of each of the images; a color correction step, which is for The central area performs color correction according to a color correction model to generate a standard color information; a first color conversion step converts the standard color information into a plurality of first stimulus values according to a first standard light source; a second color The conversion step is to convert the first stimulus values into a plurality of second stimulus values according to a second standard light source; and a third color conversion step is to convert the second stimulus values to a standard chromaticity relationship according to a standard chromaticity relationship. These standard chromaticity parameters. The tool wear prediction method using an evolutionary fuzzy neural network as described in claim 5, wherein the dynamic grouping collaborative differential evolution algorithm comprises: an initialization step, comprising the second type fuzzy translation in the interval, the average The value, the standard deviation and the post-identification weight are regarded as an individual and encoded, and the interval type II fuzzy neural network model includes a plurality of the individuals; a grouping step is to group these individuals according to a group threshold forming a plurality of groups; a leader confirming step is to confirm whether each of the individuals in each of the groups is a leader, and the leader represents the smallest of the predicted tool wear values of the plurality of the individuals of each of the groups; A mutation step is obtained by performing a mutational evolution relation on these individuals to a mutation vector; an exchange step is to execute an exchange relationship according to the mutation vector to obtain a test vector; and a selection step is to execute a selection relationship according to the test vector to obtain a selected individual vector, the The interval second type fuzzy neural network model generates the lateral evolution prediction tool wear value according to the individual vector operation. The tool wear prediction method using evolutionary fuzzy neural network as described in claim 7, wherein the dynamic grouping collaborative differential evolution algorithm further comprises: a leader adjustment step of arranging the individuals into p×n individual vectors, and divide the p×n individual vectors into n groups of p×1 individual vectors, and then find n vertical leaders of the n groups of p×1 individual vectors. The second type of fuzzy neural network model in this interval is based on n The longitudinal evolution prediction tool wear value is generated by the longitudinal leader operation; an update step is to compare these predicted tool wear values, the lateral evolution prediction tool wear value and the longitudinal evolution prediction tool wear value, and select the one with the smallest error to obtain updating to the best predicted tool wear value; and an iteration number judging step for judging whether an iteration number of the grouping step reaches a preset number of times; if not, re-execute the grouping step and the leader confirmation step; if so , the dynamic grouping cooperative differential evolution algorithm ends.

Images