TWI831603B - Tool remaining life prediction method based on current information and system thereof - Google Patents

Abstract

A tool remaining life prediction method based on current information is configured to predict a remaining life of a tool and includes performing a data obtaining step, a data normalizing step, a data dividing step and a model training and verifying step. The data obtaining step includes obtaining a plurality of tool processing data from a memory. The tool processing data include a current increase rate. The data normalizing step includes normalizing the tool processing data to generate a plurality of normalized tool processing data. The data dividing step includes dividing the normalized tool processing data into a training data set and a verifying data set. The model training and verifying step includes training a first and a second tool remaining life models by using all and a portion of the training data set, and verifying the first and the second tool remaining life models by using the verifying data set to generate a first and a second accuracies, and comparing the first and the second accuracies. Therefore, the tool remaining life prediction method based on current information of the present disclosure can determine an appropriate scenario model to predict the remaining life of the tool at optimum efficiency.

Description

Tool remaining life prediction method and system based on current information

The present invention relates to a tool remaining life prediction method and its system, and in particular to a tool remaining life prediction method and its system based on current information.

The machinery industry has laid the foundation for the vigorous development of the entire society and today retains a complete and strong supply chain of machine tools and precision machinery parts. However, with the improvement of people's wages and education levels, most people will think twice about the factory environment and day and night shift system. For many manufacturers that perform mechanical processing, they are currently encountering the dilemma of rising operating costs and manpower shortages. Therefore, Industry 4.0 smart manufacturing is promoted to reduce the number of employees in the factory. Unmanned factories are the overall trend in the current machinery industry, and monitoring tool wear and estimating processing quality and processing abnormalities is an extremely important part of smart manufacturing. Because unmanned factories If the processing quality is ignored, it is very easy to produce a large number of scrap products or repair parts, resulting in unnecessary waste of time and cost, and a considerable negative impact on the company's operations. To sum up, tool wear monitoring can bring three main benefits. The first benefit is to improve the quality of workpiece processing, the second benefit is to improve processing efficiency, and the third benefit is to reduce operating costs.

Tool wear is closely related to the processing quality of the workpiece. Increased tool wear will cause a significant increase in cutting resistance, resulting in a corresponding increase in the size variation range of the workpiece after processing and the surface roughness of the machined surface, ultimately affecting the processing quality. Today's machining processes often use at least multiple knives (tools for different purposes) for processing. When the tools are severely worn, it will significantly increase the risk of roughing tool breakage for the purpose of material removal. When the roughing tool breaks, it will be processed. The reserved amount will be much larger than before the tool breaks, causing the subsequent processing tools to break all together or insert the tool due to a chain reaction due to the excessive reserved amount. Severe wear or breakage of finishing tools will lead to risks such as poor machining accuracy. Therefore, severe wear or breakage of finishing tools will have a significant impact on the machine utilization rate. Technicians need to stop production, change tools, set tools, and repair workpieces. The current management and control methods for tool wear mainly refer to the long-standing rules of experience of technicians. In order to reduce risks and accidents during processing, technicians often use past experience as an indicator to develop a fixed process for processing. After the number of workpieces, the tool replacement action is carried out uniformly, or during the handover process with colleagues on the next shift, it is difficult to clearly explain the use of the tools in a quantitative way, resulting in the personnel on the next shift replacing new ones before starting processing. Knives, the above situation will virtually waste many useful knives. It can be seen that there is currently a lack of a tool remaining life prediction method and system based on current information that requires only a small amount of data training, is low-cost, cost-effective, and has a certain degree of accuracy, so relevant researchers are looking for solutions. way.

Therefore, the purpose of the present invention is to provide a method and system for predicting the remaining life of a tool based on current information, which can predict the tool life at different times through current increase rate and General Regression Neural Network (GRNN) training. Materials and remaining life models under different scenarios can use a small amount of data to train a cost-effective scenario model and accurately predict the remaining life of the tool, thus solving the conventional model building rules that require a large amount of experimental data and predictions. Must be based on personnel rules of thumb and excessive cost issues.

An embodiment of the method aspect of the present invention provides a method for predicting the remaining life of a tool based on current information, which is used to predict the remaining life of a tool and includes the following steps: a data acquisition step, and data normalization steps, a data segmentation step and a scenario model training and verification step. The data acquisition step includes driving an arithmetic processor to acquire a plurality of tool processing data from a memory, and the tool processing data includes a current increase rate. The current increase rate is a ratio of an immediate load current to a new knife's initial load current. The data normalization step includes driving the computing processor to normalize the tool processing data to generate plural normalized tool processing data, so that a size scale between the normalized tool processing data is the same. The data segmentation step includes driving the computing processor to segment the normalized tool machining data into a training data group and a verification data group. The scenario model training and verification step includes driving the computing processor to train a first tool remaining life model and a second tool remaining life model using all of the training data set and a part of the training data set, and using the verification data set to verify the first trained tool remaining life model. The tool remaining life model and the second tool remaining life model generate a first accuracy and a second accuracy, and the first accuracy and the second accuracy are compared to determine whether to use the second tool remaining life model to predict the tool remaining life. lifespan.

In this way, the tool remaining life prediction method based on current information of the present invention can predict the remaining life model of the tool under different materials and different situations through current increase rate and GRNN training, so that a cost-effective model can be trained using a small amount of data. situation model and can accurately predict the remaining life of the tool.

Other examples of the aforementioned embodiments are as follows: the aforementioned tool processing data further include a tool usage time, a current slope trend of the machining current, a processing cutting width and at least a hardness of the processing material. The current increase rate is used to determine the end of the remaining life.

Other examples of the aforementioned implementation are as follows: the aforementioned scenario model training and verification step further includes a first training and verification step, a second training and verification step and an accuracy comparison step. The first training and verification step includes driving the computing processor to use the entire training data set to train the first tool remaining life model, and using the verification data set to verify the trained first tool remaining life model to generate the first accuracy, wherein the training data set All of them are all data of plural materials. The second training and verification step includes driving the computing processor to use part of the training data set to train the second tool remaining life model, and using the verification data set to verify the trained second tool remaining life model to generate the second accuracy, wherein the training data set Part of it is all the data of one of these materials. The accuracy comparison step includes driving the computing processor to compare the first accuracy and the second accuracy to generate an accuracy comparison result, and determine whether to use the second tool remaining life model to predict the remaining life of the tool based on the accuracy comparison result. The number of hardnesses of the at least one processing material is a plurality. The hardnesses of the processing materials respectively correspond to the materials, and the hardnesses of the processing materials are different from each other.

Other examples of the aforementioned implementation are as follows: the aforementioned scenario model training and verification step further includes a first training and verification step, a second training and verification step and an accuracy comparison step. The first training and verification step includes driving the computing processor to use the entire training data set to train the first tool remaining life model, and using the verification data set to verify the trained first tool remaining life model to generate the first accuracy, wherein the training data set All of them are all data of plural materials. The second training and verification step includes driving the computing processor to use part of the training data set to train the second tool remaining life model, and using the verification data set to verify the trained second tool remaining life model to generate the second accuracy, wherein the training data set A portion of the data that is part of one of these materials is mixed with all the data of the rest of these materials. The accuracy comparison step includes driving the computing processor to compare the first accuracy and the second accuracy to generate an accuracy comparison result, and determine whether to use the second tool remaining life model to predict the remaining life of the tool based on the accuracy comparison result. The number of hardnesses of the at least one processing material is a plurality. The hardnesses of the processing materials respectively correspond to the materials, and the hardnesses of the processing materials are different from each other.

Other examples of the aforementioned implementation are as follows: the aforementioned scenario model training and verification step further includes a first training and verification step, a second training and verification step and an accuracy comparison step. The first training and verification step includes driving the computing processor to use the entire training data set to train the first tool remaining life model, and using the verification data set to verify the trained first tool remaining life model to generate the first accuracy, wherein the training data set All of them are all data of plural materials. The second training and verification step includes driving the computing processor to use part of the training data set to train the second tool remaining life model, and using the verification data set to verify the trained second tool remaining life model to generate the second accuracy, wherein the training data set Part of it is the middle and later data of one of these materials. The accuracy comparison step includes driving the computing processor to compare the first accuracy and the second accuracy to generate an accuracy comparison result, and determine whether to use the second tool remaining life model to predict the remaining life of the tool based on the accuracy comparison result. The number of hardnesses of the at least one processing material is a plurality. The hardnesses of the processing materials respectively correspond to the materials, and the hardnesses of the processing materials are different from each other.

An embodiment of the structural aspect of the present invention provides a tool remaining life prediction system based on current information, which is used to predict the remaining life of a tool and includes a memory and a computing processor. The memory stores a plurality of tool processing data. The tool processing data includes a current increase rate. The current increase rate is a ratio of an immediate load current and a new tool initial load current. The computing processor is electrically connected to the memory and receives the tool machining data. The computing processor is configured to perform an operation including the following steps: a data normalization step, a data segmentation step and a scenario model training and verification step. The data normalization step includes normalizing the tool processing data to generate plural normalized tool processing data, so that a large scale between the normalized tool processing data is the same. The data segmentation step includes segmenting the normalized tool machining data into a training data group and a verification data group. The scenario model training and verification step includes using the entire training data set and a part of the training data set to train a first tool remaining life model and a second tool remaining life model, and using the verification data set to verify the trained first tool remaining life model. and the second tool remaining life model to generate a first accuracy and a second accuracy, and compare the first accuracy and the second accuracy to determine whether to use the second tool remaining life model to predict the remaining life of the tool.

In this way, the tool remaining life prediction system based on current information of the present invention can predict the remaining life model of the tool under different materials and different situations through current increase rate and GRNN training, so that it can train an efficient model using a small amount of data. situation model and can accurately predict the remaining life of the tool.

Other examples of the aforementioned embodiments are as follows: the aforementioned tool processing data further include a tool usage time, a current slope trend of the machining current, a processing cutting width and at least a hardness of the processing material. The current increase rate is used to determine the end of the remaining life.

Other examples of the aforementioned implementation are as follows: the aforementioned scenario model training and verification step further includes a first training and verification step, a second training and verification step and an accuracy comparison step. The first training and verification step includes driving the computing processor to use the entire training data set to train the first tool remaining life model, and using the verification data set to verify the trained first tool remaining life model to generate the first accuracy, wherein the training data set All of them are all data of plural materials. The second training and verification step includes driving the computing processor to use part of the training data set to train the second tool remaining life model, and using the verification data set to verify the trained second tool remaining life model to generate the second accuracy, wherein the training data set Part of it is all the data of one of these materials. The accuracy comparison step includes driving the computing processor to compare the first accuracy and the second accuracy to generate an accuracy comparison result, and determine whether to use the second tool remaining life model to predict the remaining life of the tool based on the accuracy comparison result. The number of hardnesses of the at least one processing material is a plurality. The hardnesses of the processing materials respectively correspond to the materials, and the hardnesses of the processing materials are different from each other.

Other examples of the aforementioned implementation are as follows: the aforementioned scenario model training and verification step further includes a first training and verification step, a second training and verification step and an accuracy comparison step. The first training and verification step includes driving the computing processor to use the entire training data set to train the first tool remaining life model, and using the verification data set to verify the trained first tool remaining life model to generate the first accuracy, wherein the training data set All of them are all data of plural materials. The second training and verification step includes driving the computing processor to use part of the training data set to train the second tool remaining life model, and using the verification data set to verify the trained second tool remaining life model to generate the second accuracy, wherein the training data set A portion of the data that is part of one of these materials is mixed with all the data of the rest of these materials. The accuracy comparison step includes driving the computing processor to compare the first accuracy and the second accuracy to generate an accuracy comparison result, and determine whether to use the second tool remaining life model to predict the remaining life of the tool based on the accuracy comparison result. The number of hardnesses of the at least one processing material is a plurality. The hardnesses of the processing materials respectively correspond to the materials, and the hardnesses of the processing materials are different from each other.

Other examples of the aforementioned implementation are as follows: the aforementioned scenario model training and verification step further includes a first training and verification step, a second training and verification step and an accuracy comparison step. The first training and verification step includes driving the computing processor to use the entire training data set to train the first tool remaining life model, and using the verification data set to verify the trained first tool remaining life model to generate the first accuracy, wherein the training data set All of them are all data of plural materials. The second training and verification step includes driving the computing processor to use part of the training data set to train the second tool remaining life model, and using the verification data set to verify the trained second tool remaining life model to generate the second accuracy, wherein the training data set Part of this is one of the later data of these materials. The accuracy comparison step includes driving the computing processor to compare the first accuracy and the second accuracy to generate an accuracy comparison result, and determine whether to use the second tool remaining life model to predict the remaining life of the tool based on the accuracy comparison result. The number of hardnesses of the at least one processing material is a plurality. The hardnesses of the processing materials respectively correspond to the materials, and the hardnesses of the processing materials are different from each other.

Several embodiments of the present invention will be described below with reference to the drawings. For the sake of clarity, many practical details will be explained together in the following narrative. However, it will be understood that these practical details should not limit the invention. That is to say, in some embodiments of the present invention, these practical details are not necessary. In addition, for the sake of simplifying the drawings, some commonly used structures and components are shown in the drawings in a simple schematic manner; and repeated components may be represented by the same numbers.

In addition, when a certain component (or unit or module, etc.) is "connected" to another component in this article, it may mean that the component is directly connected to the other component, or it may mean that one component is indirectly connected to the other component. , meaning that there are other elements between the said element and another element. When it is stated that an element is "directly connected" to another element, it means that no other elements are interposed between the element and the other element. Terms such as first, second, third, etc. are only used to describe different components without limiting the components themselves. Therefore, the first component can also be renamed the second component. Moreover, the combination of components/units/circuit in this article is not a combination that is generally known, conventional or customary in this field. Whether the component/unit/circuit itself is common knowledge cannot be used to determine whether its combination relationship is easily understood in the technical field. Easily accomplished by the average person with knowledge.

Please refer to Figures 1 and 2 together. Figure 1 is a schematic flowchart illustrating the tool remaining life prediction method 100 based on current information according to the first embodiment of the present invention; and Figure 2 is a schematic flow chart showing A schematic diagram of a tool remaining life prediction system 200 based on current information according to the second embodiment of the present invention. As shown in the figure, the tool remaining life prediction method 100 based on current information is applied to the tool remaining life prediction system 200 based on current information, and is used to predict the remaining life of the tool. The current information includes the current increase rate 122.

In Figure 1, the tool remaining life prediction method 100 based on current information includes the following steps: data acquisition step S2, data normalization step S4, data segmentation step S6 and situation model training and verification step S8. Referring to FIG. 2 , the data acquisition step S2 includes driving the arithmetic processor 220 to acquire the plurality of tool machining data 120 from the memory 210 . These tool processing data 120 include a current increase rate 122, which is the ratio of the immediate load current to the initial load current of the new tool. The data normalization step S4 includes driving the arithmetic processor 220 to normalize the tool processing data 120 to generate plural normalized tool processing data 140, so that the size scales between the normalized tool processing data 140 are the same. The data segmentation step S6 includes driving the arithmetic processor 220 to segment the normalized tool machining data 140 into a training data group 162 and a verification data group 164 . The situation model training and verification step S8 includes driving the computing processor 220 to use the entire training data set 162 to train the first tool remaining life model, using a part of the training data set 162 to train the second tool remaining life model, and using the verification data set 164 to verify the training. The remaining life of the first tool after The life model and the second tool remaining life model are used to generate the first accuracy and the second accuracy, and the first accuracy and the second accuracy are compared to determine whether to use the second tool remaining life model to predict the remaining life of the tool. The situation model training and verification step S8 includes a first situation S82, a second situation S84 and a third situation S86.

In Figure 2, the tool remaining life prediction system 200 based on current information includes a memory 210 and a computing processor 220. The computing processor 220 is electrically connected to the memory 210 . Referring to FIG. 1 , the memory 210 stores tool processing data 120 . The computing processor 220 receives the tool machining data 120 and is configured to perform the data normalization step S4, the data segmentation step S6, and the scenario model training and verification step S8. Thus, the tool remaining life prediction method 100 based on current information and the tool remaining life prediction system 200 based on current information of the present invention can use the current increase rate 122 and the general regression neural network (General Regression Neural Network); GRNN) trains and predicts the remaining life model of tools in different materials and different scenarios, so that a cost-effective scenario model can be trained using a small amount of data, and the remaining life of the tool can be accurately predicted.

In the aforementioned embodiment, the tool machining data 120 may further include tool usage time, current machining current slope trend, machining cutting width and at least one machining material hardness. These data are parameters that affect the remaining life and are also inputs to the model. The output of the model is the corresponding actual remaining life. Furthermore, when the current increase rate 122 is greater than or equal to the critical life threshold rate, the tool is judged to have reached the critical life, that is, the remaining life of the tool is 0; in other words, the current increase rate 122 is used to determine the end of the remaining life. The critical life threshold ratio can be greater than 1.3 and less than 1.5, and the preferred one is 1.4. In the data segmentation step S6, the normalized tool machining data 140 can be evenly segmented into 80% of the data as the training data set 162 and 20% of the data as the verification data set 164 in the entire data range, but this is not the case in the present invention. limit. also. The memory 210 may be a random access memory (RAM) or other types of dynamic storage devices that can store information and instructions for execution by the computing processor 220, but the invention is not limited thereto. The computing processor 220 may be a processor (Processor), a microprocessor (Microprocessor), a central processing unit (CPU), a computer, a mobile device processor, a cloud processor or other electronic computing processors, but the present invention Not limited to this.

Please refer to Figures 1 to 3 together. Figure 3 is a schematic flowchart illustrating the application of the situation model training and verification step S8 of Figure 1 to the first situation S82. When the application context is the first context S82 (the amount of data is sufficient), the context model training and verification step S8 includes a first training verification step S822, a second training verification step S824 and an accuracy comparison step S826.

The first training verification step S822 includes driving the computing processor 220 to use the entire training data set 162 to train the first tool remaining life model, and to use the verification data set 164 to verify the trained first tool remaining life model to generate the first accuracy, The entire training data set 162 is all data of plural materials. Specifically, the first training verification step S822 includes steps S822a, S822b, and S822c, where step S822a is The three-material benchmark model is predicted using full data training, that is, all the training data set 162 is used to train the first tool remaining life model. Step S822b uses the K-equal cross-validation method to find the optimal smoothing parameters of the model, that is, the K-equal cross-validation method is used to adjust the model smoothing parameters to minimize the prediction error. Step S822c uses the verification data prediction accuracy as a comparison benchmark, that is, uses the verification data set 164 to verify the trained first tool remaining life model to generate the first accuracy.

The second training and verification step S824 includes driving the computing processor 220 to use part of the training data set 162 to train the second tool remaining life model, and using the verification data set 164 to verify the trained second tool remaining life model to generate the second accuracy, Part of the training data set 162 is all the data of one of these materials. Specifically, the second training verification step S824 includes steps S824a, S824b, and S824c. Step S824a is to train a remaining life prediction model for predicting a specific material, that is, using part of the training data set 162 to train the second tool remaining life model. Step S824b is the same as step S822b and will not be described again. Step S824c is to use the verification data to predict the accuracy, that is, to use the verification data set 164 to verify the trained second tool remaining life model to generate the second accuracy.

The accuracy comparison step S826 compares the prediction accuracy of the training model of the two methods (ie, the first training verification step S822 and the second training verification step S824); in other words, the accuracy comparison step S826 includes driving the computing processor 220 to compare the first accuracy and the second accuracy to generate an accuracy comparison result, and decide whether to use the second tool remaining life model to predict the remaining life of the tool based on the accuracy comparison result.

In one embodiment, the K-equal cross-validation method may be a 10-equal cross-validation method. Accuracy can be Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), or a combination thereof. The number of hardnesses of the processing materials may be a plural number, and the hardnesses of the processing materials respectively correspond to the materials, and the hardnesses of the processing materials are different from each other. The hardness of the processed materials can include HRC20, HRC45 and HRC55 (i.e. three materials), and the corresponding data can be obtained through repeated experiments, but the present invention is not limited to the above. In addition, in the accuracy comparison step S826, when the accuracy comparison result is that the difference between the first accuracy and the second accuracy is less than or equal to the preset difference, the second tool remaining life model is used to predict the remaining life of the tool. .

Please refer to Figures 1 to 4 together. Figure 4 is a schematic flowchart illustrating the application of the situation model training and verification step S8 of Figure 1 to the second situation S84. When the application context is the second context S84 (the target data volume is small and the relevant data volume is large), the context model training and verification step S8 includes a first training verification step S842, a second training verification step S844 and an accuracy comparison step S846.

The first training verification step S842 includes steps S842a, S842b, and S842c. The first training verification step S842 is the same as the first training verification step S822 in Figure 3 and will not be described again. The second training and verification step S844 includes driving the computing processor 220 to use part of the training data set 162 to train the second tool remaining life model, and using the verification data set 164 to verify the trained second tool remaining life model to generate a second accuracy. Specifically, the second training verification step S844 includes steps S844a, S844b, S844c, wherein step S844a is to reduce the target data amount and mix the relevant data to train the model, that is, use part of the training data set 162 to train the second tool remaining life model, where the part of the training data set 162 is one of these materials ( Part of the data (target data) is mixed with all the data of the rest of these materials (related data). Step S844b is the same as step S842b. Step S844c is to use the verification data to predict the accuracy, that is, to use the verification data set 164 to verify the trained second tool remaining life model to generate the second accuracy.

The accuracy comparison step S846 is to compare how much target data is needed so that the accuracy is close to the benchmark model; in other words, the accuracy comparison step S846 includes driving the computing processor 220 to compare the first accuracy and the second accuracy to generate an accuracy comparison result, and based on The accuracy comparison result determines whether to use the second tool remaining life model to predict the remaining life of the tool. In one embodiment, the "part" of one of these materials (target data) is 75%, 50% or 25% of the total. The accuracy of these three target data is compared, and the accuracy comparison result is selected to meet the requirements ( If it is greater than or equal to the preset accuracy threshold), the second tool remaining life model is trained with the minimum amount of target data.

Please refer to Figures 1 to 5 together. Figure 5 is a schematic flowchart illustrating the application of the situation model training and verification step S8 of Figure 1 to the third situation S86. When the application scenario is the third scenario S86 (the model focuses on predicting mid- to late-stage tool life), the scenario model training and verification step S8 includes a first training and verification step S862, a second training and verification step S864, and an accuracy comparison step S866.

The first training verification step S862 includes steps S862a, S862b, and S862c. The first training verification step S862 is the same as the first training verification step S822 in Figure 3 and will not be described again. The second training and verification step S864 includes driving the computing processor 220 to use part of the training data set 162 to train the second tool remaining life model, and using the verification data set 164 to verify the trained second tool remaining life model to generate a second accuracy. Specifically, the second training verification step S864 includes steps S864a, S864b, and S864c. Step S864a is a full-data training model for the middle and late stages of the tool life, that is, part of the training data set 162 is used to train the second tool remaining life model, where the training Part of the data set 162 is the later data of these materials. The above-mentioned middle and late data can be data where the remaining life of the tool is less than 50%, but the present invention is not limited to this. Step S864b is the same as step S862b. Step S864c is to use the verification data to predict the accuracy, that is, to use the verification data set 164 to verify the trained second tool remaining life model to generate the second accuracy. This eliminates factors that make model predictions inaccurate in the early and middle stages of tool life, which not only increases model accuracy but also reduces the amount of training data.

It can be seen from the above embodiments that the present invention has the following advantages: First, the remaining life of the tool can be accurately predicted through the current increase rate. Second, use GRNN to train and predict the remaining life model of tools in different materials and different situations. A cost-effective situation model can be trained using a small amount of data, and the remaining life of the tool can be accurately predicted, thereby solving common problems. Model building rules require a large amount of experimental data, predictions must be based on human experience rules, and the cost is too high.

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 modifications and modifications without departing from the spirit and scope of the present invention. Therefore, the protection of the present invention is The scope shall be determined by the appended patent application scope.

100: Tool remaining life prediction method based on current information

120: Tool processing data

122: Current increase rate

140: Normalized tool processing data

162:Training data group

164: Verification data group

200: Tool remaining life prediction system based on current information

210:Memory

220:Arithmetic processor

S2: Data acquisition steps

S4: Data normalization steps

S6: Data segmentation steps

S8: Scenario model training and verification steps

S82: First Situation

S822, S842, S862: first training and verification step

S822a,S822b,S822c,S824a,S824b,S824c,S842a,S842b,S842c,S844a,S844b,S844c,S862a,S862b, S862c, S864a, S864b, S864c: steps

S824, S844, S864: second training and verification step

S826, S846, S866: Accuracy comparison steps

S84: The second situation

S86: The third situation

Figure 1 is a schematic flow chart illustrating a tool remaining life prediction method based on current information according to the first embodiment of the present invention; Figure 2 is a schematic diagram of a tool remaining life prediction system based on current information according to the second embodiment of the present invention; Figure 3 is a schematic flowchart illustrating the application of the scenario model training and verification steps in Figure 1 to the first scenario; Figure 4 is a schematic flow chart illustrating the application of the situation model training and verification steps in Figure 1 to the second situation; and Figure 5 is a schematic flowchart illustrating the application of the scenario model training and verification steps in Figure 1 to the third scenario.

100: Tool remaining life prediction method based on current information

120: Tool processing data

122: Current increase rate

140: Normalized tool processing data

162:Training data group

164: Verification data group

S2: Data acquisition steps

S4: Data normalization steps

S6: Data segmentation steps

S8: Scenario model training and verification steps

S82: First Situation

S84: The second situation

S86: The third situation

Claims (10)

A tool remaining life prediction method based on current information is used to predict the remaining life of a tool. The tool remaining life prediction method based on current information includes the following steps: a data acquisition step, including driving an arithmetic processor Obtaining multiple tool processing data from a memory, the tool processing data includes a current increase rate, the current increase rate is a ratio of an immediate load current and a new tool initial load current; a data normalization step includes driving The arithmetic processor normalizes the tool processing data to generate plural normalized tool machining data so that a size scale between the normalized tool machining data is the same; a data segmentation step includes driving the arithmetic processor to convert the These normalized tool machining data are divided into a training data group and a verification data group; and a scenario model training verification step includes driving the computing processor to use the training data group to fully train a first tool remaining life model, using the A part of the training data set trains a second tool remaining life model, and uses the verification data set to verify the trained first tool remaining life model and the second tool remaining life model to generate a first accuracy and a second tool remaining life model. Accuracy, and comparing the first accuracy with the second accuracy to determine whether to use the second tool remaining life model to predict the remaining life of the tool. The tool remaining life prediction method based on current information as described in claim 1, wherein the tool processing data further includes a tool usage time, a current machining current slope trend, a machining cutting width and At least a certain hardness of the processed material, the current increase rate is used to determine the end of the remaining life. The tool remaining life prediction method based on current information as described in claim 2, wherein the scenario model training and verification step further includes: a first training and verification step, including all training that drives the computing processor to use the training data set the first tool remaining life model, and use the verification data set to verify the trained first tool remaining life model to generate the first accuracy, wherein all of the training data set is all data of the plurality of materials; a second The training and verification step includes driving the computing processor to use the part of the training data set to train the second tool remaining life model, and using the verification data set to verify the trained second tool remaining life model to generate the second accurate degree, wherein the portion of the training data set is all data of one of the materials; and an accuracy comparison step includes driving the computing processor to compare the first accuracy and the second accuracy to generate an accuracy and based on the accuracy comparison result, it is decided whether to use the second tool remaining life model to predict the remaining life of the tool; wherein, the number of hardnesses of the at least one processing material is a plural number, and the hardnesses of the processing materials respectively correspond to the Some materials have different hardnesses from each other. As claimed in claim 2, the tool remaining life prediction method based on current information, wherein the scenario model training and verification step further includes: A first training verification step, including driving the computing processor to use the entire training data set to train the first tool remaining life model, and using the verification data set to verify the trained first tool remaining life model to generate the third tool remaining life model. an accuracy, wherein all of the training data set is all data of the plurality of materials; a second training verification step, including driving the computing processor to use the part of the training data set to train the second tool remaining life model, and using The verification data set verifies the second tool remaining life model after training to generate the second accuracy, wherein the part of the training data set is a part of the data of one of the materials mixed with all of the rest of the materials. data; and an accuracy comparison step, including driving the computing processor to compare the first accuracy and the second accuracy to generate an accuracy comparison result, and deciding whether to use the second tool remaining based on the accuracy comparison result. The life model predicts the remaining life of the tool; wherein the number of hardnesses of the at least one processing material is a plural number, the hardnesses of the processing materials respectively correspond to the materials, and the hardnesses of the processing materials are different from each other. The tool remaining life prediction method based on current information as described in claim 2, wherein the scenario model training and verification step further includes: a first training and verification step, including all training that drives the computing processor to use the training data set the first tool remaining life model, and use the verification data set to verify the trained first tool remaining life model to generate the first accuracy, wherein all of the training data set is all of the plurality of materials part of the data; a second training and verification step, including driving the computing processor to use the part of the training data set to train the second tool remaining life model, and using the verification data set to verify the trained second tool remaining life model. and generating the second accuracy, wherein the part of the training data set is one of the mid- and late-stage data of the materials; and an accuracy comparison step includes driving the computing processor to compare the first accuracy with the second accuracy to generate an accuracy comparison result, and decide whether to use the second tool remaining life model to predict the remaining life of the tool based on the accuracy comparison result; wherein the number of hardnesses of the at least one processing material is a plural number, and the number of processing materials is a complex number. The material hardness corresponds to the materials respectively, and the hardness of the processed materials is different from each other. A tool remaining life prediction system based on current information is used to predict the remaining life of a tool. The tool remaining life prediction system based on current information includes: a memory that stores a plurality of tool processing data. The tools The processing data includes a current increase rate, which is a ratio of an immediate load current to a new tool's initial load current; and an operation processor that is electrically connected to the memory and receives the tool processing data, and the operation processor The processor is configured to perform operations including: a data normalization step, including normalizing the tool processing data to generate a plurality of normalized tool processing data such that a large or small scale between the normalized tool processing data is achieved. same; a data segmentation step, including segmenting the normalized tool machining data into a training data group and a verification data group; and a scenario model training verification step, including using all of the training data group to train a first tool remaining life model , use a part of the training data set to train a second tool remaining life model, and use the verification data set to verify the trained first tool remaining life model and the second tool remaining life model to generate a first accuracy and a second accuracy, and comparing the first accuracy with the second accuracy to determine whether to use the second tool remaining life model to predict the remaining life of the tool. A tool remaining life prediction system based on current information as described in claim 6, wherein the tool processing data further includes a tool usage time, a current machining current slope trend, a processing cutting width and at least a processing material hardness , the current increase rate is used to determine the end of the remaining life. As claimed in claim 7, the tool remaining life prediction system based on current information, wherein the scenario model training and verification step further includes: a first training and verification step, including all training that drives the computing processor to use the training data set the first tool remaining life model, and use the verification data set to verify the trained first tool remaining life model to generate the first accuracy, wherein all of the training data set is all data of the plurality of materials; a second The training verification step includes driving the computing processor to use the training The part of the training data set is used to train the second tool remaining life model, and the verification data set is used to verify the trained second tool remaining life model to generate the second accuracy, wherein the part of the training data set is All data of one of these materials; and an accuracy comparison step, including driving the computing processor to compare the first accuracy and the second accuracy to generate an accuracy comparison result, and making a decision based on the accuracy comparison result Whether to use the second tool remaining life model to predict the remaining life of the tool; wherein the number of hardnesses of the at least one processing material is a plural number, the hardnesses of the processing materials respectively correspond to the materials, and the hardnesses of the processing materials are different from each other. As claimed in claim 7, the tool remaining life prediction system based on current information, wherein the scenario model training and verification step further includes: a first training and verification step, including all training that drives the computing processor to use the training data set the first tool remaining life model, and use the verification data set to verify the trained first tool remaining life model to generate the first accuracy, wherein all of the training data set is all data of the plurality of materials; a second The training and verification step includes driving the computing processor to use the part of the training data set to train the second tool remaining life model, and using the verification data set to verify the trained second tool remaining life model to generate the second accurate degree, wherein the portion of the training data set is a portion of data from one of the materials mixed with all data from the remainder of the materials; and An accuracy comparison step includes driving the computing processor to compare the first accuracy and the second accuracy to generate an accuracy comparison result, and determine whether to use the second tool remaining life model prediction based on the accuracy comparison result. The remaining life of the tool; wherein the number of hardnesses of the at least one processing material is a plurality, the hardnesses of the processing materials respectively correspond to the materials, and the hardnesses of the processing materials are different from each other. As claimed in claim 7, the tool remaining life prediction system based on current information, wherein the scenario model training and verification step further includes: a first training and verification step, including all training that drives the computing processor to use the training data set the first tool remaining life model, and use the verification data set to verify the trained first tool remaining life model to generate the first accuracy, wherein all of the training data set is all data of the plurality of materials; a second The training and verification step includes driving the computing processor to use the part of the training data set to train the second tool remaining life model, and using the verification data set to verify the trained second tool remaining life model to generate the second accurate degree, wherein the part of the training data set is one of the mid- and late-stage data of the materials; and an accuracy comparison step includes driving the computing processor to compare the first accuracy and the second accuracy to generate an accuracy Compare the results, and decide whether to use the second tool remaining life model to predict the remaining life of the tool based on the accuracy comparison result; wherein the number of hardnesses of the at least one processing material is a complex number, and the number of processing materials is a complex number. The material hardness corresponds to the materials respectively, and the hardness of the processed materials is different from each other.

Images