Tool life prediction system

Abstract

A tool life prediction system capable of predicting the life of a tool with higher accuracy is provided. The tool life prediction system (1) includes: a machine tool main body (11) that performs machining of a workpiece (W) using a tool (T); a detector (13) that detects observable state data in the machine tool main body while machining the workpiece; a learned model storage unit (81) that stores, for each of a plurality of life cases, a learned model generated by performing machine learning using a training data set including explanatory variables of the state data detected by the detector and a target variable of the number of remaining machining of the workpiece before reaching the life in one tool; and a remaining machining number prediction unit (83) that selects one learned model from the plurality of learned models based on the state data and predicts the number of remaining machining using the selected one learned model and the state data.

Description

Technical Field

[0001] This invention relates to a tool life prediction system. Background Technology

[0002] From a tool cost perspective, predicting the lifespan of tools used in machining is crucial. Traditionally, tool lifespan was determined by pre-determining the number of machining operations required to reach its lifespan, taking into account individual tool variations and safety margins. However, this method of determining lifespan also considers tools to have reached their lifespan even if they haven't actually reached it, and the tools are then replaced.

[0003] Related tool life prediction systems use machining information, such as the drive current of the spindle motor, during tool processing to predict tool life (see, for example, Patent Document 1). In this prediction method, a computational model determined based on machining information is used to predict tool life. Furthermore, the computational model used for tool life prediction is selected from multiple computational models, for example, based on the workpiece material and machining conditions.

[0004] Patent Document 1: JP2019-82836A

[0005] In tool life prediction, different prediction results arise depending on the computational model used for prediction. In other words, determining the computational model used for tool life prediction is a crucial factor. For example, even with the same tool type, workpiece material, and processing conditions, individual tool differences can lead to variations in tool life. Summary of the Invention

[0006] The purpose of this invention is to provide a tool life prediction system that can predict tool life with higher accuracy.

[0007] The tool life prediction system of the present invention comprises: a machining machine body for machining a workpiece using a tool; a detector for detecting observable state data in the machining machine body during the machining of the workpiece; a learned model storage unit for storing learned models generated by machine learning using a training dataset containing the above-mentioned explanatory variables and the remaining number of machining operations of the workpiece before reaching the end of its life in a tool, for each tool life (life mode) of a plurality of tools; and a remaining machining operation prediction unit for selecting one of the learned models from the plurality of learned models based on the above-mentioned state data and predicting the remaining number of machining operations using the selected learned model and the above-mentioned state data.

[0008] According to the tool life prediction system, a learned model is selected from multiple learned models, and this selected learned model is used to predict the life of the currently used tool. The selection of the learned model used for tool life prediction uses state data from the machining process performed using the tool. Therefore, the selected learned model becomes the learned model corresponding to that tool. In this way, by making the learned model used for tool life prediction a model corresponding to that tool, the tool's life can be predicted with high accuracy. Attached Figure Description

[0009] Figure 1 This is a diagram showing the structure of a tool life prediction system.

[0010] Figure 2 This is a diagram showing an example of a processing machine.

[0011] Figure 3 This is a diagram showing the functional module structure of the tool life prediction system.

[0012] Figure 4 It is a diagram illustrating the calculation of the remaining processing times in the remaining processing times calculation section, and it is also a diagram representing the training dataset.

[0013] Figure 5 This is a diagram showing the correspondence between tool No. and the learned model No.

[0014] Figure 6 It is a graph representing multiple learned models stored in the model storage section.

[0015] Figure 7 This is a flowchart representing the processing in the remaining processing times prediction section.

[0016] Figure 8 This is a graph representing information about the amount of p times obtained by the remaining processing times prediction department.

[0017] Figure 9 It is a diagram used to illustrate the concept of similarity between the values ​​of the predicted objects in the remaining processing times prediction section, and it is a diagram showing multiple learned models (white circles) and the values ​​of the predicted objects (black circles) in the feature space (n-dimensional space).

[0018] Figure 10 This is a graph showing the content displayed on the display device, specifically the content displayed after 15 actual processing cycles. The dashed line represents the lifetime prediction trend based on the selected learned model.

[0019] Figure 11This is a graph showing the content displayed on the display device, specifically the content displayed after 30 actual processing cycles. The dashed line represents the lifetime prediction trend based on the selected learned model.

[0020] Figure 12 This is a graph showing the content displayed on the display device, specifically the content displayed after 50 actual processing cycles. The dashed line represents the lifetime prediction trend based on the selected learned model.

[0021] Figure 13 This is a graph showing the content displayed on the display device, specifically the content displayed after 80 actual processing cycles (when the predicted lifespan is reached). The dashed line represents the lifespan prediction trend based on the selected learned model. Detailed Implementation

[0022] (1. The types of machining machinery and tools to which tool life prediction systems are applied)

[0023] A tool life prediction system predicts the lifespan of a tool used in machining processes. Here, tool life refers not only to a state where the tool is completely unusable, but also to a state where tool adjustments are required. For example, tool adjustments include resharpening the tool, reshaping the grinding machine, and dressing.

[0024] As machining machinery, this includes machining centers, lathes, gear machining devices, boring machines, and other machining machines that perform cutting operations. In this case, the tools are cutting tools. For example, the tools in a machining center include: drills, milling cutters, boring tools, gear machining tools, turning tools, etc. The tools in a lathe at least include turning tools, and the tools in a compound lathe, in addition to turning tools, also include drills, milling cutters, and other tools similar to those in a machining center. As a gear machining device, this includes gear peeling machines, hobbing machines, shaping machines, etc., and the tools in this gear machining device include gear peeling tools, hobbing tools, shaping tools, and other tools used for gear machining.

[0025] In addition, as processing machinery, it includes grinding machines for performing grinding operations. The tools in a grinding machine include a grinding mill. Furthermore, as processing machinery, it includes forging machinery, such as stamping presses and rolling mills. The tools in stamping presses and rolling mills include forging punches and forging dies.

[0026] (2. Overview of the structure of tool life prediction system 1)

[0027] Reference Figure 1The structure of the tool life prediction system 1 is described in general. The tool life prediction system 1 includes at least one machining machine 10 and a computing unit (20, 30). The machining machine 10 can be used as the object, or... Figure 1 The example shown uses multiple machining machines 10 as objects. In this case, we take the tool life prediction system 1 as an example where multiple machining machines 10 are involved.

[0028] The machining machine 10 includes at least: a machining machine body 11 for machining workpiece W using a tool T; and a detector 13 for detecting observable state data in the machining machine body during the machining of workpiece W. That is, the detector 13 detects observable states in the machining machine body during the machining of workpiece W and acquires information containing those observable states as state data.

[0029] The computing devices (20, 30) predict the lifespan of tool T by using state data detected by detector 13 and applying machine learning. Figure 1 The diagram shows that the computing devices (20, 30) consist of a learning processing device 20 and a prediction computing device 30. The learning processing device 20 and the prediction computing device 30 are independent structures, but they can also be configured as a single device. In addition, part or all of the computing devices (20, 30) can also be embedded in the processing machine 10.

[0030] In this example, we consider the case where the learning processing unit 20 and the prediction computing unit 30 are independent structures. Furthermore, the learning processing unit 20 has a so-called server function and is communicatively connected to multiple processing machines 10. On the other hand, the prediction computing unit 30 is configured one-to-one with each processing machine 10 and is communicatively connected to each processing machine 10. In other words, the multiple prediction computing units 30 function as so-called edge computers and are capable of high-speed computational processing.

[0031] (3. Detailed description of the structure of Tool Life Prediction System 1)

[0032] Reference Figure 1 The structure of the tool life prediction system 1 will be described in more detail below. The tool life prediction system 1 includes: multiple machining machines 10, a learning processing device 20 that functions as part of a computing device, and multiple prediction computing devices 30 that function as another part of the computing device.

[0033] Each processing machine 10, as described above, can be used with various processing machines. The processing machine 10 includes: a processing machine body 11 for machining workpiece W using tool T, a control device 12 for controlling the processing machine body 11, a detector 13, and an interface 14.

[0034] The machining machine body 11 has a tool T and a structure that supports the workpiece W and allows the tool T and workpiece W to move relative to each other. In other words, the machining machine body 11 includes a structure and a drive device that drives the structure. The control device 12 includes a CNC device and a PLC device. The control device 12 controls the drive device in the machining machine body 11. The interface 14 is a device that enables the machining machine body 11, the control device 12, and the detector 13 to communicate with external devices.

[0035] Detector 13 detects observable state data within the machining machine body 11 during workpiece W machining. Detector 13 detects data related to machining load, drive load of the drive device, etc. Detector 13 may be, for example, a current sensor that detects drive current data of the motor (i.e., detects the drive current of the motor and acquires data including the state of that drive current as drive current data), a vibration sensor that detects vibration data of the components of the machining machine body 11 (i.e., detects the vibration of the components of the machining machine body 11 and acquires data including the state of that vibration as vibration data), or a microphone that detects sound data during machining (i.e., detects sound during machining and acquires data including the state of that sound as sound data). In other words, the state data includes, for example, drive current data, vibration data, sound data, etc.

[0036] The learning processing device 20 is configured to include a processor 21, a storage device 22, an interface 23, etc. Furthermore, the learning processing device 20 has server functionality and can be connected to multiple processing machines 10 in a communicative manner.

[0037] The learning processing device 20 generates a learned complete model for predicting the lifespan of tool T by applying machine learning based on the state data detected by detector 13. Specifically, the learning processing device 20 generates learned complete models for the individual tool lifespans (each lifespan pattern) of multiple tools T. In other words, the learning processing device 20 generates multiple learned complete models.

[0038] For example, in the learning processing apparatus 20, multiple learned models are generated by generating a learned model for one tool T among multiple tools T of a certain type. In other words, the learning processing apparatus 20 generates a learned model for each of the multiple lifetime patterns by treating each tool T as a separate lifetime pattern and generating a learned model for each of the multiple lifetime patterns.

[0039] Each prediction processing unit 30 is configured to include a processor 31, a storage device 32, an interface 33, etc. Furthermore, the prediction processing unit 30 is communicatively connected to the learning processing unit 20, which acts as a server, and the corresponding processing machine 10.

[0040] Each predictive computing unit 30 is configured close to each machining machine 10 and functions as a so-called edge computer. The predictive computing unit 30 uses multiple learned models generated by the learning processing unit 20 to predict the lifespan of the tool T based on state data detected by the detector 13 during the machining of workpiece W. In this example, the predictive computing unit 30 predicts the remaining machining operations, i.e., the remaining number of times workpiece W can be machined.

[0041] In particular, in this example, the prediction processing unit 30 selects one model from a plurality of learned models and uses the selected learned model to predict the lifetime of the tool T. Specific examples of the selection method will be described later.

[0042] The tool life prediction system 1 also includes a common display device 40 and multiple discrete display devices 50. However, the tool life prediction system 1 may also have a structure without the common display device 40 or without the discrete display devices 50. The common display device 40 is configured corresponding to the learning processing device 20. In addition, the discrete display devices 50 are configured corresponding to each processing machine 10.

[0043] (4. Example of machining machine body 11)

[0044] As an example of the main body 11 of the processing machinery, refer to Figure 2 A gear machining apparatus for performing gear peeling is described. Furthermore, as described above, the gear machining apparatus of the machining machine body 11 is an example, and can also be applied to other machining machines.

[0045] like Figure 2 As shown, the machining machine body 11, serving as a gear machining device, acts as the drive axis for changing the relative position and posture of the workpiece W and the tool T, and utilizes a structure of a 5-axis machining center with three forward axes and two rotary axes. In this example, the machining machine body 11 has three orthogonal axes (X-axis, Y-axis, and Z-axis) as forward axes, and two rotary axes, B-axis and Cw-axis. The B-axis is a rotary axis about the Y-axis, and the Cw-axis is a rotary axis about the central axis of the workpiece W.

[0046] The machining machine body 11 includes a tool spindle 61, which supports a tool T (rotary tool) and is rotatable about the Ct axis, and is movable in the Y-axis and Z-axis directions respectively. Furthermore, the machining machine body 11 includes a workpiece spindle 62, which supports a workpiece W and is rotatable about the Cw axis, and is rotatable about the B axis, and is movable in the X-axis direction. The machining machine body 11 includes a motor, which serves as a drive device for driving (moving) in the directions of each axis (X-axis, Y-axis, Z-axis, B-axis, Cw-axis, Ct-axis).

[0047] (5. Functional module structure of tool life prediction system 1)

[0048] Reference Figure 3 The functional modules of the tool life prediction system 1 are described below. The tool life prediction system 1 includes a detector 13, a counter 12a, a learning processing device 20, a prediction calculation device 30, and display devices 40 and 50.

[0049] As described above, detector 13 detects observable state data in the machining machine body 11 during the machining of workpiece W. State data includes, for example, drive load data in the motor that drives the rotary tool T. State data also includes, for example, drive load data in the motor that drives the workpiece W. Drive load data corresponds to the motor's drive current data. Additionally, state data may also include vibration data, machining sound data, etc. State data is a time-series data of a workpiece W from the start to the end of machining.

[0050] The counter 12a is included in the control device 12 of the machining machine 10 and counts the number of times workpiece W is processed for each tool T. In other words, the counter 12a counts the number of workpiece W processed for each tool T since it was first used. In addition, the counter 12a may also have a detector 13 itself in addition to the control device 12.

[0051] The learning processing device 20 generates a learned model for predicting the lifespan of tool T based on data including state data detected by detector 13 and processing counts acquired by counter 12a. The learning processing device 20 includes: a training dataset acquisition unit 71, a training dataset storage unit 72, and a model generation unit 73.

[0052] The training dataset acquisition unit 71 acquires a training dataset for machine learning for each of the multiple tools T in one class. This means that the training dataset acquisition unit 71 acquires a training dataset for each of the multiple lifetime patterns by treating each tool T as a separate lifetime pattern and acquiring a training dataset for each tool T.

[0053] The training dataset acquisition unit 71 includes: a state data acquisition unit 71a, a feature quantity calculation unit 71b, a processing number acquisition unit 71c, a lifespan information acquisition unit 71d, and a remaining processing number calculation unit 71e. The state data acquisition unit 71a acquires state data detected by the detector 13 during the processing of the workpiece W. In one processing operation performed by a tool T, the state data acquisition unit 71a acquires state data for the number of detectors 13. Furthermore, the state data acquisition unit 71a acquires state data for each of the multiple tools T of the same type.

[0054] The feature quantity calculation unit 71b calculates multiple feature quantities of the state data acquired by the state data acquisition unit 71a. Here, the feature quantities use various statistics from the state data. For example, the feature quantities are the maximum, minimum, mean, variance, standard deviation, skewness, kurtosis, median, etc., of the state data. Additionally, the feature quantities may also include statistics for data obtained by differentiating the state data, statistics for data after frequency analysis of the state data, etc. The feature quantity calculation unit 71b can calculate all of the above-mentioned feature quantities, or it can calculate only a portion of the feature quantities.

[0055] The machining count acquisition unit 71c acquires the number of machining operations from the counter 12a for each tool T since the start of machining. The tool life information acquisition unit 71d receives input indicating that the tool T of the object has reached its lifespan. Alternatively, the operator can determine whether the tool T of the object has reached its lifespan and input this information into the tool life information acquisition unit 71d. The determination of whether the tool T has reached its lifespan can be made, for example, by checking whether scratches called tool marks have formed on the surface of the workpiece W. Alternatively, the determination can be made based on whether the machining accuracy of the workpiece W has significantly decreased. In addition, if the determination is made by an inspection device, the tool life information acquisition unit 71d can also input the information through an inspection device.

[0056] The remaining processing count calculation unit 71e determines the remaining processing count (first remaining processing count) of tool T based on information including the processing count acquisition unit 71c (first processing count) and the lifespan completion information acquisition unit 71d (lifespan completion information acquisition unit 71d). The remaining processing count is set to zero when the lifespan is completed, and increases in ascending order back to before the lifespan was completed. Figure 4 As shown in the left sidebar, when the product is deemed to have reached the end of its lifespan after N processing cycles, as follows... Figure 4 As shown in the right column, the remaining number of processing cycles is zero when the lifespan is reached, initially being (N-1).

[0057] The training dataset storage unit 72 stores the training dataset acquired by the training dataset acquisition unit 71. Specifically, as follows: Figure 4As shown in the right column, the training dataset storage unit 72 establishes and stores the association between the feature quantities DATA(1)~DATA(N) of the state data and the remaining number of processing times (N-1)~(0). Figure 4 The content shown in the right-hand column is the training dataset associated with one tool T. In other words, the training dataset storage section 72 stores the training dataset associated with each of the multiple tools T.

[0058] The model generation unit 73 performs machine learning using the training dataset stored in the training dataset storage unit 72. Specifically, the model generation unit 73 performs machine learning for each lifetime mode, i.e., for each tool T, by setting the feature quantities of the state data detected by the detector 13 as explanatory variables and setting the remaining processing times (first remaining processing times) as the objective variable. Furthermore, the model generation unit 73 generates a fully learned model for predicting the lifetime of tool T.

[0059] Here, the learned models are generated for each lifetime pattern, i.e., for each tool T. Therefore, the model generation unit 73 generates a large number of learned models of the same type and for multiple tools T. In other words, the correspondence between tool T and learned models is as follows: Figure 5 As shown. For example, tool No.a corresponds to the learned model No.A, and the following are the same correspondences. Furthermore, Figure 5 All the tools T shown are of the same type.

[0060] Machine learning methods, for example, can utilize regression. Linear regression, ridge regression, Lasso, elastic net regression, and random forest regression are useful, for instance. In particular, these methods allow for understanding the influence of multiple features and can be used for feature selection as needed. Furthermore, machine learning methods can also be applied to other approaches besides regression.

[0061] The prediction calculation device 30 predicts the lifespan of the tool T used in machining based on the state data during machining in the corresponding machining machine 10. The prediction calculation device 30 includes: a model storage unit 81, a prediction data acquisition unit 82, a remaining machining number prediction unit 83, and an output unit 84.

[0062] The model storage unit 81 stores multiple learned models generated by the model generation unit 73. For example... Figure 6 As shown, the model storage unit 81 stores multiple learned models such as model No. A, B, ... . These multiple learned models, as described above, correspond to each lifetime mode, and in this example, to each tool T.

[0063] The prediction data acquisition unit 82 acquires prediction data during processing using the tool T of the prediction object. The prediction data acquisition unit 82 includes: a state data acquisition unit 82a, a feature quantity calculation unit 82b, and a processing number acquisition unit 82c.

[0064] The state data acquisition unit 82a acquires state data detected by the detector 13 during the machining of workpiece W using the tool T of the prediction object. Here, the tool T of the prediction object is of the same type as the tool T used when acquiring the data in the training dataset acquisition unit 71.

[0065] The feature calculation unit 82b calculates the feature quantities of the state data acquired by the state data acquisition unit 82a. Here, the feature quantities use various statistics from the state data. Moreover, the feature quantities are of the same type as those calculated in the feature calculation unit 71b of the training dataset acquisition unit 71. The processing number acquisition unit 82c acquires from the counter 12a the number of processing times in the tool T of the prediction object since the start of processing.

[0066] Here, the state data acquisition unit 82a and the feature calculation unit 82b perform the same processing as the state data acquisition unit 71a and the feature calculation unit 71b of the training dataset acquisition unit 71. Furthermore, in this example, the state data acquisition unit 82a and the feature calculation unit 82b are described as different elements from the state data acquisition unit 71a and the feature calculation unit 71b of the training dataset acquisition unit 71. However, it is also possible to use the elements 71a and 71b of the training dataset acquisition unit 71 and the elements 82a and 82b of the prediction data acquisition unit 82. In other words, the functions of elements 71a and 71b in the learning processing device 20 are shared with some functions of the prediction calculation device 30.

[0067] The remaining machining count prediction unit 83 selects a learned model from a plurality of learned models stored in the model storage unit 81, and predicts the remaining machining count of the tool T based on the selected learned model and the data acquired by the prediction data acquisition unit 82. In other words, the remaining machining count prediction unit 83 predicts the number of workpieces W that the tool T can process, or the number of times the tool T can process one workpiece W. Details of the processing by the remaining machining count prediction unit 83 will be described later.

[0068] The output unit 84 outputs information related to the remaining processing times of the tool T to the display devices 40 and 50, as predicted by the remaining processing times prediction unit 83. Furthermore, in addition to the remaining processing times, the output unit 84 can also output the current processing times and the prediction trend made by the selected learned model for prediction.

[0069] Display devices 40 and 50 display information output from the output unit 84 of the prediction calculation device 30. Details regarding the content displayed by display devices 40 and 50 will be described later.

[0070] (6. Processing of the remaining processing times prediction unit 83)

[0071] Reference Figure 7 - Figure 9 The processing by the remaining processing times prediction unit 83 (hereinafter referred to as the "prediction unit") will be explained. Figure 7 As shown, the prediction unit 83 obtains multiple feature quantities of the state data of the most recent p quantities from the prediction data acquisition unit 82 (step S1). In other words, the prediction unit 83 obtains multiple feature quantities of the processing state data of the most recent p quantities of the workpiece W from the prediction object tool T.

[0072] Next, the prediction unit 83 obtains from the prediction data acquisition unit 82 the actual number of times the tool T is processed, corresponding to the number of times multiple features are acquired (step S2). In other words, as... Figure 8 As shown, the prediction unit 83 obtains information that establishes a correlation between multiple feature quantities DATA(N1) to DATA(Np) of the state data of the most recent p times and the actual number of processing operations (N1) to (Np) corresponding to that state data. The value of p can be, for example, around 5 to 10, but is not limited to 5 to 10. If p is set to a large value, initial prediction cannot be performed. However, if p is set to a small value, it will affect the prediction accuracy. Therefore, as mentioned above, the value of p is preferably 5 to 10. In addition, the value of p can also be appropriately changed according to the number of processing operations up to the life of tool T.

[0073] Next, the prediction unit 83 selects a fully learned model from multiple fully learned models based on multiple feature quantities of the state data of the most recent p quantities and the corresponding actual number of processing times (hereinafter referred to as "prediction object data") (step S3). The prediction unit 83 compares the prediction object data of the most recent p quantities with the data sets in the training dataset used to generate each of the multiple fully learned models, and selects the fully learned model with the highest similarity.

[0074] Reference Figure 9 This section explains the concept of selecting a learned model. Figure 9 In the feature space (n-dimensional space) where each of the n features is represented as a dimensional element, the transitions of the data (hereinafter referred to as "base data") in the training dataset used to generate each of the multiple learned models are represented by white circles, arrows, and × symbols.

[0075] For example, in Figure 9 In the diagram, the white circle representing the completed learning model No. A lies in the coordinates of the n-dimensional space of the underlying data related to model No. A. Furthermore, the arrows indicate the direction of movement of the coordinates within the underlying data related to model No. A, from the start of processing by tool T to the attainment of its lifespan. The × symbol indicates the underlying data at the point of reaching its lifespan. In other words, as... Figure 9 As shown, in the feature space (n-dimensional space), the changes in the basic data related to multiple learned models vary at various locations and in various directions.

[0076] Here, in Figure 9 In the diagram, black circles and arrows represent the data to be predicted. Figure 9 The example shown is the case of prediction target data for a quantity of 5. Furthermore, the prediction unit 83 selects data similar to the prediction target data (black circles) for the multiple quantities from the base data related to each of the multiple learned models. For example, in... Figure 9 In the diagram, the portion of the basic data related to the learned model No.A, enclosed by an elliptical dashed line, represents basic data similar to the prediction target data (black circle).

[0077] The above can be understood conceptually as follows: The prediction unit 83 compares the changes in multiple feature quantities of the state data with the increase in the actual number of processing operations (first change) with the changes in multiple feature quantities of the state data with the increase in the number of processing operations in the training dataset used to generate each of the multiple learned models (second change). Furthermore, based on the comparison of changes, the prediction unit 83 selects the learned model with the highest similarity. In other words, it selects a learned model generated from the training dataset with a second change that has a higher similarity to the first change.

[0078] Furthermore, such as Figure 9 As shown, for similarity, in the feature space (n-dimensional space), the combination of basic data and the combination of prediction object data are compared. This is equivalent to a smaller distance value in n-dimensional space indicating higher similarity. The so-called combination of basic data refers to the combination of multiple features of the state data in the training dataset as the number of processing iterations increases. The so-called combination of prediction object data refers to the combination of multiple features of the state data in the prediction object as the number of actual processing iterations increases.

[0079] Furthermore, similarity is determined using any one of the following distances: Euclidean distance, Mahalanobis distance, Manhattan distance, and Chebyshev distance. In other words, the prediction unit 83 uses any of these distances to select the learned model with the highest similarity (the smallest distance equivalent value).

[0080] Next, the prediction unit 83 uses a selected fully learned model and multiple features of the state data to predict the remaining number of processing steps (second remaining number of processing steps) (step S4). As described above, the fully learned model is generated by machine learning using multiple features of the state data as explanatory variables and the remaining number of processing steps (first remaining number of processing steps) as the objective variable. Therefore, by setting the multiple features as explanatory variables as input data, the fully learned model can output the remaining number of processing steps as the objective variable.

[0081] Next, the prediction unit 83, based on the prediction result, determines whether the tool T of the prediction object has reached its lifespan (step S5), and returns to step S1. If it has not reached its lifespan, it predicts the remaining number of processing times again for the next processing (step S5: No). If the prediction unit 83 determines that the tool T of the prediction object has reached its lifespan (step S5: Yes), it ends the prediction process.

[0082] (7. Display content of display devices 40 and 50)

[0083] Next, refer to Figure 10 - Figure 13 Examples of the display content on display devices 40 and 50 will be described below. The content that can be displayed on any of display devices 40 and 50 is as described above. However, the display content is just an example and is not limited to this.

[0084] like Figure 10 - Figure 13 As shown, on display devices 40 and 50, for example, a graph is shown with the actual number of processing operations of the tool T being predicted as the horizontal axis and the predicted remaining number of processing operations as the vertical axis. Black circles depict points that have been processed, and white circles depict points of the currently predicted object. In other words, the change in the prediction of the remaining number of processing operations to date is displayed using white circles and multiple black circles.

[0085] Furthermore, in Figure 10 - Figure 13 The top column displays the learned model currently used for prediction. As explained in the processing of prediction unit 83 above, since the learned model used for prediction is selected each time the number of processing cycles increases, there is a possibility that it will change sequentially. Of course, there is also a possibility that the same learned model will be continuously selected from the start of processing until the end of its lifespan. Therefore, the currently selected learned model is displayed on display devices 40 and 50.

[0086] Furthermore, the display devices 40 and 50 display the lifespan prediction trend (slanted dashed line) using the currently selected learned model. If prediction continues using this learned model, it indicates that the prediction will move along the lifespan prediction trend line. Moreover, the point on the lifespan prediction trend line where the remaining processing times are zero indicates the number of times the tool T of the predicted object can be processed (the second processing times).

[0087] The following is a detailed explanation of each figure. Figure 10 The diagram shows a scenario where the tool T used for prediction has performed 15 processing iterations, leaving approximately 60 processing iterations remaining. Furthermore, the currently used, fully learned model is No. B. Figure 10 The diagram shows the predicted results when the actual number of processing cycles is 5 and 10. It can be seen that when the actual number of processing cycles is 5, the predicted remaining processing cycles are approximately 70, and when it is 10, the predicted remaining lifetime is approximately 65. Furthermore, in... Figure 10 In the middle, it is drawn every 5 times, but it can also be drawn every time.

[0088] exist Figure 11 The diagram shows the case where the tool T, used to predict the object, performs 30 actual processing operations, leaving approximately 65 remaining. Furthermore, the currently used, fully learned model is No. G. Additionally, in... Figure 11 The diagram shows the predicted results for actual processing times of 5, 10, 15, 20, and 25 cycles. It can be seen that the lifetime prediction trend for the first 15 cycles differs from the lifetime prediction trend for 20 to 30 cycles.

[0089] Suppose that, given the tool life of a tool T very similar to the tool being predicted, and considering that the state of the workpiece W machined by the learned tool T is very similar to the state of the workpiece being predicted, there is a possibility of predicting the remaining machining cycles with minimal deviation and high accuracy. However, due to subtle differences in the surface treatment of tool T, subtle differences in the quenching state of workpiece W, etc., it is rare to reproduce a machining state that is very similar to that learned in the past. Therefore, each time a prediction process is performed, the selected learned model changes, and there is a possibility of prediction deviation. Therefore, the occurrence of prediction deviation itself is not a problem; rather, it has the advantage of indicating that the prediction accuracy is not very high. Furthermore, the display devices 40 and 50 display the prediction result with the highest accuracy at that moment.

[0090] exist Figure 12 The diagram shows the case where the tool T, used for prediction, performs 50 actual processing operations, leaving approximately 30 remaining. Furthermore, the learned model used for the current prediction is No.D. Figure 12 The results shown are predicted with increments of 5 for actual processing times ranging from 5 to 50.

[0091] exist Figure 13 The diagram shows the situation when the tool T, the object of the prediction, reaches its lifespan. The actual number of processing cycles at the time of lifespan is 80. Furthermore, the learned model used for prediction at the time of lifespan is No.D.

[0092] (8. Effect)

[0093] According to the tool life prediction system 1, a learned model is selected from multiple learned models, and the selected learned model is used to predict the life of the currently used tool T. The selection of the learned model used for tool T life prediction uses state data from the machining process performed by the tool T. Therefore, the selected learned model is the learned model corresponding to the tool T. In this way, by making the learned model used for tool T life prediction correspond to the tool T, the life of the tool T can be predicted with high accuracy.

[0094] Furthermore, the display devices 40 and 50 display the remaining number of machining operations for the tool T, which is the subject of the lifespan prediction. Therefore, the operator can easily determine the remaining number of machining operations for the tool T by checking the display devices 40 and 50.

Claims

1. A lifespan prediction system for a predictive object tool, comprising: The main body of the machining machinery is configured to use tools to perform mechanical processing on workpieces; The detector is configured to detect observable states in the machining machine body during the machining of the workpiece, and to acquire data containing the observable states as state data. The learned model storage unit stores learned models according to the tool life of each of the aforementioned tools. Each learned model is generated by machine learning using multiple training datasets containing the aforementioned descriptive and objective variables, with the aforementioned state data as explanatory variables and the first remaining number of machining operations based on the first number of machining operations performed on the workpiece before reaching the end of its tool life. The remaining machining count prediction unit is configured to select one of the learned models from a plurality of learned models based on the aforementioned state data, and use the learned model and the aforementioned state data to predict a second remaining machining count, which is a predicted second machining count based on the possibility of machining the workpiece before the predicted tool reaches its lifespan. The aforementioned remaining processing count prediction unit is configured to: acquire, for the aforementioned prediction target tool, the aforementioned status data of multiple processing times of the aforementioned workpiece recently, and the actual number of processing times performed using the aforementioned tool, which is the same as the aforementioned recent processing count; Based on the obtained state data and the actual number of processing operations, one of the learned models is selected.

2. The lifespan prediction system for the prediction object tool according to claim 1, wherein, The aforementioned remaining processing times prediction unit is configured as follows: The first transition of the state data accompanying the increase in the actual number of processing operations performed using the aforementioned prediction tool is compared with the second transition of the state data accompanying the increase in the number of processing operations performed using the aforementioned tool in the aforementioned multiple training datasets. Based on comparison, the learned model is generated from the training dataset that has the highest similarity to the first transition among the multiple training datasets mentioned above. The selected learned model and the state data are used to predict the second remaining number of processing steps.

3. The lifespan prediction system for the prediction object tool according to claim 2, wherein, The aforementioned remaining processing times prediction unit is configured such that the first transition is a change in multiple feature quantities of the aforementioned state data accompanying the increase of the aforementioned actual processing times, and the second transition is a change in multiple feature quantities of the aforementioned state data in the aforementioned training dataset accompanying the increase of the aforementioned processing times.

4. The lifespan prediction system for the prediction object tool according to claim 3, wherein, Regarding the aforementioned similarity, in the feature space of elements representing the aforementioned features as dimensions, the combination of multiple features of the aforementioned state data in the aforementioned training dataset that increase with the aforementioned number of processing times, and the combination of multiple features of the aforementioned state data that increase with the aforementioned actual number of processing times are compared. The smaller the value equivalent to distance in the aforementioned feature space, the higher the similarity is judged.

5. The lifespan prediction system for the prediction object tool according to claim 4, wherein, The above similarity is determined using any one of the following: Euclidean distance, Mahalanobis distance, Manhattan distance, and Chebyshev distance.

6. The life prediction system for the prediction target tool according to any one of claims 1 to 5, wherein, In a category of multiple tools described above, multiple learned models are generated by generating one learned model from one of the tools described above. The aforementioned learning-completed model storage unit is configured to store the learning-completed models for each of the aforementioned tools, treating them as the same type and each of the aforementioned tools having a different tool lifespan, and to store the learning-completed models for each of the multiple tool lifespans.

7. The life prediction system for the prediction target tool according to any one of claims 1 to 5, wherein, The aforementioned machining machine body is configured to use a rotary tool to cut or grind the aforementioned workpiece. The aforementioned status data includes drive load data in the motor that drives the aforementioned rotary tool.

8. The lifespan prediction system for the predictive object tool according to claim 7, wherein, The aforementioned machining machine body is configured to perform cutting or grinding of the workpiece while rotating it using the aforementioned rotary tool. The aforementioned status data also includes drive load data in the motor that rotates the aforementioned workpiece.

9. The life prediction system for the prediction target tool according to any one of claims 1 to 5, wherein, The aforementioned tool life prediction system also includes a display device that displays the predicted second remaining number of machining operations.

10. The lifespan prediction system for the predictive object tool according to claim 9, wherein, The aforementioned display device is configured to also display the aforementioned learning completion model, which is used to predict the second remaining number of processing steps, among a plurality of the aforementioned learning completion models.

11. The lifespan prediction system for the prediction target tool according to any one of claims 1 to 5, wherein, The aforementioned tool life prediction system also includes a model generation unit, which is configured to generate the plurality of learned models for each of the plurality of tool lifes by performing machine learning on the aforementioned training dataset containing the aforementioned explanatory variables and the aforementioned objective variables.

Images