TWI792011B - Adaptive model adjustment system of tool life prediction model and method thereof - Google Patents

Abstract

The present invention provides an adaptive model adjustment system of tool life prediction model and method thereof. A database stores an original tool life model, key factors, and an original weight coefficient therein. With a calculation by the key factor analysis module according to the key factors, a target tool life model is generated. With a calculation by the weight assessment module according to the original weight coefficient, a target weight coefficient is acquired. The model adjustment module accordingly generates the adjusted tool life model, such that the life prediction module predicts the tool life based on the adjusted tool life model, with an adaptive module updating the original tool life model and the original weight coefficient. The adjustment of the tool life model is regularly generated after a predetermined tool processing times for predicting the tool life, achieving the adaptive model adjustment.

Description

Tool Life Prediction Model Fast Adaptive Modification System and Method

The invention relates to a tool life prediction technology, in particular to a tool life prediction model fast self-adaptive mold repair system and method.

The conventional tool life prediction method is to build a tool life model during the process of processing a tool from a new one to wear out, to judge the available hours of the tool life, so as to warn when the life of the same tool is approaching. To remind the user to replace the tool with a new one at that time.

When this conventional tool life model is established, it is necessary to collect a complete signal of a tool that is continuously processed from a new state to when it must be replaced, and such tool life prediction or warning can usually only be carried out under the same processing conditions. Changes in processing conditions, such as the replacement of different specifications of tools, replacement of different models of processing machines, or changes in processing conditions (such as changing processing speed or workpiece material), must re-establish a new tool life model. When introducing practical applications, it is inconvenient because of time and cost.

In order to solve the above problems, the present invention provides a tool life prediction model rapid self-adaptive mold repair system, which can maintain the existing tool life through adaptive mold repair technology without re-establishing a new tool life model under changing machining conditions. Tool life model continuation prediction.

One embodiment of the present invention provides a tool life prediction model rapid self-adaptive repair system, which includes a database, a key factor analysis module, a weight estimation module, a model correction module, a life prediction module and an adaptive module. The database stores a pre-established initial tool life model of a tool, and stores a key factor and a weight coefficient before correction. The key factor is selected from at least one signal feature extracted from the tool due to machining wear to a total number of machining tools The key factor analysis module is electrically connected to the database, and the key factor analysis module generates a tool life model to be repaired through calculation according to the key factors; the weight estimation module is electrically connected to the database, and the weight estimation module is based on the weight coefficient before correction A corrected weight coefficient is obtained through calculation; the model correction module is electrically connected to the key factor analysis module and the weight estimation module, and the model correction module uses the product of the corrected weight coefficient and the tool life model to be repaired to generate a Correct the tool life model; the life prediction module is electrically connected to the model correction module, and the life prediction module uses the corrected tool life model to predict the tool life of the tool after a predetermined number of cuts; the self-adaptive module is electrically connected to the database, the key Factor analysis module, weight estimation module, model correction module and life prediction module, after the tool life prediction, the self-adaptive module updates the initial tool life model and its corresponding key factors, and updates the Weight coefficient; at every other number of machining knives, the corrected tool life model is generated by the key factor analysis module, weight estimation module, and model correction module as described above, and the tool life is predicted by the life prediction module , according to this cycle to quickly and adaptively modify the tool life prediction model.

One embodiment of the present invention provides a method for rapid and adaptive repair of a tool life prediction model, which includes the steps of storing model data, the steps of model repair and tool life prediction, and the step of self-adaptive cycle model repair. In the step of storing model data, an initial tool life model established in advance of a tool is stored, and a key factor corresponding to the initial tool life model and a weight coefficient before correction are stored. The key factor is selected from the tool due to machining wear to a At least one signal feature extracted from the total number of machining tools; in the steps of mold repair and tool life prediction, a tool life model to be repaired is generated through calculation based on key factors, and a correction is obtained through calculation based on the weight coefficient before correction The final weight coefficient, and the product of the corrected weight coefficient and the tool life model to be repaired is used to generate a corrected tool life model, and the corrected tool life model is used to predict the tool life of the tool after a predetermined number of cuts; in the adaptive cycle In the mold repairing step, after the mold repairing and tool life prediction steps, the initial tool life model and its corresponding key factors are updated, and the weight coefficient before correction is updated; The modified tool life model is used to predict the tool life, and the cycle is used to quickly and adaptively modify the tool life prediction model.

In this way, when the initial tool life model is established and stored in the database, even if the processing conditions are changed during the use of the tool, such as the replacement of different specifications of tools, the replacement of different models of processing machines, or changes in processing conditions, etc. When conditions are met, the present invention generates a corrected tool life model every preset number of machining tools to predict the tool life, and then circulates to achieve fast self-adaptive model repair of the tool life prediction model, thereby avoiding the need for remodeling. It is inconvenient to consume time and cost because of the model, but it is conducive to the introduction of practical applications.

In order to illustrate the central idea of the present invention expressed in the column of the above-mentioned summary of the invention, it is expressed in specific embodiments. Various objects in the embodiments are drawn according to proportions, sizes, deformations or displacements suitable for illustration, rather than drawn according to the proportions of actual components, which are described first.

Referring to FIG. 1 to FIG. 5 , the present invention provides a system 100 and a method 200 for fast adaptive mold repair of a tool life prediction model. The system 100 includes a database 10, a key factor analysis module 20, a weight estimation module 30, a model correction module 40, a life prediction module 50, and an adaptive module 60, and in In this embodiment, a model building module 70 is included, wherein the database 10 stores an initial tool life model 11, and the database 10 also stores a key factor 12 and a weight coefficient 13 before correction; the key factor analysis module 20, The weight estimation module 30 and the model building module 70 are respectively electrically connected to the database 10; the model correction module 40 is electrically connected to the key factor analysis module 20 and the weight estimation module 30; the life prediction module 50 is electrically connected to the model correction The module 40 and the adaptive module 60 are electrically connected to the database 10 , the key factor analysis module 20 , the weight estimation module 30 , the model modification module 40 , and the life prediction module 50 .

The method 200 for rapid adaptive mold repair of the tool life prediction model in this embodiment includes a model building step 201, a model data storage step 202, a mold repair and tool life prediction step 203, and an adaptive loop mold repair step 204, in:

In the model building step 201, the model building module 70 extracts at least one signal feature 71 and applies a dimensionality reduction algorithm to find a number of key factors 12 during the process of machining a tool to a total number of tools. And through Logis regression algorithm to establish an initial tool life model 11 . The cutting tool in this embodiment is a milling cutter as an example, and the applied processing machine is a central processing machine.

As shown in Figure 3, it is the signal features 71 extracted for the milling cutters with diameters of ψ6 and ψ10 respectively, where the abscissa is classified according to the feature attributes, and the ordinate is the score corresponding to each feature, and then the reduction is applied. Dimension algorithm to find out some key factors12. The established initial tool life model 11 can be visualized as shown in Figure 4. As shown in Figure 4, after the tool reaches the 982nd number of cuts, the tool life begins to decline significantly, and when the machining reaches the 982nd When the number of knives is 1182, the tool life is reduced to 0, which means that the tool has been worn out to the extent that it needs to be replaced, and 1182 knives is the total number of knives processed by this tool.

Next, in the step 202 of storing model data, the initial tool life model 11 is stored in the database 10, and the several key factors 12 and the weight coefficients before correction 13 are stored in the database 10. The several key factors 12 are selected At least one signal feature 71 extracted from the tool wear due to machining to the total number of machining tools. Preferably, the signal feature 71 is a vibration signal measured by an accelerometer during machining of the tool, including a time-domain feature and/or a frequency-domain feature, and the time-domain feature includes root mean square (RMS), average Value (Mean), standard deviation (STD), kurtosis (Kurtosis), skewness (Skewness); the frequency domain features include rotational speed frequency, 1 times blade frequency, 2 times blade frequency and/or 3 times blade frequency.

Next, in the mold repairing and tool life prediction step 203 , the key factor analysis module 20 generates a tool life model 21 to be repaired through calculation according to the several key factors 12 . In this embodiment, in the model building step 201 , the initial tool life model 11 is established with the aforementioned ψ6 tool, and the key factor 12 and weight coefficient 13 selected for the tool corresponding to ψ10 are stored in the database 10 . Assume that due to processing needs, the tool must be replaced with a ψ6 tool for milling of the workpiece. When the 40th tool is processed, the vibration signal picked up by the accelerometer is divided into the signal feature 71 of the first 20 tools and the signal of the last 20 tools. feature 71, and in the mold repairing and tool life prediction step 203, the corresponding key factor 12 is analyzed by the key factor analysis module 20 with the signal features 71 of the first 20 knives and the last 20 knives, and the key factor 12 The tool life model 21 to be repaired is generated through a Logis regression algorithm.

As above, in the mold repairing and tool life prediction step 203 , after the tool life model 21 to be repaired is generated, the weight estimation module 30 calculates the weight coefficient 13 before correction to obtain a weight coefficient 31 after correction. In this embodiment, the weight estimation module 30 applies a genetic algorithm to generate multiple weight coefficient values, and then uses steps such as selection, replication, mating, and mutation to match the required weight coefficient values to obtain the desired weight coefficient values. The modified weight coefficient 31 is described above. In the model repairing and tool life prediction step 203, the product of the corrected weight coefficient 31 and the tool life model 21 to be repaired is used to generate a corrected tool life model 41, whereby the corrected tool life model 41 predicts the tool life in a Tool life after the preset number of knives, the preset number of knives mentioned here is set to 20 knives in this embodiment.

Next, in the self-adaptive loop model repairing step 204 , after the model repairing and tool life prediction step 203 , the initial tool life model 11 and its corresponding key factor 12 are updated, and the weight coefficient 13 before correction is updated. When every other number of machining knives (preset to every 20 knives in this embodiment), the corrected tool life model 41 is generated as described above and the tool life is predicted, and the tool life prediction model is quickly and adaptively corrected in a loop . In this embodiment, the signal features 71 extracted from the 0th tool to the 40th tool are completed, and the first corrected tool life model 41 is generated as in the aforementioned model repair and tool life prediction step 203 to predict the 40th tool Up to the tool life of the 60th tool, and carry out the self-adaptive cycle model repair step 204, and when the number of machining tools from the 40th tool to the 60th tool is completed, the second corrected tool life model is generated using the extracted signal characteristics 71 41, to predict the tool life of the 60th to 80th tool, and complete the extracted signal characteristics with the number of processing tools from the 40th to the 60th tool 71. Compare the first corrected tool life model 41 to get Know the accuracy rate of tool life prediction, and so on.

In this embodiment, the accuracy rate of the tool life prediction is gradually corrected by the reinforcement learning algorithm, and as shown in Figure 5, when the number of machining tools of the tool reaches the 140th tool, the 6th corrected tool has been generated Life model 41, and when reaching the 160th machining tool number, the signal feature 71 comparison completed by the 6th corrected tool life model 41 and the 140th to 160th knife machining tool number is completed, The accuracy of the verifiable prediction can reach 88%. Similarly, the 8th revised tool life model 41 is compared with the signal features 71 extracted from the number of machining tools from the 180th to the 200th tool, which can verify the prediction. The accuracy rate reached 89.16%; the 10th revised tool life model 41 was compared with the signal features 71 extracted from the 220th to 240th tool, and the verifiable prediction accuracy reached 97.69% , so the higher the number of tool life model corrections, the more accurate the prediction accuracy.

From the above description, it is not difficult to find that the feature of the present invention is that when the initial tool life model 11 is established and stored in the database 10, even if the processing conditions are changed during the use of the tool, the present invention will change the tool life at every preset tool life. The corrected tool life model 41 is generated every few times to predict the tool life, and the cycle is used to achieve the fast adaptive model modification of the tool life prediction model. By selecting the key factor 12, it is possible to avoid over-fitting caused by subtle conditions. The impact of changes leads to the distortion of the tool life prediction model, so as to avoid the inconvenience of time and cost due to remodeling, and is conducive to the introduction of practical applications to meet the diverse processing needs of the processing industry.

Furthermore, in the present embodiment, the tool life model 21 to be repaired is generated through the Logis regression algorithm, and the corrected weight coefficient 31 is obtained through a genetic algorithm to generate a corrected tool life model 41 , can cope with different processing conditions, such as the replacement of different specifications of tools, replacement of different models of processing machines, or changes in processing conditions such as processing conditions, can still obtain accurate tool life prediction results, compared with other algorithms, such as Tool life prediction through Taylor life formula, Logis regression algorithm combined with dynamic neural network algorithm can only be applied to the same processing conditions after analysis, and a new tool life model must be re-established when the processing conditions change. In this embodiment, the existing tool life prediction model can be extended to cope with different processing conditions.

The above-mentioned embodiments are only used to illustrate the present invention, and are not intended to limit the scope of the present invention. All modifications or changes that do not violate the spirit of the present invention belong to the intended protection category of the present invention.

100: system 10: Database 11: Initial tool life model 12: Key Factors 13: Weight coefficient before correction 20: Key factor analysis module 21: Tool life model to be repaired 30: Weight Estimation Module 31: Corrected weight coefficient 40:Model correction module 41: Corrected tool life model 50:Life Prediction Module 60: Adaptive Module 70:Model building module 71:Signal Characteristics 200: method 201: Model building steps 202: Step of saving model data 203: Steps for mold repair and tool life prediction 204: Steps of self-adaptive loop mold repair

Fig. 1 is a block diagram of a tool life prediction model fast self-adaptive mold repairing system according to an embodiment of the present invention. Fig. 2 is a flow chart of a fast adaptive mold repair method for a tool life prediction model according to an embodiment of the present invention. Fig. 3 is a schematic diagram of key factor analysis of the embodiment of the present invention. Fig. 4 is a schematic diagram of a tool life model of an embodiment of the present invention. Fig. 5 is a schematic diagram of tool life model correction and prediction according to an embodiment of the present invention.

200: method

201: Model building steps

202: Step of saving model data

203: Steps for mold repair and tool life prediction

204: Steps of self-adaptive loop mold repair

Claims (11)

A tool life prediction model fast self-adaptive mold repair system, which includes: a database, which stores a pre-established initial tool life model of a tool, and stores a key factor and a weight coefficient before correction, the key factor is selected At least one signal feature extracted from the tool wear due to machining to a total number of machining tools; a key factor analysis module, which is electrically connected to the database, and the key factor analysis module is generated by calculation according to the key factor A tool life model to be repaired; a weight estimation module, which is electrically connected to the database, the weight estimation module applies a genetic algorithm, and calculates a weight coefficient after correction according to the weight coefficient before correction; a A model correction module, which is electrically connected to the key factor analysis module and the weight estimation module, and the model correction module uses the product of the corrected weight coefficient and the tool life model to be repaired to generate a corrected tool life model; a life prediction module, which is electrically connected to the model correction module, and the life prediction module uses the corrected tool life model to predict the tool life of the tool after a predetermined number of cuts; and an adaptive module, It is electrically connected to the database, the key factor analysis module, the weight estimation module, the model modification module, and the life prediction module. After the tool life prediction, the self-adaptive module updates the initial The tool life model and its corresponding key factors are updated, and the weight coefficient before correction is updated; at every other number of machining tools, the key factor analysis module, the weight estimation module, and the model correction module are as described above The corrected tool life model is generated, and the tool life is predicted by the life prediction module, and the cycle is repeated to quickly and adaptively correct the tool life prediction model. The tool life prediction model fast self-adaptive mold repair system as described in claim 1, further includes a model building module, the model building module is from the tool processing to the total machining tool In the process of counting, the key factor is found by extracting the at least one signal feature and applying a dimension reduction algorithm, and the initial tool life model is established by a Logis regression algorithm. The tool life prediction model rapid self-adaptive mold repairing system as described in Claim 2, wherein the key factor analysis module applies a Logis regression algorithm to generate the tool life model to be repaired according to the key factor calculation. The tool life prediction model fast self-adaptive mold repairing system as described in Claim 1, wherein the signal feature is a vibration signal measured by an accelerometer when the tool is being processed, including a time-domain feature and/or a frequency-domain feature. The tool life prediction model fast adaptive mold repairing system as described in Claim 4, wherein the time domain features include root mean square, average value, standard deviation, kurtosis and/or skewness; the frequency domain features include Speed frequency, 1 times blade frequency, 2 times blade frequency and/or 3 times blade frequency. A tool life prediction model fast self-adaptive model repair method, which includes the following steps: store model data step: store a tool through the pre-established initial tool life model, and store a key factor and a weight coefficient before correction, the key factor It is selected from at least one signal feature extracted from the tool due to machining wear to a total number of machining tools; model repair and tool life prediction steps: generate a tool life model to be repaired through calculation based on the key factors, and apply a genetic algorithm method to generate a plurality of weight coefficient values, and then use selection, replication, mating, and mutation to match the required weight coefficient values to obtain a modified weight coefficient, and use the modified weight coefficient and the service life of the tool to be repaired multiplying the models to generate a corrected tool life model, using the corrected tool life model to predict the tool life of the tool after a predetermined number of cuts; and an adaptive loop model repair step: said model repair and tool life prediction steps After that, update the initial tool life The life model and its corresponding key factors are updated, and the weight coefficient before correction is updated; at every other number of machining tools, the corrected tool life model is generated as described above and the tool life is predicted, and the cycle is repeated to quickly Adaptive modification of tool life prediction model. The tool life prediction model fast self-adaptive mold repair method as described in claim item 6, which further includes a model building step, the model building step is before the storing model data step, and the model building step is after the tool is processed to the In the process of processing the total number of tools, the key factor is found by extracting the at least one signal feature and applying a dimension reduction algorithm, and the initial tool life model is established through a Logis regression algorithm. According to claim 7, the tool life prediction model rapid self-adaptive model modification method, wherein the signal feature is a vibration signal measured by an accelerometer when the tool is being processed, including a time-domain feature and/or a frequency-domain feature. The tool life prediction model fast self-adaptive model modification method as described in Claim 8, wherein the time domain features include root mean square, average value, standard deviation, kurtosis and/or skewness; the frequency domain features include Speed frequency, 1 times blade frequency, 2 times blade frequency and/or 3 times blade frequency. The tool life prediction model rapid self-adaptive mold repair method as described in claim item 7, wherein, in the mold repair and tool life prediction steps, a Logis regression algorithm is applied, and the key factor calculation is used to generate the waiting time Modify the tool life model. The method for fast adaptive mold repairing of the tool life prediction model according to claim 6, wherein the number of machining knives is 20 knives; the predetermined number of knives is 20 knives.

Images