Machining state inference device, machining state inference method, machining state inference program, training method, and trained model

Abstract

This machining state inference device infers the machining state indicating at least one state of a workpiece or a tool of a machine tool. The machining state inference device comprises a processor. The processor acquires a machining waveform indicating a machining load by the machine tool, and divides the machining waveform to acquire a plurality of section waveforms. The processor determines a parameter representing an approximate curve of the machining waveform by approximating each of the section waveforms, and infers the machining state on the basis of the determined parameter.

Description

Machining state estimation device, machining state estimation method, machining state estimation program, learning method, and trained model

[0001] The present disclosure relates to a machining state estimating device, a machining state estimating method, a machining state estimating program, a learning method, and a trained model.

[0002] Patent Document 1 discloses a technology in which a data logger device installed in a machine tool periodically collects various data from a numerical control device and predicts abnormalities through machine learning. The data logger device in Patent Document 1 classifies feature vectors created based on the coefficients of an approximate polynomial through clustering, while also deriving an approximate polynomial for currently collected data through regression analysis, determining which cluster the feature vector belongs to, and further scoring and ranking abnormalities that have occurred within the cluster to which the feature vector belongs, and notifying the user in the order of ranking.

[0003] Japanese Patent Application Laid-Open No. 2018-97616

[0004] The present disclosure provides a machining state estimating device that can reduce the processing load for estimating the machining state of a machine tool.

[0005] A machining state estimating device according to one aspect of the present disclosure is a machining state estimating device that estimates a machining state indicating at least one state of a tool or a workpiece of a machine tool, and includes a processor, wherein the processor acquires a machining waveform that indicates a machining load applied by the machine tool, divides the machining waveform to obtain a plurality of section waveforms, determines parameters that represent an approximation curve of the machining waveform by approximating each of the plurality of section waveforms, and estimates the machining state based on the parameters.

[0006] A machining state estimation method according to one aspect of the present disclosure is a machining state estimation method for estimating, by a processor, a machining state indicating at least one state of a tool or workpiece of a machine tool, and includes the steps of: acquiring a machining waveform indicating a machining load applied by the machine tool; acquiring a plurality of section waveforms by dividing the machining waveform; determining parameters representing an approximation curve of the machining waveform by approximating the plurality of section waveforms; inputting the parameters into a trained model that has been machine-learned so as to output a machining state value indicating the machining state when parameters representing a curve related to a load are input; and estimating the machining state based on the machining state value output from the trained model.

[0007] A program according to one aspect of the present disclosure is a machining state estimating program for causing a processor to execute the machining state estimating method.

[0008] A learning method according to one aspect of the present disclosure is a learning method for generating a trained model that estimates a machining state that indicates at least one state of a tool or workpiece of a machine tool, and includes the steps of: preparing a training dataset that includes, as input information, parameters that represent a load-related curve that is at least partially straight and at least partially curved, and that includes, as correct information, a machining state value that indicates the machining state; and performing machine learning on the model using the training dataset.

[0009] A trained model according to one aspect of the present disclosure is a trained model that estimates a machining state indicating at least one state of a tool or workpiece of a machine tool, learns the relationship between a learning parameter that represents a load-related curve that is at least partially straight and at least partially curved, and a machining state value that indicates the machining state, and outputs the machining state value when a parameter that represents the load-related curve is input.

[0010] According to the present disclosure, the load of processing for estimating the machining state of a machine tool can be reduced.

[0011] 5 is a block diagram showing an example of the configuration of a machining state estimation device according to a first embodiment of the present disclosure. FIG. 6 is a schematic cross-sectional view showing a press machine to which the load sensor and distance sensor shown in FIG. 1 are attached. FIG. 7 is a schematic graph showing an example of a measurement waveform according to the first embodiment using a load sensor and a distance sensor. FIG. 8 is a flowchart illustrating a procedure for estimating a machining state executed by a CPU of the machining state estimation device of FIG. 1. FIG. 9 is a graph showing an approximation curve obtained by approximating a machining waveform with a straight line and a curve. FIG. 10 is a diagram showing a cross-sectional state obtained by a simulation of the punching process shown in FIG. 2. FIG. 11 is a diagram showing a cross-sectional state obtained by a simulation of the punching process shown in FIG. 2. FIG. 12 is a diagram showing a cross-sectional state obtained by a simulation of the punching process shown in FIG. 2. FIG. 13 is a diagram showing a cross-sectional state obtained by a simulation of the punching process shown in FIG. 2. FIG. 14 is a diagram showing a cross-sectional state obtained by a simulation of the punching process shown in FIG. 2. FIG. 15 is an enlarged view of a portion of the approximation curve shown in FIG. 5. FIG. 16 is a graph showing a plurality of load waveforms obtained by changing a clearance in a punching simulation. FIG. 17 is a graph showing a plurality of load waveforms obtained by changing a punch wear amount in a punching simulation. 15 is a graph showing a plurality of load waveforms obtained by changing the die wear amount in a punching simulation. 16 is a graph showing a plurality of load waveforms obtained by changing the workpiece thickness in a punching simulation. 17 is a flowchart explaining a learning operation executed by the learning device of FIG. 9, which is a block diagram showing an example configuration of a learning device according to a first embodiment of the present disclosure. 18 is a schematic cross-sectional view showing an initial state of a press machine according to a second embodiment of the present disclosure. 19 is a schematic cross-sectional view showing an operating state of a press machine according to a second embodiment of the present disclosure. 20 is a schematic graph showing an example of a waveform measured by a load sensor and a distance sensor according to the second embodiment. 21 is a graph showing an approximation curve obtained by approximating the machining waveform shown in FIG. 13 with straight lines and curves for each section. 22 is a block diagram showing an example configuration of a machining state estimation device according to a first modified example. 23 is a diagram showing an example data configuration of the reference parameter DB of FIG. 25. 24 is a flowchart illustrating an example procedure of a machining state estimation process according to a first modified example.FIG. 13 is a block diagram showing a configuration example of a machining state estimating system according to an eighth modified example.

[0012] (Findings that form the basis of the present disclosure) In machining using machine tools, the state of the tool and the workpiece is reflected in the specific work performed using the tool to make a product. For example, it is known that the state of the tool and the workpiece appears in the machining waveform obtained by sensing the machining point.

[0013] For example, in press working, it is conceivable to estimate the machining state value from the machining waveform obtained by sensing the load applied to the tool as the machining point acting to make the product. When the machining waveform obtained by one cycle of machining is divided into multiple sections depending on the tool state and the workpiece state, the slope of the machining waveform within each section generally differs.

[0014] Therefore, it is conceivable to estimate the machining state from the change in the slope of the machining waveform. If the machining state can be estimated, it will be possible to know the appropriate timing for tool maintenance. Furthermore, in order to improve the stability of machining quality, it is preferable to be able to know the state value of the workpiece being machined.

[0015] Here, the "tool" in press working refers to at least one of a punch and a die. The "working state value" refers to at least one of the "tool state value" and "workpiece state value" used in working. The "tool state value" refers to at least one of the tool wear amount and clearance, and the "workpiece state value" refers to at least one of the workpiece thickness, workpiece hardness, and workpiece working shape.

[0016] The load applied to the punch or workpiece during machining depends on the values ​​of the punch wear amount, die wear amount, clearance, workpiece thickness, workpiece hardness, and workpiece machining shape, etc.

[0017] The punch wear amount is an example of a punch wear parameter that is an index showing the degree of wear of the punch, and the die wear amount is an example of a die wear parameter that is an index showing the degree of wear of the die. Tool wear amounts such as punch wear amount and die wear amount are expressed, for example, as dimensional changes from the design values ​​of the tool. Tool wear amount may also be expressed as the amount of change such as shape change, volume change, or mass change. Furthermore, tool wear amount may also be expressed as the radius of an arc when wear is approximated as an arc.

[0018] The clearance is the gap between the die and the punch. For example, the clearance is the gap between the die and the punch when punching a hole in a workpiece. The clearance may be expressed as the ratio of the gap between the die and the punch to the thickness of the workpiece.

[0019] The work thickness is the thickness of the material to be processed, and is measured as a length using, for example, a micrometer, a laser displacement meter, or the like, and its unit is, for example, μm.

[0020] The workpiece hardness is a value indicating the hardness of the workpiece material, and is measured as Vickers hardness by, for example, a Vickers hardness test, and its unit is, for example, HV.

[0021] The workpiece shape is a value that indicates the shape of the workpiece after processing, such as the size of burrs when punching the workpiece.

[0022] Since the machining waveform obtained by sensing the machining point in cycle machining depends on these parameters, it is possible to estimate the amount of punch wear, the amount of die wear, the clearance, the workpiece thickness, the workpiece hardness, and the machined shape of the workpiece from the machining waveform. For example, if the tool condition can be estimated from changes in the slope of the machining waveform, it will be possible to know the appropriate time for maintenance.

[0023] Furthermore, if the workpiece condition can be estimated inline, for example, for each processing cycle by a press machine, there is no need to measure the workpiece condition using a measuring device installed in a location separate from the press machine. The workpiece condition measured in this way can not only be used for traceability, but also can be used for process management by observing its trends.

[0024] In addition, if the workpiece condition can be estimated for each processing cycle, fluctuations in the workpiece condition within a lot and between lots can be managed or displayed as trends. The estimated workpiece condition can be used for status management of workpieces fed into a press, process management, and process improvement by analyzing them.

[0025] Furthermore, since the estimation of the tool status value and / or the workpiece status value varies due to various factors, it is preferable to perform the estimation for each machining cycle. However, the prior art does not disclose the estimation of the tool status value and / or the workpiece status value for each machining cycle.

[0026] The inventors have conducted extensive research to accurately estimate the tool state and workpiece state in cycle machining using a machine tool that repeats the same machining process, and as a result have discovered the following: By dividing the machining waveform obtained by sensing the machining point into multiple sections, the machining state can be estimated more accurately from changes in the slope of the machining waveform within each section.

[0027] The inventors have also found that by dividing the waveform as described above and linearly approximating the machining waveform in a portion of the section, it is possible to reduce the processing load and estimate the machining state with high accuracy.

[0028] Hereinafter, embodiments of the present disclosure will be described in detail with appropriate reference to the drawings. However, more detailed description than necessary may be omitted. For example, detailed description of already well-known matters or redundant description of substantially identical configurations may be omitted. This is to avoid unnecessary redundancy in the following description and to facilitate understanding by those skilled in the art. Note that the inventors provide the accompanying drawings and the following description to enable those skilled in the art to fully understand the present disclosure, and do not intend for them to limit the subject matter described in the claims.

[0029] 1 is a block diagram showing an example of the configuration of a machining state estimating device 100 according to a first embodiment of the present disclosure. The machining state estimating device 100 includes a CPU 1, a storage device 2, an input interface (I / F) 3, and an output interface (I / F) 4.

[0030] The CPU 1 performs information processing to realize the estimation processing of the machining state estimation device 100, which will be described later. Such information processing is realized, for example, by the CPU 1 operating in accordance with instructions of a program 21 stored in the storage device 2. The CPU 1 is an example of a processor of the present disclosure. The processor is not limited to a CPU as long as it includes an arithmetic circuit that performs calculations for information processing. For example, the processor may be configured by a circuit such as an MPU or FPGA.

[0031] The storage device 2 is a recording medium for recording various information including data such as the trained model 22 and a program 21 required to implement the estimation process of the machining state estimation device 100. The storage device 2 is realized, for example, by a semiconductor storage device such as a flash memory or a solid state drive (SSD), a magnetic storage device such as a hard disk drive (HDD), or other recording media, either alone or in combination. The storage device 2 may include a volatile memory such as an SRAM or a DRAM.

[0032] The trained model 22 is a model that has been trained to output the machining state value when parameters representing a curve related to load are input. Such a model has a neural network structure, for example, a convolutional neural network (CNN). The model is trained by a learning device using, for example, an error backpropagation method. The trained model 22 will be described in detail later.

[0033] The input interface 3 is an interface circuit that connects the machining state estimating device 100 to an external device in order to input information such as detection results by the load sensor 11 and the distance sensor 12 to the machining state estimating device 100. Such external devices are, for example, the load sensor 11, the distance sensor 12, other information processing terminals, etc. The input interface 3 may be a communication circuit that performs data communication in accordance with an existing wired communication standard or wireless communication standard.

[0034] The output interface 4 is an interface circuit that connects the machining state estimating device 100 to an external output device in order to output information from the machining state estimating device 100. Such an output device is, for example, a display or another information processing terminal. The output interface 4 may be a communication circuit that performs data communication in accordance with an existing wired communication standard or wireless communication standard. The input interface 3 and the output interface 4 may be realized by similar hardware.

[0035] Fig. 2 is a schematic cross-sectional view showing a press 50 to which the load sensor 11 and distance sensor 12 shown in Fig. 1 are attached. For ease of explanation, Fig. 2 shows an X-axis, a Y-axis, and a Z-axis that are perpendicular to each other. The Z-axis indicates the vertical direction.

[0036] The press 50 is an example of a machine tool that performs cyclic machining, repeating the same machining process. The press 50 includes a bolster 51 and a slide 52 that repeatedly performs a cyclical up-and-down motion from the top dead center to the bottom dead center relative to the bolster 51. A die backing plate 61 is attached to the top of the bolster 51, and a die plate 62 is attached to the top of the die backing plate 61. The die plate 62 holds a die 63.

[0037] A punch backing plate 71 is attached to the bottom of the slide 52, and a punch plate 72 is attached to the bottom of the punch backing plate 71. The punch plate 72 grips the punch 73. The press 50 further includes a stripper plate 74. The stripper plate 74 is attached to fasteners such as bolts and the punch plate 72 or punch backing plate 71 via positioning guides such as posts (not shown), for example. The stripper plate 74 is biased downward by, for example, a compression spring, and has the function of guiding the punch 73 to keep the position of the punch 73 constant, the function of removing material adhering to the punch 73 after punching the workpiece 80, and / or the function of fixing the workpiece 80 during punching.

[0038] The load sensor 11 is installed, for example, between the punch 73 and the punch backing plate 71. The load sensor 11 is, for example, a piezoelectric force sensor, a semiconductor strain sensor, or an electric force sensor such as a strain gauge type, and measures the load applied to the punch 73 when the punch 73 punches out the workpiece 80.

[0039] The distance sensor 12 is installed on, for example, the die backing plate 61. The distance sensor 12 is, for example, an eddy current gap sensor or a laser displacement meter, and measures the distance to, for example, the punch plate 72 that faces it in the Z-axis direction.

[0040] Fig. 3 is a schematic graph showing an example of waveforms measured by the load sensor 11 and the distance sensor 12. The horizontal axis of the graph in Fig. 3 represents the position of the punch 73 relative to the initial position. The position of the punch 73 is measured by the distance sensor 12. The horizontal axis of the graph in Fig. 3 can also be said to represent the distance that the punch 73 has advanced in the negative direction of the Z axis.

[0041] The vertical axis of the graph in Fig. 3 represents the load measured by the load sensor 11. The graph in Fig. 3 shows a mountain-shaped waveform in which, during punching, a load begins to be applied to the workpiece 80, and therefore to the punch 73 and load sensor 11, from the moment the punch 73 moves down and makes contact with the workpiece 80, and the load suddenly decreases to near zero after the workpiece 80 has been punched out.

[0042] 1-2. Operation Fig. 4 is a flowchart illustrating the procedure of the machining state estimation process executed by the CPU 1 of the machining state estimation device 100 of Fig. 1 .

[0043] First, the CPU 1 acquires, from the load sensor 11 and the distance sensor 12, a load waveform that indicates the measurement results of the load and position applied to the load sensor 11 during press working by the press machine 50 (S1).

[0044] Next, based on the load waveform obtained in step S1, CPU 1 determines a machining waveform corresponding to the start point to the end point of one cycle of cyclic machining (for example, from the punching start position to the punching end position in the case of punching) (S2). Referring to FIG. 3, for example, CPU 1 calculates the peak position where the load reaches its peak value, and searches for load values ​​before and after the peak position to determine the start and end points of the waveform. CPU 1 determines the load waveform from the determined start point to the end point as machining waveform W1. Machining waveform W1 corresponds to a waveform obtained by sensing a machining point in a specific operation (for example, punching) performed using a tool to make a product.

[0045] Alternatively, the CPU 1 may determine a load waveform included in a preset position range as the processed waveform W1.

[0046] Next, the CPU 1 divides the processed waveform W1 obtained in step S2 into a plurality of sections (S3), and performs fitting of an approximate curve in steps S4 and S5. Steps S3 to S5 will be described below with reference to FIGS.

[0047] 5 is a graph showing an approximation curve C1 obtained by approximating the processed waveform W1 with straight lines and curves. Fig. 5 shows the multiple sections A1, A2, A3, A4, A5, B12, B23, B34, and B45 divided in step S3. The approximation curve C1 is made up of straight lines L1 to L5 and curves M12 to M45, which will be described later.

[0048] The processed waveform W1 is a waveform based on time-series data measured by the load sensor 11 and the distance sensor 12. In step S3, the CPU 1 divides the processed waveform W1 into a plurality of sections, for example, by dividing the time-series data into sections each having a predetermined time width. Alternatively, in step S3, the CPU 1 may divide the processed waveform W1 into a plurality of sections by dividing the processed waveform W1 into sections each having a predetermined ratio to the width of the processed waveform W1 (the distance from the start point to the end point).

[0049] In sections A1, A2, A3, A4, and A5, the processed waveform W1 is approximated by a straight line as described below, and therefore sections A1, A2, A3, A4, and A5 are referred to as "straight line sections." In sections B12, B23, B34, and B45, the processed waveform W1 is approximated by a curve, and therefore sections B12, B23, B34, and B45 are referred to as "curved line sections."

[0050] The machining waveform obtained from machining by a machine tool includes sections that can be partially approximated as straight lines. For example, these include sections where monotonic states corresponding to the state during machining, such as monotonically maintaining, monotonically increasing, or monotonically decreasing levels, are continuous. Therefore, it is desirable to determine the straight line sections A1 to A5 so that they correspond to sections where monotonic states are continuous, i.e., sections that can be approximated as straight lines.

[0051] On the other hand, the processed waveform also includes sections that can be partially approximated as curves. For example, it includes sections that continuously change from at least one of monotonically maintaining, monotonically increasing, monotonically decreasing, etc. the level to at least one of monotonically maintaining, monotonically increasing, monotonically decreasing, etc. the level with a different slope. Therefore, it is desirable that the curve sections B12, B23, B34, and B45 be determined to correspond to continuously changing sections, i.e., sections that can be approximated as curves.

[0052] Figures 6A to 6E are diagrams showing the state of a cross section obtained by simulating the punching process shown in Figure 2. Figure 6A shows the state of the workpiece 80 in the straight line section A1 (straight line L1), Figure 6B shows the state of the workpiece 80 in the straight line section A2 (straight line L2), Figure 6C shows the state of the workpiece 80 in the straight line section A3 (straight line L3), Figure 6D shows the state of the workpiece 80 in the straight line section A4 (straight line L4), and Figure 6E shows the state of the workpiece 80 in the straight line section A5 (straight line L5).

[0053] There are various phases in the changes of the workpiece during machining. For example, Fig. 6A shows the phase in which the tool and workpiece come into contact, and Fig. 6B shows the phase in which the workpiece is elastically deformed. Fig. 6C shows the phase in which the workpiece is pulled mainly by the tool and plastically deformed by stress. Fig. 6D shows the phase in which the workpiece is plastically deformed by shear stress. The shear stress is mainly applied to the workpiece by the tool. Fig. 6E shows the phase in which the workpiece is broken.

[0054] 6A to 6E, the change in the processed waveform W1 is observed as a change that can be approximated by a straight line. In contrast, between each phase (for example, between the phase in FIG. 6A and the phase in FIG. 6B), the change in the processed waveform W1 is observed as a change in the slope of the corresponding straight line before and after the phase.

[0055] Therefore, it is preferable that the straight line sections A1 to A5 are set to correspond to phases in a series of changes in the workpiece caused by machining. For example, the straight line section A1 is set to correspond to the contact phase between the tool and the workpiece, the straight line section A2 to the elastic deformation phase of the workpiece, the straight line section A3 to the plastic deformation phase mainly caused by tensile stress, the straight line section A4 to the plastic deformation phase mainly caused by shear stress, and the straight line section A5 to the fracture phase of the workpiece.

[0056] 4, after dividing the processed waveform W1 in step S3, the CPU 1 determines initial values ​​of the parameters of the approximation curve (S4). In this embodiment, in step S4, the CPU 1 approximates the processed waveform W1 in the straight line sections A1, A2, A3, A4, and A5 by straight lines.

[0057] In step S4, the CPU 1 approximates linear equations representing the lines L1, L2, L3, L4, and L5 in the linear sections A1, A2, A3, A4, and A5, respectively. Specifically, for each section, the CPU 1 approximates the intercept and slope of the linear equation so that it corresponds to the processed waveform W1 obtained in step S1. For example, for the line L2 corresponding to section A2, the load indicating the starting point of the line L2 is obtained from the processed waveform W1 obtained in step S1, thereby approximating the intercept of the line L2. Furthermore, the load indicating the end point of the line L2 is similarly obtained from the processed waveform W1, and the slope of the line L2 is approximated from the starting point and the end point. In this way, the intercept and slope of the line L2 are approximated, and the linear equation for the line L2 is obtained. While the explanation has been given for the line L2, the same applies to the lines L1, L3, L4, and L5. The parameters representing the approximate curve C1 are composed of the intercepts and slopes representing the lines L1 to L5.

[0058] The tangents and slopes of the approximated straight lines L1 to L5 are determined as the initial values ​​of the parameters that represent the approximate curve C1.

[0059] In this embodiment, the tangents and slopes of the approximated straight lines L1 to L5 are calculated from the machining waveform W1, but the present disclosure is not limited to this. For example, the CPU 1 may set the parameters representing the previous approximate curve C1 in cycle machining as the initial values, or may set values ​​given by settings as the initial values.

[0060] Next, the CPU 1 adjusts parameters representing the approximate curve C1 obtained in step S4 and fits the approximate curve C1 to the processed waveform W1 (S5). In this embodiment, the processed waveform W1 in the curve sections B12, B23, B34, and B45 is approximated by a curve.

[0061] First, the CPU 1 calculates approximate curve equations that represent the curves M12, M23, M34, and M45 in the curve sections B12, B23, B34, and B45, respectively. For example, the CPU 1 represents the curves M12, M23, M34, and M45 as Bezier curves that are obtained from the straight lines L1 to L5 located at both ends of each curve.

[0062] 7 is an enlarged view of only the straight line section A2, the curved line section B23, and the straight line section A3 of the approximate curve C1 shown in FIG. P23 ​​is a control point of the Bézier curve determined by the straight lines L2 and L3, and is set, for example, at the intersection of the straight lines L2 and L3. Curve M23 is a Bézier curve to which control point P23 is assigned, and the starting point of curve M23 coincides with the end point of line L2, and the end point of curve M23 coincides with the starting point of line L3.

[0063] The slope of the Bezier curve at the start point and the end point coincides with the slope of the line connecting the control point adjacent to the start point and the end point. Therefore, by setting the control point P23 to be on the extension of both the lines L2 and L3, the curve M23 is expressed as a curve tangent to the lines L1 and L2. While the curve approximation for the curve M23 has been described in Figure 7, the approximation curves for the other curves in Figure 5, i.e., the curves M12, M34, and M45, can also be obtained in a similar manner.

[0064] In this embodiment, the curves M12, M23, M34, and M45 are represented as Bezier curves having only one control point, but the present disclosure is not limited to this. For example, the CPU 1 may determine the curves M12, M23, M34, and M45 by setting multiple control points. Alternatively, the CPU 1 may determine the curves M12 to M45 as n-th degree curves (n≧1).

[0065] In step S5, the CPU 1 fits parameters so that the approximation curve C1 closely approximates the processed waveform W1. For example, the parameters are fitted using the Levenberg-Marquardt (LM) method. Alternatively, the CPU 1 may use the Gauss-Newton method or the gradient descent method, or a method that combines multiple fitting algorithms, for fitting the parameters.

[0066] The approximation curve C1 is represented by a repetition of straight lines and nonlinear curves. Therefore, a fitting algorithm similar to the nonlinear least squares method may be used to fit the entire approximation curve C1 to the processed waveform W1.

[0067] Returning to FIG. 4, the CPU 1 inputs the parameters representing the approximate curve C1 determined in step S5 into the trained model 22 (S6).

[0068] The principle of estimating the processing state will be explained using Figures 8A to 8D, which are graphs showing load waveforms obtained by a punching simulation.

[0069] 8A is a graph showing load waveforms obtained by a simulation in which the punch wear amount was 3 μm, the die wear amount was 3 μm, the workpiece thickness was 35 μm, and the workpiece hardness was 600 HV. In the simulation of FIG. 8A, the clearance was varied. The graph of FIG. 8A shows the simulation results for clearances of 3 μm, 5 μm, and 10 μm. The arrows in FIG. 8A indicate the direction in which the load waveform changes when the clearance is increased.

[0070] 8B is a graph showing the load waveform obtained by a simulation in which the die wear amount was 3 μm, the clearance was 3 μm, the workpiece thickness was 35 μm, and the workpiece hardness was 600 HV. In the simulation of FIG. 8B, the punch wear amount was varied. The graph of FIG. 8B shows the simulation results in which the punch wear amount was 3 μm, 10 μm, 15 μm, and 25 μm. The arrows in FIG. 8B indicate the direction in which the load waveform changes when the punch wear amount is increased.

[0071] 8C is a graph showing the load waveform obtained by a simulation in which the punch wear amount was 3 μm, the clearance was 3 μm, the workpiece thickness was 35 μm, and the workpiece hardness was 600 HV. The die wear amount was varied in the simulation of FIG. 8C. The graph in FIG. 8C shows the simulation results in which the die wear amount was 3 μm, 10 μm, 15 μm, and 25 μm. The arrows in FIG. 8C indicate the direction in which the load waveform changes as the die wear amount increases.

[0072] FIG. 8D is a graph showing the load waveform obtained by a simulation in which the punch wear amount was 3 μm, the die wear amount was 3 μm, the clearance was 3 μm, and the workpiece hardness was 600 HV. The workpiece thickness was varied in the simulation of FIG. 8D. The graph in FIG. 8D shows the simulation results for workpiece thicknesses of 32 μm, 35 μm, and 38 μm. The arrows in FIG. 8D indicate the direction in which the load waveform changes when the workpiece thickness is increased or decreased.

[0073] By performing the processes corresponding to steps S1 to S5 in FIG. 4 on each of the load waveforms in FIGS. 8A to 8D, parameters representing the approximate curve C1 corresponding to each load waveform can be obtained.

[0074] As described above, the simulation provides parameters representing the approximation curve C1 corresponding to each of the load waveforms shown in Figures 8A to 8D, as well as the punch wear amount, die wear amount, clearance, workpiece thickness, and workpiece hardness corresponding to each parameter. The relationships between the parameters representing the approximation curve C1 and the punch wear amount, die wear amount, clearance, workpiece thickness, and workpiece hardness may be obtained by a simulation different from the simulation described above, or by actual measurement. The relationships between the parameters representing the approximation curve C1 obtained in this manner and the punch wear amount, die wear amount, clearance, workpiece thickness, and workpiece hardness can be used as training data (training dataset) for supervised learning of the trained model 22.

[0075] The trained model 22 is a model that has been trained through supervised learning using the training data described above to estimate at least one machining state value when parameters representing the approximation curve C1 are input.

[0076] As described above, in step S6, the CPU 1 inputs the parameters representing the approximate curve C1 determined in step S5 into the trained model 22.

[0077] Next, the CPU 1 estimates the machining state by acquiring the machining state value output from the trained model 22 (S7).

[0078] After step S7, the CPU 1 may display the machining state value for the current machining on a display device such as a liquid crystal display, etc. This allows the user to refer to the machining state value and perform maintenance or adjust the machining conditions as necessary.

[0079] 2. Learning Device 2-1. Configuration FIG. 9 is a block diagram showing an example configuration of a learning device 200 that performs machine learning to generate a trained model 22.

[0080] The learning device 200 includes a CPU 201 , a storage device 202 , an input interface 203 , and an output interface 204 .

[0081] The CPU 201 performs information processing and performs machine learning operations. Such information processing is realized, for example, by the CPU 201 operating in accordance with instructions from a program 205 stored in the storage device 202. The CPU 201 is an example of a processor of the present disclosure. The processor is not limited to a CPU as long as it includes an arithmetic circuit that performs calculations for information processing. For example, the processor may be configured by a circuit such as an MPU or FPGA.

[0082] The storage device 202 is a recording medium that records various information including data such as a model 206 to be learned and a program 205. The storage device 202 may store a trained model 22 generated by performing a learning operation on the model 206. The storage device 202 is realized, for example, by a semiconductor storage device, a magnetic storage device, or other recording medium, either alone or in combination. The storage device 202 may include a volatile memory such as an SRAM or a DRAM.

[0083] The input interface 203 is an interface circuit that connects the learning device 200 to an external device in order to input information such as a training data set into the learning device 200. The input interface 203 may be a communication circuit that performs data communication in accordance with an existing wired communication standard or wireless communication standard.

[0084] The output interface 204 is an interface circuit that connects the learning device 200 to an external device in order to output information from the learning device 200. Such an external device is, for example, the machining state estimation device 100. The output interface 204 may be a communication circuit that performs data communication in accordance with an existing wired communication standard or wireless communication standard. The input interface 203 and the output interface 204 may be realized by similar hardware.

[0085] 10 is a flowchart illustrating the learning operation executed by the learning device 200. The processing of this flowchart is executed by the CPU 201 of the learning device 200, for example.

[0086] First, the CPU 201 acquires a training data set (S11). The training data set is a data set that indicates the relationship between parameters that represent the approximation curve C1 and the punch wear amount, die wear amount, clearance, workpiece thickness, and workpiece hardness as correct answer information. As described above, the training data set may be obtained by simulation or actual measurement.

[0087] Next, the CPU 201 trains the model 206 on the training data set obtained in step S11 to generate a trained model 22 (S12).

[0088] The CPU 201 transmits the obtained trained model 22 to the machining state estimating device 100 (S13).

[0089] As a result, the machining state estimating device 100 can acquire the trained model 22 and use it to estimate the machining state.

[0090] 3. Effects, etc. As described above, the machining state estimating device 100 according to this embodiment includes a CPU 1 (an example of a processor) and estimates a machining state indicating at least one state of a tool of a machine tool or a workpiece 80. The CPU 1 acquires a machining waveform W1 indicating a machining load applied by the machine tool (S2). The CPU 1 divides the machining waveform W1 to obtain a plurality of section waveforms (S3). The CPU 1 determines parameters representing an approximation curve C1 of the machining waveform W1 by approximating each section waveform (S4, S5). The CPU 1 estimates the machining state based on the parameters representing the approximation curve C1.

[0091] According to this configuration, by dividing the machining waveform W1, the processing load for estimating the machining state of the machine tool can be reduced.

[0092] In this embodiment, the CPU 1 inputs parameters representing the approximate curve C1 to the trained model 22 that has been machine-learned to output a machining state value indicating the machining state when parameters representing a curve related to a load are input (S6). The CPU 1 estimates the machining state based on the machining state value output from the trained model 22 (S7).

[0093] According to this configuration, the processing state can be estimated with high accuracy while reducing the processing load.

[0094] In this embodiment, the CPU 1 determines parameters that represent the approximate curve C1 by approximating some of the multiple section waveforms with straight lines and approximating other parts of the multiple section waveforms with curves.

[0095] According to this configuration, by approximating parts of the waveforms of the plurality of sections with straight lines, it is possible to reduce the load of processing including approximation.

[0096] In this embodiment, a section including a section waveform approximated by a straight line and a section including a section waveform approximated by a curve are arranged adjacent to each other.

[0097] According to this configuration, the processing state can be estimated with high accuracy while reducing the processing load.

[0098] The learning method according to this embodiment is a learning method for generating a trained model 22 that estimates a machining state that indicates at least one state of a tool of a machine tool or a workpiece. The learning method includes a step (S11) of preparing a training data set that includes, as input information, parameters that represent a load-related curve that is at least partially linear and at least partially curved, and that includes, as correct answer information, a machining state value that indicates the machining state. The learning method also includes a step (S12) of performing machine learning on a model 206 using the training data set.

[0099] The trained model 22 according to the present embodiment estimates a machining state indicating at least one state of a tool of a machine tool or a workpiece. The trained model 22 is generated by learning the relationship between parameters representing a load-related curve, at least a portion of which is linear and at least a portion of which is curved, and a machining state value indicating the machining state.

[0100] Second Embodiment In the first embodiment, an example in which the machining state estimation method according to the present disclosure is applied to punching using a press machine 50 as shown in Fig. 2 has been described, but the machining state estimation method can also be applied to processes other than punching. In the second embodiment, an example in which a machining state value is estimated in bending will be described.

[0101] 11 and 12 are schematic cross-sectional views showing a press machine 53 to which the load sensor 11 and the distance sensor 12 according to this embodiment are attached. The press machine 53 raises and lowers a punch 73 to perform bending processing on a workpiece 80. Fig. 11 shows the press machine 53 in an initial state before the punch 73 is lowered, and Fig. 12 shows the press machine 53 in a state in which the punch 73 has lowered to the bottom dead center.

[0102] 13 is a schematic graph showing an example of waveforms measured by the load sensor 11 and the distance sensor 12 according to this embodiment. In this embodiment, the punch 73 descends from the top dead center to the bottom dead center, and then ascends.

[0103] In this embodiment, the CPU 1 determines a machining waveform W2 corresponding to the start point to the end point of one cycle of cyclic machining based on the measured waveform (corresponding to S2 in the first embodiment shown in FIG. 4). Referring to FIG. 13, for example, the CPU 1 determines the point at which load detection begins as the start point. If the CPU 1 detects that the load is zero or substantially zero for a predetermined period or more, or a predetermined number of times or more, after the start point, it determines the start point of the predetermined period as the end point of the machining waveform. Alternatively, the CPU 1 may determine the machining waveform W2 from a preset range of positions.

[0104] The CPU 1 divides the processed waveform W2 into a plurality of sections and performs fitting of an approximate curve. Fig. 14 is a graph showing an approximate curve C2 obtained by approximating the processed waveform W2 shown in Fig. 13 with straight lines and curves for each section. Similar to the approximate curve C1, the approximate curve C2 has straight lines L1, L3, L4, and L5 and curves M12, M23, M34, and M45. The method of fitting the approximate curve is the same as in the first embodiment (see S4 and S5 in Fig. 4).

[0105] In this embodiment, the lines L1, L3, L4, and L5 correspond to various phases that indicate changes in the state of the workpiece during bending. For example, the line L1 corresponds to the elastic deformation of the workpiece, the line L2 corresponds to the plastic deformation that mainly bends the workpiece, the line L3 corresponds to the completion of the plastic deformation, the line L4 corresponds to the sliding resistance between the punch and the workpiece as it is pushed down, and the line L5 corresponds to the sliding resistance between the punch and the workpiece as it is raised.

[0106] The CPU 1 inputs parameters representing the approximate curve C2 into the trained model 22 and obtains the machining state value output from the trained model 22, thereby estimating the machining state.

[0107] According to this embodiment, it is possible to estimate the machining state value of the machine tool that performs bending. For example, by training the trained model 22 to estimate the bending angle corresponding to the parameters representing the approximate curve C2, the CPU 1 can estimate the bending angle from the parameters representing the approximate curve C2.

[0108] (Other Embodiments) As described above, the embodiments have been described as examples of the technology in the present disclosure. However, the technology in the present disclosure is not limited to these, and can be applied to embodiments in which modifications, substitutions, additions, omissions, etc. are made as appropriate. Furthermore, it is also possible to combine the components described in the above embodiments to create new embodiments. Therefore, below, examples of modified examples as other embodiments will be described.

[0109] (First Modification) In the above embodiment, an example has been described in which the machining state is estimated by inputting parameters representing the approximate curve C1 into the trained model 22 and acquiring the machining state value output from the trained model 22 (see S6 and S7 in FIG. 4 ). However, the use of the trained model 22 is not essential for the machining state estimation method according to the present disclosure.

[0110] Fig. 15 is a block diagram showing an example of the configuration of a machining state estimating device 100a according to Modification 1. Compared with the machining state estimating device 100 of Fig. 1, the storage device 2 of the machining state estimating device 100a of Fig. 15 stores a reference parameter database (DB) 23 instead of the trained model 22.

[0111] 16 is a diagram showing an example of the data configuration of the reference parameter DB 23. The reference parameter DB 23 stores parameters (hereinafter referred to as "reference parameters") that represent an approximation curve C1 for various values ​​of punch wear amount, die wear amount, clearance, and workpiece thickness, which are examples of machining state values. The parameters that represent the approximation curve C1 may be obtained by the above-mentioned simulation or by actual measurement.

[0112] Fig. 17 is a flowchart illustrating the procedure of the processing state estimation process according to this modification. Compared with the flowchart according to the first embodiment in Fig. 4, Fig. 17 includes step S6A for performing a coincidence determination process instead of step S6.

[0113] In the coincidence determination process S6A, the CPU 1 calculates the coincidence between each of the reference parameters (X1, X2, . . . ) in the reference parameter DB 23 and the parameters representing the approximate curve C1 determined in step S5.

[0114] Here, the degree of similarity is an index showing the degree of similarity between two waveforms, and may be, for example, the cosine similarity, Euclidean distance, or Manhattan distance between two waveforms during the punching period.

[0115] In the matching degree determination process S6A, the CPU 1 searches among the plurality of reference parameters for a reference parameter that has the highest degree of matching with the parameter representing the approximate curve C1 determined in step S5.

[0116] The CPU 1 estimates the processing state by obtaining the processing state value corresponding to the reference parameter with the highest degree of coincidence (S7).

[0117] (Second Modification) In the first embodiment, the processed waveform W1 is divided into a plurality of sections, and the width of each section corresponds to a predetermined time width related to the time-series data or a predetermined ratio to the processed waveform W1. In contrast, in the second modification, the width of each section is made variable. That is, in this modification, the width of at least one of the straight-line sections A1 to A5 and the curved sections B12, B23, B34, and B45 is a variable parameter rather than a fixed value.

[0118] In fitting the approximate curve C1, not only the tangents and slopes constituting the linear equations of the straight lines L1, L2, L3, L4, and L5, but also the widths of at least one of the straight line sections A1 to A5 and the curved line sections B12, B23, B34, and B45 are subject to fitting. For example, in the phases of the series of changes during machining shown in Figures 6A to 6E, it is possible that the phase changes may be shifted due to tool state values, etc.

[0119] According to the second variant, even if the position of each phase changes due to tool state values, etc., it is possible to construct straight sections A1 to A5 and curved sections B12, B23, B34, and B45 that correspond to more appropriate phases, and estimations can be made that correspond to a wider range of machining state values.

[0120] (Third Modification) In the first embodiment, an example of estimating a machining state value in a punching process was described, and in the second embodiment, an example of estimating a machining state value in a bending process was described. However, the machining state estimating device according to the present disclosure may be configured to estimate a machining state value in other plastic processing. For example, the machining state estimating device according to this modification may be configured to estimate a tool state value or a workpiece state value in a forging process, i.e., a shape after forging.

[0121] Furthermore, for example, the machining state estimating device according to this modified example may be configured to estimate the tool state values ​​and workpiece state values ​​in the drawing process, that is, the shape after the drawing process.

[0122] (Fourth Modification) In the above embodiment, an example has been shown in which a machining state value in one machining process is estimated. However, the machining state estimating device according to the present disclosure may be configured to estimate a machining state value in integrated machining in which a plurality of machining processes are combined.

[0123] Progressive dies are used in many press processes. On the progressive dies, processes such as punching, bending, forging, and drawing are combined, and the workpiece is fed forward in cycles, and a product is formed by passing through each process. Therefore, for example, if a product is subjected to a punching process and then a bending process, the subsequent bending process can use parameters representing the approximate curve obtained in the previous punching process.

[0124] For example, in the preceding punching process, if the amount of wear increases compared to when the amount of wear was small, the punch and / or die come into contact with and bite into the workpiece, resulting in an increase in sagging and / or burrs. In other words, the punched fracture surface has a difference in shape depending on the amount of wear.

[0125] On the other hand, for example, in the bending process, which is a subsequent process, the bending process is performed including sagging and / or burrs that occurred on the fracture surface of the workpiece in the previous process, and this can result in differences in the shape after the bending process, for example, the bending angle. Therefore, in the bending process, which is a subsequent process, parameters representing the approximate curve obtained in the punching process, which is a previous process, and parameters representing the approximate curve obtained in the bending process, which is a subsequent process, are combined to estimate the processing state value in an integrated manner, thereby making it possible to estimate the processing state value with higher accuracy.

[0126] In the state value estimation calculation in this modified example, the CPU 1 estimates the machining state value in the target process by combining a first parameter representing an approximation curve corresponding to the machining waveform of the target process for which the machining state value is to be estimated, and a second parameter representing an approximation curve corresponding to a machining process that precedes the target process and is sent sequentially to the target process.

[0127] For example, in this modified example, the training data set acquired in step S11 is a data set indicating the relationship between the first parameter, the second parameter, and the bending angle as correct answer information. In step S12 of generating a trained model 22 in this modified example, the CPU 201 trains the model 206 on the training data set acquired in step S11 to generate the trained model 22. In step S7 of estimating the machining state in this modified example, the CPU 201 estimates the machining state, i.e., the bending angle, by acquiring the machining state value output from the trained model 22. This configuration makes it possible to estimate the bending angle in a target process. In other words, the machining state estimation device according to this modified example can estimate a shape that depends on the machining state of the preceding punching process in a target process (subsequent process).

[0128] In this modified example, an example in which a plurality of machining steps using a progressive die is combined to estimate a machining state value has been described, but the present disclosure is not limited to this. For example, the plurality of machining steps in the estimation of the machining state value may each be configured using a different die. Furthermore, at least some of the plurality of machining steps may be a machining device that does not use a die.

[0129] Furthermore, in this modified example, an example of estimating the processing state values ​​for a plurality of processing steps that are a combination of a punching process as a preceding process and a bending process as a succeeding process has been described, but the present disclosure is not limited to this. For example, the plurality of processing steps for estimating the processing state values ​​may be a combination of two or more processes, such as punching, bending, forging, and drawing. For example, if a product undergoes a punching process and then two bending processes with different bending angles, i.e., if the product is bent to a predetermined angle in two separate steps, the final bending process, which is the target process, may use parameters representing the approximate curves obtained in the previous bending process, which bends the product to a different bending angle from the target process, and the punching process, which is the process before that. That is, the machining state estimation device according to this modified example may estimate a machining state value in a target process that depends on the machining states of the previous process and the process before that by combining a first parameter representing an approximation curve corresponding to the machining waveform of the target process with second and third parameters representing approximation curves corresponding to the processing processes that precede the target process and are sent sequentially to the target process.

[0130] According to this configuration, parameters from a processing step prior to the target step can be used to estimate the processing state value for the target step, thereby enabling more accurate estimation of the processing state value.

[0131] (Fifth Modification) In the above embodiment, an example has been shown in which an approximation curve is determined by approximating five straight lines and four curved lines. However, the number of straight lines and curved lines used for the approximation is not limited to this. For example, the machining state estimating device according to the present disclosure may be configured to determine an approximation curve by, for example, three straight lines and two curved lines in response to a change in the machining phase, and estimate the machining state.

[0132] (Sixth Modification) In the above embodiment, an example was shown in which the machining waveform was determined from a waveform indicating the load relative to the position. However, the machining waveform is not limited to a waveform indicating the relationship between the position and the load. Because the machining device performs cyclic machining in which the same machining is repeated, the machining waveform may be determined from a waveform indicating the load relative to time, for example. With this configuration, it is not necessary to measure the distance, and the machining state value can be estimated with a simpler configuration.

[0133] (Seventh Modification) In the above embodiment, an example has been shown in which a machining state value is estimated from a measured waveform of a load applied to a tool. However, the machining state estimating device according to the present disclosure may estimate the machining state value based on, for example, pressure, strain, torque, flow rate, or temperature, or may estimate the machining state value based on, for example, a voltage or current corresponding to the pressure, strain, torque, flow rate, or temperature.

[0134] With this configuration, for example, even if it is not possible to install the load sensor 11, it is possible to estimate the processing state value based on the pressure, strain, torque, flow rate, temperature, voltage, or current. Furthermore, in injection molding that performs cyclic processing, it is possible to estimate processing state values ​​such as the shape in injection molding by determining the processing waveform from the temperature measurement waveform.

[0135] (Eighth Modification) In the first embodiment, an example has been described in which the trained model 22 is stored in the storage device 2 of the machining state estimating device 100, but the trained model 22 may also be stored in an external device.

[0136] 18 is a block diagram showing an example of the configuration of a machining state estimating system 400 according to an eighth modified example. The machining state estimating system 400 includes a machining state estimating apparatus 100 and a server apparatus 300. The machining state estimating apparatus 100 and the server apparatus 300 are connected to each other so as to be able to communicate with each other via a network, for example, when necessary.

[0137] The server device 300 includes a CPU 301 , a storage device 302 , an input interface 303 , and an output interface 304 .

[0138] The CPU 301 performs information processing by operating in accordance with instructions from a program 305 stored in the storage device 302. The CPU 301 is an example of a processor of the present disclosure. The processor is not limited to a CPU as long as it includes an arithmetic circuit that performs calculations for information processing. For example, the processor may be configured with a circuit such as an MPU or FPGA.

[0139] The storage device 302 is a recording medium that records various information including the trained model 22 and the program 305. The storage device 302 is realized, for example, by a semiconductor storage device, a magnetic storage device, or other recording medium, either alone or in combination. The storage device 302 may include a volatile memory such as an SRAM or a DRAM.

[0140] The input interface 303 is an interface circuit that connects the server device 300 to an external device in order to input information such as parameters representing the approximate curve C1 to the server device 300. The output interface 304 is an interface circuit that connects the server device 300 to an external device in order to output information from the server device 300. Such an external device is, for example, the machining state estimating device 100. The input interface 303 and the output interface 304 may be communication circuits that perform data communication in accordance with an existing wired communication standard or wireless communication standard. The input interface 303 and the output interface 304 may be realized by similar hardware.

[0141] In the first embodiment, step S6 in Fig. 4 is executed by the machining state estimating device 100, whereas in this modification, step S6 in Fig. 4 is executed by the server device 300. Therefore, in this modification, it is possible to reduce the load on the machining state estimating device 100 related to the processing of estimating the machining state.

[0142] Furthermore, the functions of the server device 300 can be used by other devices besides the machining state estimating device 100. Any information processing device such as a general-purpose computer, smartphone, or tablet terminal can acquire the machining state value, which is the output of the trained model 22, from the server device 300 as long as it can transmit parameters representing the approximate curve C1 to the server device 300. This allows various users to use the machining state estimating function of the server device 300.

[0143] (Examples of Aspects) Examples of aspects of the present disclosure are described below.

[0144] <Aspect 1> A machining state estimating device that estimates a machining state indicating at least one state of a tool or a workpiece of a machine tool, comprising a processor, wherein the processor: acquires a machining waveform that indicates a machining load applied by the machine tool; divides the machining waveform to obtain a plurality of section waveforms; determines parameters that represent an approximation curve of the machining waveform by approximating each of the plurality of section waveforms; and estimates the machining state based on the parameters.

[0145] <Aspect 2> The machining state estimating device according to Aspect 1, further comprising a storage device for storing a plurality of reference parameters corresponding to a plurality of candidate values ​​for a machining state value indicating the machining state, wherein the plurality of reference parameters each represent a curve relating to load, and wherein the processor: determines a degree of agreement which is an index of the degree of agreement between each of the plurality of reference parameters and the parameter; searches for a reference parameter from the plurality of reference parameters which has the greatest degree of agreement; determines each of the candidate values ​​corresponding to the searched reference parameter as a machining state value indicating the machining state; and estimates the machining state based on the determined machining state values.

[0146] <Aspect 3> The machining state estimating device according to Aspect 1 or 2, wherein the processor inputs parameters representing a curve related to a load to a trained model that has been machine-learned to output a machining state value indicating the machining state when the parameters are input, and estimates the machining state based on the machining state value output from the trained model.

[0147] <Aspect 4> The machining state estimating device according to Aspect 2, further comprising a storage device, wherein the trained model is stored in the storage device.

[0148] <Aspect 5> The machining state estimating device according to Aspect 2 or 4, wherein the trained model is stored in an external server device.

[0149] <Aspect 6> The machining state estimating device according to any one of Aspects 1 to 5, wherein the processor determines the parameters by approximating at least one of the plurality of section waveforms by a straight line and approximating at least one other of the plurality of section waveforms by a curve.

[0150] <Aspect 7> The machining state estimating device according to Aspect 6, wherein the section to which the section waveform approximated by the straight line belongs and the section to which the section waveform approximated by the curve belong are arranged adjacent to each other.

[0151] <Aspect 8> The machining state estimating device according to any one of Aspects 1 to 7, wherein the machine tool is a press machine.

[0152] <Aspect 9> The machining state estimating device according to Aspect 1, wherein the machining state is estimated in machining consisting of a plurality of steps, and the processor: determines a first parameter representing an approximation curve of the machining waveform in a first step among the plurality of steps; determines a second parameter representing an approximation curve of the machining waveform in a second step among the plurality of steps that is performed before the first step; and estimates the machining state based on the first parameter and the second parameter.

[0153] <Aspect 10> The machining state estimating device according to Aspect 8, wherein the press machine uses a progressive die and performs machining consisting of a plurality of steps, and the processor: determines a first parameter representing an approximation curve of the machining waveform in a first step of the plurality of steps; determines a second parameter representing an approximation curve of the machining waveform in a second step of the plurality of steps that is performed before the first step; and estimates the machining state based on the first parameter and the second parameter.

[0154] <Aspect 11> A machining state estimating method for estimating a machining state indicating at least one state of a tool or a workpiece of a machine tool by a processor, the machining state estimating method comprising the steps of: acquiring a machining waveform indicating a machining load applied by the machine tool; acquiring a plurality of section waveforms by dividing the machining waveform; determining parameters representing an approximation curve of the machining waveform by approximating each of the plurality of section waveforms; inputting the parameters into a trained model that has been machine-learned so as to output a machining state value indicating the machining state when parameters representing a curve related to a load are input; and estimating the machining state based on the machining state value output from the trained model.

[0155] <Aspect 12> A machining state estimating program for causing a processor to execute the machining state estimating method according to aspect 11.

[0156] <Aspect 13> A learning method for generating a trained model that estimates a machining state that indicates at least one state of a tool or a workpiece of a machine tool, the learning method comprising: a step of preparing a training data set that includes, as input information, parameters that represent a load-related curve that is at least partially straight and at least partially curved, and that includes, as ground truth information, a machining state value that indicates the machining state; and a step of performing machine learning on the model using the training data set.

[0157] <Aspect 14> A trained model that estimates a machining state that indicates at least one state of a tool or a workpiece of a machine tool, the trained model learning the relationship between a learning parameter that indicates a load-related curve that is at least partially straight and at least partially curved, and a machining state value that indicates the machining state, and when the parameter that indicates the load-related curve is input, the trained model outputs the machining state value.

[0158] The present disclosure is applicable to a machining state estimation device, a machining state estimation method, a machining state estimation program, a learning method, and a trained model related to a machine tool.

[0159] 2 Storage device 3 Input interface 4 Output interface 11 Load sensor 12 Distance sensor 21 Program 22 Trained model 23 Reference parameter database 50 Press machine 51 Bolster 52 Slide 53 Press machine 61 Die backing plate 62 Die plate 63 Die 71 Punch backing plate 72 Punch plate 73 Punch 74 Stripper plate 80 Workpiece 100 Machining state estimation device 200 Learning device 201 CPU 202 Storage device 203 Input interface 204 Output interface 205 Program 206 Model 300 Server device 302 Storage device 303 Input interface 304 Output interface 305 Program 400 Machining state estimation system

Claims

1. A machining state estimation device that estimates a machining state that indicates the state of at least one of a tool or a workpiece of a machine tool, comprising a processor, wherein the processor: acquires a machining waveform that indicates the machining load applied by the machine tool; divides the machining waveform to obtain a plurality of section waveforms; determines parameters that represent an approximation curve of the machining waveform by approximating each of the plurality of section waveforms; and estimates the machining state based on the parameters.

2. A machining state estimating device according to claim 1, further comprising a storage device for storing a plurality of reference parameters corresponding to a plurality of candidate values ​​for a machining state value indicating said machining state, said plurality of reference parameters each representing a curve relating to load, said processor: determining a degree of agreement which is an index of the degree of agreement between each of said plurality of reference parameters and said parameter; searching for a reference parameter from said plurality of reference parameters which has the greatest degree of agreement; determining each of said candidate values ​​corresponding to said searched reference parameter as a machining state value indicating said machining state; and estimating said machining state based on said determined machining state values.

3. The machining state estimation device according to claim 1, wherein the processor inputs parameters representing a curve related to load into a trained model that has been machine-learned to output a machining state value indicating the machining state when the parameters are input, and estimates the machining state based on the machining state value output from the trained model.

4. The machining state estimation device according to claim 3, further comprising a storage device, wherein the trained model is stored in the storage device.

5. The machining state estimation device according to claim 3, wherein the trained model is stored in an external server device.

6. The machining state estimating device according to claim 1, wherein the processor determines the parameters by approximating at least one of the plurality of section waveforms by a straight line and approximating at least one other of the plurality of section waveforms by a curve.

7. The machining state estimating device according to claim 6, wherein the section to which the section waveform approximated by the straight line belongs and the section to which the section waveform approximated by the curve belong are arranged adjacent to each other.

8. The machining state estimation device according to claim 1, wherein the machine tool is a press machine.

9. The machining state estimating device according to claim 1, wherein the machining state estimating device estimates the machining state in machining consisting of a plurality of steps, and the processor determines a first parameter representing an approximation curve of the machining waveform in a first step among the plurality of steps, determines a second parameter representing an approximation curve of the machining waveform in a second step among the plurality of steps that is performed before the first step, and estimates the machining state based on the first parameter and the second parameter.

10. The machining state estimation device according to claim 8, wherein the press machine uses a progressive die and performs machining consisting of a plurality of steps, and the processor: determines a first parameter representing an approximation curve of the machining waveform in a first step of the plurality of steps; determines a second parameter representing an approximation curve of the machining waveform in a second step of the plurality of steps that is performed before the first step; and estimates the machining state based on the first parameter and the second parameter.

11. A machining state estimation method for estimating, by a processor, a machining state indicating at least one state of a tool or workpiece of a machine tool, comprising: a step of acquiring a machining waveform indicating the machining load applied by the machine tool; a step of acquiring a plurality of section waveforms by dividing the machining waveform; a step of determining parameters representing an approximation curve of the machining waveform by approximating each of the plurality of section waveforms; a step of inputting the parameters into a trained model that has been machine-learned so as to output a machining state value indicating the machining state when parameters representing a curve related to a load are input; and a step of estimating the machining state based on the machining state value output from the trained model.

12. A machining state estimating program for causing a processor to execute the machining state estimating method according to claim 11.

13. A learning method for generating a trained model that estimates a machining state that indicates at least one state of a tool or workpiece of a machine tool, the learning method comprising the steps of: preparing a training dataset that includes, as input information, parameters that represent a load-related curve that is at least partially straight and at least partially curved, and that includes, as correct answer information, a machining state value that indicates the machining state; and performing machine learning on the model using the training dataset.

14. A trained model that estimates a machining state that indicates at least one state of a tool or workpiece of a machine tool, the trained model learning the relationship between training parameters that represent a load-related curve that is at least partially straight and at least partially curved, and a machining state value that indicates said machining state, and outputs said machining state value when a parameter that represents the load-related curve is input.

Images