Learning device, prediction device, diagnosis device, diagnosis system, model generation method, prediction method, diagnosis method, and program

Abstract

A learning device (10) is provided with: a simulation data acquisition unit (102) that acquires first simulation data by performing a machining simulation performed by a first machining program; an actual machining data acquisition unit (103) that performs actual machining performed by the first machining program and acquires first actual machining data; and a model generation unit (104) that learns a relationship between the first simulation data and the first actual machining data, and generates second simulation data for use in performing a machining simulation performed by a second machining program. And a trained model for predicting second actual machining data obtained by performing actual machining by the second machining program.

Description

Technical Field

[0001] The present invention relates to a learning device, a prediction device, a diagnostic device, a diagnostic system, a model generation method, a prediction method, a diagnostic method, and a program. Background Art

[0002] In the field of machine tools, based on a machining program, the positional relationship between a tool mounted on the machine tool and a workpiece (workpiece to be machined) is controlled, and machining is performed to cut the workpiece into a desired shape using the tool. In machining, when the tool is not in a specified shape (for example, when the wear of the tool progresses), it may not be possible to machine the workpiece into the desired shape. In recent years, various attempts have been made to diagnose the machining state of machine tools in order to achieve desired machining.

[0003] As a technology related to the above situation, Patent Document 1 discloses a machining state diagnostic device that learns the relationship between estimated machining state information obtained by virtually machining a workpiece based on control information and actual machining state information obtained by actually machining based on the control information. The machining state diagnostic device compares the machining state information obtained by machining based on specified control information with the actual machining state information derived from the estimated machining state information estimated from the corresponding control information and the learned relationship, and can diagnose whether the machining state is abnormal.

[0004] Patent Document 1: Japanese Unexamined Patent Application Publication No. 2021-026598 Summary of the Invention

[0005] However, the machining state diagnostic device described in Patent Document 1 learns the relationship between estimated machining state information including multiple machining processes such as straight machining, groove machining, hole machining, and corner R machining, and there is a large error between the machining state information obtained by machining based on specified control information and the actual machining state information derived from the estimated machining state information estimated from the corresponding control information and the learned relationship. In addition, since the machining state is diagnosed by comparing the machining state information with a large error and the actual machining state information, the diagnosis result of the machining state is also unreliable.

[0006] An object of the present invention is to provide a diagnostic device that accurately estimates machining state information and highly reliably diagnoses whether the machining state is abnormal in view of the above situation.

[0007] To achieve the above object, the learning device according to the present invention includes:

[0008] A simulation data acquisition unit that acquires first simulation data through a machining simulation performed by a first machining program;

[0009] An actual machining data acquisition unit that acquires first actual machining data through actual machining performed by the first machining program; and

[0010] A model generation unit that learns the relationship between the first simulation data and the first actual machining data, and generates a trained model for predicting the second actual machining data obtained by performing actual machining through a second machining program based on the second simulation data obtained by performing a machining simulation through the second machining program.

[0011] Advantages of the Invention

[0012] According to the present invention, it is possible to accurately predict actual machining data and highly reliably diagnose whether machining is abnormal. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 FIG. is a diagram showing the overall structure of a machining system according to Embodiment 1 of the present invention.

[0014] Figure 2 FIG. is a diagram for explaining the functional structure of a learning device according to Embodiment 1 of the present invention.

[0015] Figure 3 FIG. is a diagram for explaining the functional structure of a prediction device according to Embodiment 1 of the present invention.

[0016] Figure 4 FIG. is a diagram for explaining the functional structure of a diagnostic device according to Embodiment 1 of the present invention.

[0017] Figure 5 FIG. is a diagram for explaining the functional structure of a terminal device according to Embodiment 1 of the present invention.

[0018] Figure 6 FIG. is a diagram for explaining an example of the hardware structure of a learning device, a prediction device, a diagnostic device, and a terminal device according to Embodiment 1 of the present invention.

[0019] Figure 7 FIG. is a flowchart for explaining an example of the operation of model generation performed by the learning device according to Embodiment 1 of the present invention.

[0020] Figure 8 FIG. is a flowchart for explaining an example of the operation of predicting actual machining data performed by the prediction device according to Embodiment 1 of the present invention.

[0021] Figure 9 This is a flowchart for explaining an example of the operation of abnormal diagnosis of a cutting tool performed by the diagnostic device according to Embodiment 2 of the present invention. Detailed implementation mode

[0022] Next, embodiments of applying the learning device, prediction device, diagnostic device, and diagnostic system according to the present invention to a machining system will be described with reference to the accompanying drawings. In each of the drawings, the same or equivalent parts are denoted by the same reference numerals.

[0023] (Embodiment 1)

[0024] While referring to Figure 1 , the machining system 1 according to Embodiment 1 will be described.

[0025] The machining system 1 includes a learning device 10, a prediction device 20, a diagnostic device 30, a sensor 31, a machine tool 40, a cutting tool 41, a CNC (Computer Numerical Controller) 42, and a terminal device 50. The machining system 1 is, for example, a machining system introduced into the production site in a factory.

[0026] As described later, in the machining system 1, each stage of a machining simulation stage, an actual machining stage, a learning stage, a prediction stage, and a diagnostic stage is executed. In addition, the term "actual machining" is used to emphasize the difference from "machining simulation" representing virtual machining.

[0027] Next, the functions of each device will be briefly described. The detailed content will be described later.

[0028] The learning device 10 generates a trained model required for the prediction device 20 to predict actual machining data based on simulation data obtained by performing machining simulation through a machining program and actual machining data obtained by performing actual machining through a machining program. The simulation data is data such as axis positions, cutting tool axial feed amounts, cutting tool radial feed amounts, and cutting volumes obtained in time series in the machining simulation. The actual machining data is data such as axis positions and torque axis speeds obtained in time series in the actual machining. The learning device 10 is implemented, for example, by a personal computer, a server, or the like. The learning device 10 is an example of the learning device according to the present invention.

[0029] The prediction device 20 predicts actual machining data based on the trained model generated by the learning device 10 according to the simulation data. The prediction device 20 is implemented, for example, by a personal computer, a server, or the like. The prediction device 20 is an example of the prediction device according to the present invention.

[0030] The diagnostic device 30 compares the actual machining data obtained from the sensor 31 during actual machining with the actual machining data (predicted data) predicted by the prediction device 20, and diagnoses whether there is an abnormality during machining. The sensor 31 is various sensors provided on the machine tool 40 to detect the state of the machine tool 40. Sequential data such as the axis position and torque during actual machining are obtained by the sensor 31. The diagnostic device 30 is implemented by, for example, a PLC (Programmable Logic Controller), a personal computer, or the like. The diagnostic device 30 is an example of the diagnostic device according to the present invention.

[0031] The machine tool 40 operates based on the control of the CNC 42, and controls the positional relationship between the tool 41 mounted on the spindle of the machine tool 40 and the workpiece (workpiece to be machined) fixed on the worktable or the turning spindle of the machine tool 40. Further, the spindle or the turning spindle rotates, whereby the tool 41 or the workpiece rotates, and the tool 41 contacts the workpiece, whereby a cutting process of cutting a part of the workpiece by the tool 41 is performed.

[0032] The terminal device 50 is, for example, a personal computer for FA (Factory Automation). The user of the terminal device 50 is, for example, an operator in the factory. The terminal device 50 is equipped with CAM (Computer Aided Manufacturing), and generates a machining program for achieving desired machining through the operation of the user. The generated machining program is used to obtain simulation data generated by machining simulation or actual machining data generated by actual machining.

[0033] In the above description, various data are appropriately exchanged between multiple devices. The exchange of data can be performed through communication between devices or through exchange via a removable medium. Hereinafter, for the sake of simplicity of description, it is assumed that each device is connected through the Figure 1 shown communication line.

[0034] Next, each of the above machining simulation stage, actual machining stage, learning stage, prediction stage, and diagnosis stage will be briefly described.

[0035] (1) Machining simulation stage

[0036] In the machining simulation stage, the machining program is input into the terminal device 50, and the terminal device 50 performs machining simulation implemented by the input machining program to obtain simulation data. In the machining simulation, like the control information obtained by the execution of the machining program, the positional relationship between the tool and the workpiece is controlled in the virtual space, and the area passed by the tool is removed from the workpiece, thereby evaluating the shape of the workpiece. The simulation data is data such as axis position, tool axial feed amount, tool radial feed amount, and cutting volume obtained in time series in the machining simulation. The simulation data is used to generate a trained model in the learning stage described later.

[0037] In addition, the machining processes performed by the machining program include multiple machining processes such as straight machining, groove machining, and hole machining, but there is often no description in the machining program for identifying each machining process. In this machining simulation stage, each machining process is identified based on the obtained simulation data, and thereby the simulation data is differentiated for each machining process. Specifically, the change in the time series of the simulation data is captured, and the simulation data is differentiated for each machining process. For example, the machining process can be identified as changing to groove machining through the point where the tool radial feed amount changes to a value close to the tool diameter. For example, the machining process can be identified as changing to hole machining through the point where the tool changes to a drill bit. In addition, based on the information of the machining process identified in this machining simulation stage, in the actual machining stage described later, the actual machining data is differentiated for each machining process.

[0038] In addition, in this machining simulation stage, the machining program executed during machining simulation is different from the machining program executed during the prediction of actual machining data in the prediction stage described later. For example, in this machining simulation stage, the machining program that is the object of machining simulation is machining program X, and the machining program that is the object of prediction of actual machining data in the prediction stage is machining program Y. Hereinafter, for the sake of distinction, the machining program in this machining simulation stage is referred to as the "first machining program", and the simulation data is referred to as the "first simulation data".

[0039] (2) Actual Machining Stage

[0040] In the actual machining stage, the first machining program is input into the machine tool 40, and actual machining is performed by the machine tool 40. The diagnostic device 30 obtains the data obtained from the sensor 31 during actual machining as actual machining data. The actual machining data is data such as axis position and torque obtained in time series during actual machining. The actual machining data is used to generate a trained model in the learning stage described later.

[0041] In addition, based on the information of the machining processes identified in the above (1) machining simulation stage, the actual machining data is differentiated for each machining process. Specifically, the phase of the time series of the simulation data and the actual machining data can be made consistent, and in the same manner as the simulation data, the actual machining data is differentiated for each machining process. Hereinafter, for the purpose of differentiation, the actual machining data obtained through this actual machining stage is referred to as "first actual machining data".

[0042] In addition, in this actual machining stage, it is preferable that the cutting tool 41 of the machine tool 40 does not deteriorate. The first actual machining data obtained through this actual machining stage is used to generate a trained model in the learning stage described later. If the cutting tool 41 deteriorates, the relationship between the first simulation data and the first actual machining data will be learned through the first actual machining data obtained by cutting with the deteriorated cutting tool, so it is impossible to diagnose whether the machining is abnormal with high reliability.

[0043] (3) Learning stage

[0044] In the learning stage, the first simulation data for each machining process obtained through the above (1) machining simulation stage and the first actual machining data for each machining process obtained through the above (2) actual machining stage are input to the learning device 10, and the learning device 10 generates a trained model obtained by correlating the input first simulation data and first actual machining data. The generated trained model is used to predict the actual machining data based on the simulation data in the prediction stage described later. The "simulation data" mentioned here refers to the simulation data not used for the generation of the trained model. If this trained model is used, it is possible to predict the actual machining data based on the simulation data without performing actual machining.

[0045] In order to improve the accuracy of the prediction performed through the prediction stage described later, that is, in order to generate a trained model for performing high-precision prediction, learning can be performed through a plurality of first simulation data and first actual machining data.

[0046] (4) Prediction stage

[0047] In the prediction stage, in the same manner as in the (1) machining simulation stage, machining simulation is performed through a machining program different from the first machining program by the terminal device 50 to obtain simulation data. Hereinafter, for the purpose of differentiation, the machining program in this prediction stage is referred to as "second machining program", and the simulation data is referred to as "second simulation data".

[0048] Next, the second simulation data for each processing step obtained by the terminal device 50 and the trained model generated by the above-mentioned (3) learning stage are input to the prediction device 20, and the prediction device 20 predicts the actual processing data. The predicted actual processing data is used for diagnosis of whether the processing is abnormal in the diagnosis stage described later. In the following, for the purpose of distinction, the predicted actual processing data in the prediction stage is referred to as "second actual processing data (prediction data)". In addition, the actual processing data obtained by the actual processing implemented by the second processing program in the diagnosis stage described later is referred to as "second actual processing data".

[0049] (5) Diagnosis stage

[0050] In the diagnosis phase, in the same manner as in the (2) actual machining phase, actual machining performed by the second machining program is performed by the diagnosis device 30 to obtain second actual machining data. Next, the second actual machining data (prediction data) predicted in the above-mentioned (4) prediction phase is obtained by the diagnosis device 30. The diagnosis device 30 compares the second actual machining data with the second actual machining data (prediction data) to diagnose whether or not there is an abnormality in machining.

[0051] Through the above-mentioned stages (1) to (5), according to the processing system 1 involved in the first embodiment, the second actual processing data (prediction data) for each processing step can be predicted based on the second simulation data for each processing step without performing actual processing. In addition, according to the processing system 1 involved in the first embodiment, the second actual processing data and the second actual processing data (prediction data) can be compared to diagnose whether there is an abnormality in the processing.

[0052] Next, refer to Figure 2 , the functional configuration of the learning device 10 will be described. The learning device 10 includes a communication unit 101 , a simulation data acquisition unit 102 , an actual processing data acquisition unit 103 , a model generation unit 104 , and a storage unit 105 .

[0053] The communication unit 101 communicates with an external device and transmits and receives various data as necessary. The communication unit 101 is realized by, for example, a network interface.

[0054] The simulation data acquisition unit 102 acquires the first simulation data for each machining process generated by the terminal device 50 in the (1) machining simulation phase. The simulation data acquisition unit 102 acquires the first simulation data generated by the terminal device 50, for example, by communication via the communication unit 101. The simulation data acquisition unit 102 is an example of a simulation data acquisition unit according to the present invention.

[0055] The actual processing data acquisition unit 103 acquires the first actual processing data for each processing step obtained by the diagnostic device 30 in the (2) actual processing stage. The actual processing data acquisition unit 103 acquires the first actual processing data obtained by the diagnostic device 30, for example, through communication via the communication unit 101. The actual processing data acquisition unit 103 is an example of the actual processing data acquisition unit according to the present invention.

[0056] Based on the first simulation data for each processing step obtained by the simulation data acquisition unit 102 and the first actual processing data for each processing step obtained by the actual processing data acquisition unit 103, the model generation unit 104 generates a trained model for predicting the second actual processing data based on the second simulation data. The model generation unit 104 stores the generated trained model in the storage unit 105. The model generation unit 104 is generated, for example, by machine learning methods such as multiple regression analysis, support vector machines, random forests, and gradient boosting trees (decision trees). The model generation unit 104 is an example of the model generation unit according to the present invention.

[0057] As described above, the model generation unit 104 generates a trained model based on the first simulation data for each processing step and the first actual processing data for each processing step. Therefore, the generated trained model is optimized for each processing step. Therefore, as described later, by predicting the second actual processing data (prediction data) for each processing step, the actual processing data can be predicted with high accuracy.

[0058] The storage unit 105 stores the trained model generated by the model generation unit 104. The storage unit 105 is an example of the storage unit according to the present invention.

[0059] Next, while referring to Figure 3 , the functional structure of the prediction device 20 will be described. The prediction device 20 includes a communication unit 201, a model acquisition unit 202, a simulation data acquisition unit 203, a prediction unit 204, and a storage unit 205.

[0060] The communication unit 201 communicates with an external device and transmits and receives various data as needed. The communication unit 201 is implemented, for example, through a network interface.

[0061] The model acquisition unit 202 acquires the trained model generated by the learning device 10 in the (3) learning stage and stores it in the storage unit 205. The model acquisition unit 202 acquires the trained model stored in the storage unit 105 of the learning device 10 via the communication unit 201, for example, and stores it in the storage unit 205.

[0062] The simulation data acquisition unit 203 acquires the second simulation data for each processing step generated by the terminal device 50 in the (4) prediction stage. The simulation data acquisition unit 203 acquires the second simulation data generated by the terminal device 50, for example, through communication via the communication unit 201.

[0063] The prediction unit 204 refers to the trained model stored in the storage unit 205, outputs the second simulation data for each processing step acquired by the simulation data acquisition unit 203 to the trained model, and predicts the second actual processing data (prediction data) for each processing step. The prediction unit 204 is an example of the prediction unit according to the present invention.

[0064] The storage unit 205 stores the trained model acquired by the model acquisition unit 202.

[0065] Next, while referring to Figure 4 , the functional structure of the diagnostic device 30 will be described. The diagnostic device 30 includes a communication unit 301, an actual processing data acquisition unit 302, a prediction data acquisition unit 303, a diagnostic unit 304, and a storage unit 305.

[0066] The communication unit 301 communicates with an external device and transmits and receives various data as needed. The communication unit 301 is implemented, for example, through a network interface.

[0067] The actual processing data acquisition unit 302 acquires the data obtained from the sensor 31 via the communication unit 301 as the second actual processing data for each processing step. Specifically, the actual processing data acquisition unit 302 acquires the data obtained from the sensor 31 during the actual processing of the machine tool 40 as the second actual processing data for each processing step. The actual processing data acquisition unit 302 stores the acquired second actual processing data in the storage unit 305.

[0068] The prediction data acquisition unit 303 acquires the second actual processing data (prediction data) predicted by the prediction device 20. The prediction data acquisition unit 303 stores the second actual processing data (prediction data) in the storage unit 305. The acquired prediction data acquisition unit 303 is an example of the prediction data acquisition unit according to the present invention.

[0069] The diagnostic unit 304 compares the second actual processing data acquired by the actual processing data acquisition unit 302 with the second actual processing data (prediction data) acquired by the prediction data acquisition unit 303, and diagnoses whether there is an abnormality during processing. The diagnostic unit 304 is an example of the diagnostic unit according to the present invention.

[0070] The storage unit 305 stores the second actual processing data for each processing step obtained by the actual processing data acquisition unit 302 and the second actual processing data (prediction data) obtained by the prediction data acquisition unit 303.

[0071] Next, while referring to Figure 5 , the functional structure of the terminal device 50 will be described. The terminal device 50 includes a communication unit 501, a simulation execution unit 502, and a storage unit 503.

[0072] The communication unit 501 communicates with an external device and transmits and receives various data as needed. The communication unit 501 is implemented, for example, through a network interface.

[0073] The simulation execution unit 502 is input with a machining program, performs a machining simulation implemented by the input machining program, and obtains simulation data. The simulation execution unit 502 stores the obtained simulation data in the storage unit 503.

[0074] The storage unit 503 stores the simulation data for each processing step obtained by the simulation execution unit 502.

[0075] Next, an example of the hardware structure of the learning device 10, the prediction device 20, the diagnosis device 30, and the terminal device 50 (hereinafter referred to as "the learning device 10 etc.") will be described while referring to Figure 6 while explaining. Figure 6 The learning device 10 etc. shown are implemented, for example, by a computer such as a personal computer, a microcontroller, or a PLC.

[0076] The learning device 10 etc. include a processor 1001, a memory 1002, an interface 1003, and a secondary storage device 1004 that are connected to each other via a bus 1000.

[0077] The processor 1001 is, for example, a CPU (Central Processing Unit). The processor 1001 reads the operation program stored in the secondary storage device 1004 into the memory 1002 and executes it, thereby implementing the various functions of the learning device 10 etc.

[0078] The memory 1002 is, for example, a main storage device constituted by a RAM (Random Access Memory). The memory 1002 stores the operation program read by the processor 1001 from the secondary storage device 1004. In addition, the memory 1002 functions as a working memory when the processor 1001 executes the operation program.

[0079] The interface 1003 is, for example, an I / O (Input / Output) interface such as a serial port, a USB (Universal Serial Bus) port, or a network interface. The functions of the communication units 101, 201, 301, and 501 are implemented through the interface 1003.

[0080] The secondary storage device 1004 is, for example, a flash memory, an HDD (Hard Disk Drive), or an SSD (Solid State Drive). The secondary storage device 1004 stores the action programs executed by the processor 1001. The functions of the storage units 105, 205, 305, and 503 are implemented through the secondary storage device 1004.

[0081] Next, while referring to Figure 7 , an example of the action generated by the model implemented by the learning device 10 will be described. For example, the user operates the learning device 10 and executes the Figure 7 shown action when instructing the generation of the learning model. Additionally, when starting the Figure 7 shown action, it is assumed that (1) the machining simulation phase and (2) the actual machining phase have been executed.

[0082] The simulation data acquisition unit 102 of the learning device 10 acquires the first simulation data for each machining process (step S101).

[0083] Next, the actual machining data acquisition unit 103 of the learning device 10 acquires the first actual machining data for each machining process (step S102).

[0084] The model generation unit 104 of the learning device 10 generates a trained model based on the first simulation data obtained in step S101 and the first actual machining data obtained in step S102, and stores it in the storage unit 105 (step S103). Subsequently, the learning device 10 ends the action of generating the model.

[0085] Next, while referring to Figure 8 , an example of the action of predicting the actual machining data implemented by the prediction device 20 will be described. For example, the user operates the prediction device 20 and executes the Figure 8 shown action when instructing the prediction of the actual machining data. Additionally, when starting the Figure 8 shown action, it is assumed that (3) the learning phase has been executed and a trained model has been generated.

[0086] The model acquisition unit 202 of the prediction device 20 acquires a trained model and stores it in the storage unit 205 (step S201). However, for example, if the latest trained model has already been stored in the storage unit 205, this operation does not have to be performed.

[0087] The simulation data acquisition unit 203 of the prediction device 20 acquires second simulation data for each machining process (step S202).

[0088] The prediction unit 204 of the prediction device 20 refers to the trained model stored in the storage unit 205 through step S201, and predicts the second actual machining data (prediction data) for each machining process based on the second simulation data for each machining process obtained through step S202 (step S203). Subsequently, the prediction device 20 ends the operation of predicting the actual machining data.

[0089] As described above, the machining system 1 according to Embodiment 1 has been described. According to the machining system 1 according to Embodiment 1, referring to a trained model generated based on the first simulation data for each machining process obtained by performing a machining simulation through the first machining program and the first actual machining data for each machining process obtained by performing the actual machining through the first machining program, based on the second simulation data for each machining process obtained by performing a machining simulation through the second machining program, it is possible to predict the second actual machining data (prediction data) for each machining process without performing actual machining.

[0090] In addition, according to the machining system 1 according to Embodiment 1, by using a trained model generated based on the first simulation data for each machining process and the first actual machining data for each machining process, predicting the second actual machining data (prediction data) for each machining process, so by comparing the second actual machining data and the second actual machining data (prediction data) for each machining process by the diagnostic device 30, it is possible to diagnose whether there is an abnormality during machining.

[0091] (Embodiment 2)

[0092] As described in the following explanation of Embodiment 2, in the abnormal diagnosis of the tool, the actual machining data (prediction data) predicted in Embodiment 1 is effectively used.

[0093] If the machine tool performs actual machining multiple times, the tool possessed by the machine tool will wear. Therefore, the following technology is required, that is, diagnosing whether there is an abnormality in the tool based on the actual machining state. For example, considering diagnosing the tool based on the sensor values detected by the sensors of the machine tool during the execution of actual machining.

[0094] In the case where the manufactured product by the machine tool is a mass-produced product, the processing conditions become the same in each actual processing. Therefore, for example, a trained model obtained by learning while associating sensor values with the presence or absence of an abnormality is generated, and based on this trained model and sensor values, an abnormality diagnosis of the cutting tool can be performed.

[0095] However, in the case where the manufactured product by the machine tool is not a mass-produced product but a single product, the processing shape and processing process are different in each actual processing. Therefore, the trained model used for the abnormality diagnosis of the cutting tool used in the manufacture of a single product cannot be generated based on a plurality of the same processing data in the same manner as the above-mentioned mass-produced product.

[0096] Embodiment 2 performs an abnormality diagnosis of a cutting tool based on the actual processing data obtained in actual processing and the predicted actual processing data (predicted data) predicted from the simulation data obtained by performing a machining simulation according to a machining program.

[0097] The overall structure of the processing system 1 according to Embodiment 2 is the same as that of Figure 1 the shown Embodiment 1. However, as described below, a part of the functions of the diagnosis device 30 is different. In addition, in Embodiment 2, (1) the machining simulation stage, (2) the actual machining stage, and (3) the learning stage are executed when the cutting tool 41 has not deteriorated. The processing system 1 according to Embodiment 2 is an example of the diagnosis system according to the present invention.

[0098] In Embodiment 2, the diagnosis device 30 diagnoses whether there is an abnormality in the cutting tool 41 of the machine tool 40 based on the second actual processing data obtained by the sensor 31 and the second actual processing data (predicted data) predicted by the prediction device 20. The diagnosis device 30 according to Embodiment 2 is an example of the diagnosis device according to the present invention.

[0099] The differences from Embodiment 1 in the functional structure of the diagnosis device 30 according to Embodiment 2 will be described.

[0100] The diagnosis unit 304 diagnoses whether there is an abnormality in the cutting tool 41 based on the feature amount of the second actual processing data obtained by the actual processing data acquisition unit 302 and the feature amount of the second actual processing data (predicted data) obtained by the predicted data acquisition unit 303. The diagnosis unit 304 is an example of the diagnosis unit according to the present invention.

[0101] As the above characteristic quantity, for example, the average value, variance, maximum / minimum value, median value, skewness, kurtosis, etc. of the torque in actual machining can be used. As a simple example, it is considered that when the difference between the average value of the torque in the prediction data and the average value of the torque in the actual machining data by the diagnosis unit 304 is greater than or equal to the threshold value, it is diagnosed that there is an abnormality in the tool 41. For example, when the characteristics of the tool 41 are known and the torque increases if it deteriorates, when the average value of the torque in the actual machining data significantly exceeds the average value of the torque in the prediction data, it is considered that there is an abnormality in the tool 41. In addition, depending on the machining process, the deviation of the characteristic quantities of the prediction data and the actual machining data sometimes exceeds and sometimes falls below.

[0102] A more complex example as described below is also considered. For example, consider the characteristic quantity of the second actual machining data that changes based on time series and the characteristic quantity of the second actual machining data (prediction data), and obtain an approximate curve related to the distribution when the horizontal axis is the elapsed time and the vertical axis is the difference between the characteristic quantity of the second actual machining data and the characteristic quantity of the second actual machining data (prediction data), and perform abnormality determination of the tool 41 based on this approximate curve. In addition, the approximate curve mentioned here also includes "lines" such as regression lines.

[0103] For example, consider obtaining a regression line as the approximate curve, and when the slope of the regression line is greater than or equal to the threshold value, it is diagnosed that there is an abnormality in the tool 41. The reason is that the greater the difference between the characteristic quantity of the second actual machining data and the characteristic quantity of the second actual machining data (prediction data), the greater the slope of the regression line. That is, it is considered that the greater the slope of the regression line, the greater the deviation between the second actual machining data and the second actual machining data (prediction data), that is, the more deteriorated the tool 41 is.

[0104] Alternatively, for example, it is also considered to obtain an approximate curve by curve fitting, and when an inflection point appears in the obtained approximate curve, it is diagnosed that there is an abnormality in the tool 41. An inflection point is a point where the change rate of the difference in characteristic quantities changes from positive to negative or from negative to positive, so it can be said that it is a point where the tendency of the difference in characteristic quantities changes significantly. It is considered that when the tool 41 does not deteriorate, the above-mentioned large change does not occur, so when an inflection point appears, it is considered that the tool 41 has deteriorated.

[0105] If the diagnosis unit 304 diagnoses whether there is an abnormality in the tool 41, it notifies the diagnosis result. The diagnosis unit 304 outputs the diagnosis result to a display / alarm unit (not shown) connected to the diagnosis device 30 through an alarm sound, message, etc. In addition, the method of notifying the diagnosis result is to display a message on the screen of the CNC 42 of the machine tool 40, output an alarm sound, and abort new machining using the tool 41.

[0106] Next, while referring toFigure 9 , an example of the operation of the abnormal diagnosis of the cutting tool performed by the diagnosis device 30 will be described. Figure 9 The operation shown is performed when actual machining is carried out after the above-mentioned stages (1) to (4) are executed.

[0107] The actual machining data acquisition unit 302 of the diagnosis device 30 acquires the second actual machining data from the sensor 31 (step S301).

[0108] The prediction data acquisition unit 303 of the diagnosis device 30 acquires the second actual machining data (prediction data) predicted by the prediction device 20 (step S302).

[0109] The diagnosis unit 304 of the diagnosis device 30 diagnoses whether there is an abnormality in the cutting tool 41 based on the feature amount of the second actual machining data obtained through step S301 and the feature amount of the second actual machining data (prediction data) obtained through step S302 (step S303).

[0110] The diagnosis unit 304 notifies the diagnosis result of step S303 (step S304). Subsequently, the diagnosis device 30 ends the operation of the abnormal diagnosis of the cutting tool.

[0111] As described above, the machining system 1 according to the second embodiment has been described. According to the machining system 1 according to the second embodiment, it is diagnosed whether there is an abnormality in the cutting tool 41 based on the feature amount of the second actual machining data (prediction data) predicted based on the feature amount of the second actual machining data and the second simulation data. Since the second actual machining data and the second actual machining data (prediction data) are data distinguished for each machining process, even when the workpiece machined by the machine tool 40 is not a mass-produced product, it is possible to appropriately diagnose whether there is an abnormality in the cutting tool 41.

[0112] (Modification example)

[0113] In the second embodiment, it is assumed that the diagnosis device 30 performs the abnormal diagnosis of the cutting tool 41, but the abnormal diagnosis may be performed by a device different from the diagnosis device 30. For example, the diagnosis device 30 may output the second actual machining data to a personal computer, and the personal computer acquires the second actual machining data and the second actual machining data (prediction data) to perform the abnormal diagnosis of the cutting tool 41. In this case, it is assumed that the personal computer has functions equivalent to Figure 4 the actual machining data acquisition unit 302, the prediction data acquisition unit 303, and the diagnosis unit 304 shown. This personal computer is an example of the diagnosis device according to the present invention.

[0114] In each embodiment, the learning device 10, the prediction device 20, the diagnostic device 30, and the terminal device 50 are separate devices. However, the functions of two or more of these devices may be integrated into one device. For example, the functions of each device may be included in a server, a personal computer, or the CNC 42 of the device 40, and various necessary data may be exchanged through communication via a network, a CPU bus, or the like.

[0115] In Figure 6 the hardware structure shown, the learning device 10 and the like have a secondary storage device 1004. However, it is not limited thereto, and the secondary storage device 1004 may be provided outside the learning device 10 and the like, and connected to the learning device 10 and the like and the secondary storage device 1004 via the interface 1003. In this method, a removable medium such as a USB flash drive or a memory card can also be used as the secondary storage device 1004.

[0116] In addition, instead of Figure 6 the hardware structure shown, the learning device 10 and the like may be configured by using a dedicated circuit such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array). In addition, in Figure 6 the hardware structure shown, a part of the functions of the learning device 10 and the like may be realized by a dedicated circuit connected to the interface 1003, for example.

[0117] The program used by the learning device 10 and the like can be stored in a computer-readable recording medium such as a CD-ROM (Compact Disc Read Only), a DVD (Digital Versatile Disc), a USB flash drive, a memory card, or an HDD and distributed. By installing this program on a specific or general computer, this computer can function as the learning device 10 and the like.

[0118] In addition, the above program may be stored in a storage device of another server on the Internet, and the above program may be downloaded from the server.

[0119] The present invention can achieve various embodiments and modifications without departing from the broad spirit and scope of the present invention. In addition, the above embodiments are used to explain the present invention and do not limit the scope of the present invention. That is, the scope of the present invention is shown not by the embodiments but by the claims. Moreover, various modifications implemented within the scope of the claims and the meaning of equivalent inventions are regarded as being within the scope of the present invention.

[0120] Description of reference numerals

[0121] 1 Machining system, 10 Learning device, 20 Prediction device, 30 Diagnosis device, 31 Sensor, 40 Machine tool, 41 Tool, 42 CNC, 50 Terminal device, 101 Communication unit, 102 Simulation data acquisition unit, 103 Actual machining data acquisition unit, 104 Model generation unit, 105 Storage unit, 201 Communication unit, 202 Model acquisition unit, 203 Simulation data acquisition unit, 204 Prediction unit, 205 Storage unit, 301 Communication unit, 302 Actual machining data acquisition unit, 303 Prediction data acquisition unit, 304 Diagnosis unit, 305 Storage unit, 501 Communication unit, 502 Simulation execution unit, 503 Storage unit, 1000 Bus, 1001 Processor, 1002 Memory, 1003 Interface, 1004 Secondary storage device.

Claims

1. A learning device, comprising: A simulation data acquisition unit that performs a machining simulation implemented by a first machining program to acquire first simulation data; An actual machining data acquisition unit that performs actual machining implemented by the first machining program to acquire first actual machining data; and A model generation unit that learns the relationship between the first simulation data and the first actual machining data, and generates a trained model for predicting second actual machining data obtained by performing actual machining implemented by a second machining program based on second simulation data obtained by performing a machining simulation implemented by the second machining program.

2. The learning device according to claim 1, wherein The first simulation data includes data for each machining process, The first actual machining data includes data for each machining process, The model generation unit generates a trained model for predicting the second actual machining data for each machining process.

3. A prediction device, comprising: A storage unit that stores the trained model generated by the learning device according to claim 1 or 2; A simulation data acquisition unit that performs a machining simulation implemented by a second machining program to acquire second simulation data; and A prediction unit that predicts the second actual machining data based on the trained model and the second simulation data.

4. A diagnostic device, comprising: An actual machining data acquisition unit that performs actual machining implemented by a second machining program to acquire second actual machining data; A predicted data acquisition unit that acquires predicted data of the second actual machining data predicted by the prediction device according to claim 3; and A diagnostic unit that diagnoses whether there is an abnormality in the machining based on the feature amount of the second actual machining data and the feature amount of the predicted data of the second actual machining data.

5. The learning device according to claim 4, wherein The diagnostic unit obtains an approximate curve related to the feature amount of the second actual machining data and the feature amount of the predicted data of the second actual machining data, and diagnoses whether there is an abnormality in the tool based on the obtained approximate curve.

6. A diagnostic system, comprising the prediction device according to claim 3 and the diagnostic device according to claim 4 or 5.

7. A model generation method Performing a machining simulation implemented by a first machining program to acquire first simulation data, Performing actual machining implemented by the first machining program to acquire first actual machining data, Learning the relationship between the first simulation data and the first actual machining data, and generating a trained model for predicting second actual machining data obtained by performing actual machining implemented by a second machining program based on second simulation data obtained by performing a machining simulation implemented by the second machining program.

8. The model generation method according to claim 7, wherein The first simulation data includes data for each machining process, The first actual machining data includes data for each machining process. Generate a trained model for predicting the second actual machining data for each machining process.

9. A prediction method Perform machining simulation implemented by the second machining program to obtain second simulation data. Predict the second actual machining data based on the trained model generated by the model generation method recited in claim 7 or 8 and the second simulation data.

10. A diagnosis method Perform actual machining implemented by the second machining program to obtain second actual machining data. Diagnose whether there is an abnormality during machining based on the feature quantity of the second actual machining data and the feature quantity of the predicted data of the second actual machining data predicted by the prediction method recited in claim 9.

11. The diagnosis method according to claim 10, wherein Obtain an approximate curve related to the feature quantity of the second actual machining data and the feature quantity of the predicted data of the second actual machining data, and diagnose whether there is an abnormality in the tool based on the obtained approximate curve.

12. A program that causes a computer to function as the following units: A simulation data acquisition unit that performs machining simulation implemented by the first machining program to obtain first simulation data; An actual machining data acquisition unit that performs actual machining implemented by the first machining program to obtain first actual machining data; and A model generation unit that learns the relationship between the first simulation data and the first actual machining data, and generates a trained model for predicting the second actual machining data obtained by performing actual machining implemented by the second machining program based on the second simulation data obtained by performing machining simulation implemented by the second machining program.

13. The program according to claim 12, wherein The first simulation data includes data for each machining process. The first actual machining data includes data for each machining process. The model generation unit generates a trained model for predicting the second actual machining data for each machining process.

14. A program that causes a computer to function as the following units: A storage unit that stores the trained model generated by the program recited in claim 12 or 13; A simulation data acquisition unit that performs machining simulation implemented by the second machining program to obtain second simulation data; and A prediction unit that predicts the second actual machining data based on the trained model and the second simulation data.

15. A program that causes a computer to function as the following units: An actual machining data acquisition unit that performs actual machining implemented by the second machining program to obtain second actual machining data; A predicted data acquisition unit that acquires the predicted data of the second actual machining data predicted by the program recited in claim 14; and A diagnostic unit that diagnoses whether there is an abnormality during machining based on the characteristic quantities of the second actual machining data and the characteristic quantities of the predicted data of the second actual machining data.

16. The program according to claim 15, wherein the diagnostic unit obtains an approximate curve related to the characteristic quantities of the second actual machining data and the characteristic quantities of the predicted data of the second actual machining data, and diagnoses whether there is an abnormality in the cutting tool based on the obtained approximate curve.

Images