Tool damage probability estimation device, tool damage probability estimation method, and program

Abstract

To estimate a damage probability or a damage risk of a tool used to machine a workpiece.SOLUTION: A tool damage probability estimating device is a device for estimating a damage probability of a tool used for machining a workpiece, and comprises a first estimating unit and a second estimating unit. The first estimating unit estimates at least one damage evaluation parameter for evaluating a damage state of a tool by inputting a state quantity of the tool into a first learning model. The second estimating unit estimates a damage probability or a damage risk by inputting a damage evaluation parameter estimated by the first estimating unit into a second learning model.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present disclosure relates to a tool damage probability estimation device, a tool damage probability estimation method, and a program. [Background technology]

[0002] For example, when machining a workpiece using a tool under automatic control, the tool wears as the machining operation progresses. As this wear progresses, the tool may be damaged, which may result in problems such as scratches on the workpiece being machined.

[0003] A technology for predicting the amount of wear that occurs on a tool during machining is, for example, disclosed in Patent Document 1. This document discloses that a prediction model that defines the correlation between machining conditions and the amount of wear is constructed by machine learning, and the amount of wear that occurs on a tool is predicted using the model. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Publication No. 2019-139755 Summary of the Invention [Problem to be solved by the invention]

[0005] In the above Patent Document 1, the amount of wear that occurs on a tool during machining is predicted, but it is not possible to estimate whether the wear that occurs on the tool will actually cause damage to the tool (probability of tool damage).

[0006] At least one embodiment of the present disclosure has been made in consideration of the above-mentioned circumstances, and aims to provide a tool damage probability estimation device, a tool damage probability estimation method, and a program capable of estimating the damage probability or damage risk of a tool used to machine a workpiece. [Means for solving the problem]

[0007] In order to solve the above problem, a tool damage probability estimation device according to at least one embodiment of the present disclosure includes: A tool damage probability estimation device for estimating a damage probability or a damage risk of a tool used in machining a workpiece, comprising: a first estimation unit for estimating at least one damage assessment parameter for assessing a damage state occurring in the tool by inputting a state quantity of the tool into a first learning model; a second estimation unit for estimating the damage probability or the damage risk by inputting the at least one damage assessment parameter estimated by the first estimation unit into a second learning model; Equipped with.

[0008] In order to solve the above problem, a tool damage probability estimation method according to at least one embodiment of the present disclosure includes: A tool damage probability estimation method for estimating a damage probability or a damage risk of a tool used in machining a workpiece, comprising: a step of estimating at least one damage assessment parameter for assessing a damage state occurring in the tool by inputting a state quantity of the tool into a first learning model; estimating the damage probability or the damage risk by inputting the at least one damage assessment parameter into a second learning model; Equipped with.

[0009] In order to solve the above problem, a program according to at least one embodiment of the present disclosure includes: A program for estimating a damage probability or a damage risk of a tool used in machining a workpiece using a computer, a step of estimating at least one damage assessment parameter for assessing a damage state occurring in the tool by inputting a state quantity of the tool into a first learning model; estimating the damage probability or the damage risk by inputting the at least one damage assessment parameter into a second learning model; is possible. [Effects of the Invention]

[0010] According to at least one embodiment of the present disclosure, it is possible to provide a tool damage probability estimation device, a tool damage probability estimation method, and a program capable of estimating the damage probability or damage risk of a tool used to machine a workpiece. [Brief explanation of the drawings]

[0011] [Figure 1] 1 is a block diagram showing a hardware configuration of a tool damage probability estimation device 100 according to an embodiment. [Figure 2] 1 is a block diagram showing a functional configuration of a tool damage probability estimation device 100 according to an embodiment. [Figure 3] 1 is a flowchart illustrating a tool damage probability estimation method according to an embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0012] Hereinafter, several embodiments of the present disclosure will be described with reference to the accompanying drawings. However, the dimensions, materials, shapes, relative arrangements, etc. of components described as embodiments or shown in the drawings are merely illustrative examples and are not intended to limit the scope of the present disclosure.

[0013] A tool damage probability estimation device according to at least one embodiment of the present disclosure is a device for estimating the damage probability of a tool used in machining a workpiece. While the type of workpiece, the type of machining, and the type of tool are not limited, the following embodiment describes, as an example, a case in which a drill tool is used to drill a workpiece made of a metal material (e.g., a nickel-based heat-resistant alloy). In particular, the drill tool is automatically loaded onto a machining center (MC) during drilling and operated to drill the workpiece. During this process, the drill tool experiences considerable wear due to the load it receives during operation. As this wear progresses, it may be damaged depending on the machining conditions and the specifications of the workpiece. The tool damage probability estimation device is capable of estimating the probability of damage (damage probability) that may occur to such a tool.

[0014] (Hardware configuration) First, the hardware configuration of the tool damage probability estimation device will be described. The tool damage probability estimation device is configured by an arithmetic processing device such as a computer. FIG. 1 is a block diagram showing the hardware configuration of a tool damage probability estimation device 100 according to one embodiment. The tool damage probability estimation device 100 includes, as its hardware configuration, an input unit 110, a storage unit 120, a calculation unit 130, and an output unit 140.

[0015] The input unit 110 is configured to input various information necessary for the calculation processing performed in the tool damage probability estimation device 100. The input unit 110 may be a human interface such as a mouse, keyboard, or touch panel that can be operated by an operator, or may be an interface device for acquiring input information from another device.

[0016] The storage unit 120 is configured to store various types of information necessary for the arithmetic processing performed in the tool damage probability estimation device 100. The storage unit 120 is configured to include a computer-readable storage medium including at least one of a RAM (Random Access Memory) and a ROM (Read Only Memory). The various types of information stored in the storage unit 120 may include a program that enables this hardware configuration to function as the tool damage probability estimation device 100.

[0017] The calculation unit 130 is configured to perform various calculations of the tool damage probability estimation device 100, and is configured to include, for example, a CPU (Central Processing Unit). The calculation unit 130 reads out the programs stored in the storage unit 120 into a RAM or the like, and executes information processing and calculation processing, thereby realizing various functions of the tool damage probability estimation device 100.

[0018] The program executed by the calculation unit 130 may be stored in the storage unit 120 as described above, or may be pre-installed in a ROM or other storage medium, provided in a state stored in a computer-readable storage medium, or distributed via wired or wireless communication means, etc. Examples of computer-readable storage media include magnetic disks, magneto-optical disks, CD-ROMs, DVD-ROMs, and semiconductor memories.

[0019] The output unit 140 is configured to perform output based on the calculation result in the calculation unit 130. In this embodiment, the output unit 140 outputs the tool damage probability based on the calculation result in the calculation unit 130, and the operator can take appropriate measures and make preparations to deal with problems that may occur in the machining work due to damage to the tool based on the damage probability.

[0020] (Functional configuration) Next, a description will be given of the functional configuration of the tool damage probability estimation device 100. Fig. 2 is a block diagram showing the functional configuration of the tool damage probability estimation device 100 according to one embodiment. The tool damage probability estimation device 100 includes a state quantity acquisition unit 150, a specific information acquisition unit 160, a first estimation unit 170, and a second estimation unit 180.

[0021] The block diagram shown in FIG. 2 is an example showing the functional configuration of the tool damage probability estimation device 100 in accordance with the following description, and the blocks may be integrated with each other or further subdivided.

[0022] The state quantity acquisition unit 150 is configured to acquire the state quantities of the tool required for the calculations of the tool damage probability estimation device 100. The state quantities of the tool broadly include parameters related to factors that may affect the feature quantities of the tool when the tool is used for machining, and are input as appropriate by the above-mentioned input unit 110 as parameters that have been subjected to dimension reduction (normalized regression) in advance. Such parameters are not limited, but some specific examples include various parameters related to machining conditions (conditions set before machining a workpiece), machining noise, machining temperature, tool type, and machining environment. More specifically, the state quantities can include the material properties of the workpiece (hardness, tensile strength, crystalline state, characteristics of the machined part (dimensions, residual stress, machined layer, etc.)), characteristics of the machining center on which the tool is mounted (type, operating accuracy (axis, table, inspection interval, deterioration over time, etc.)), tool characteristics (material, shape, thinning, dimensions, tip angle, helix angle, groove length, diameter back taper, core thickness, margin, etc.), machining conditions (machining depth, machining diameter, feed rate, spindle rotation speed, coolant concentration and temperature, feed per revolution, machining time, cutting speed, etc.), and operator characteristics (tool setup skills, part setup skills, jig setup skills, etc.).

[0023] The unique information acquisition unit 160 is configured to acquire tool unique information required for the calculations of the tool damage probability estimation device 100. The tool unique information is information for identifying the tool used to machine the workpiece or the machine tool (machining center, lathe, etc.) on which the tool is mounted, and includes, for example, the type and specifications of the tool or machine tool. Specifically, the unique information corresponding to a drill tool includes the material supplier, the rough machining process, the number of holes that can be machined, and the L / D (length to diameter ratio) of the drill.

[0024] The first estimation unit 170 is configured to estimate at least one damage assessment parameter P using at least one first learning model M1. The damage assessment parameter P is a parameter for assessing the damage state of a tool. Generally, there are multiple factors that cause a tool to be damaged, and each factor affects the tool's lifespan. The first learning model M1 may have multiple damage assessment parameters P as objective variables, and in this embodiment, the multiple damage assessment parameters include a first damage assessment parameter P1 corresponding to chipping of the tool, a second damage assessment parameter P2 corresponding to burning, and a third damage assessment parameter P3 corresponding to the amount of wear.

[0025] The first damage assessment parameter P1, which refers to tool chipping, is a damage assessment parameter that is significantly affected when the tool material is hard, the feed rate is high, or the cutting edge strength is insufficient. The second damage assessment parameter P2 is a parameter that affects chipping due to embrittlement caused by cutting heat, such as hardening. The third damage assessment parameter P3 is a parameter related to the induction of cutting load and the resulting damage. In this way, the first learning model M1 has multiple damage assessment parameters P as objective variables, allowing for multifaceted damage assessment from various perspectives compared to, for example, when only the amount of wear is used as a single damage assessment parameter P.

[0026] The first learning model M1 is a computational model that shows the correlation between explanatory variables including state quantities and a target variable that is a damage assessment parameter P, and is constructed in advance using training data. In this embodiment, the first estimation unit 170 has multiple first learning models M1. The multiple first learning models M1 are prepared to correspond to the unique information acquired by the unique information acquisition unit 160. In other words, each first learning model M1 and the unique information are associated in advance, and can be selected based on the unique information acquired by the unique information acquisition unit 160.

[0027] The multiple first learning models M1 prepared to correspond to the unique information are constructed, for example, by different methods. Fig. 2 shows multiple first learning models M1 constructed by different methods A, B, .... Examples of methods that can be used to construct such first learning models M1 include regression analysis (simple regression analysis, multiple regression analysis, etc.), neural networks, random forests, and support vector machines (SVMs).

[0028] Furthermore, even if the multiple first learning models M1 prepared to correspond to unique information are constructed using the same method, they may handle different explanatory variables (for example, if the first learning model M1 is defined as Y = a x1 + b x2 + c, the explanatory variables x1 and x2) or different coefficients (for example, if the first learning model M1 is defined as Y = a x1 + b x2 + c, the coefficients a, b, c). In the embodiment of Figure 2, an example is shown in which multiple first learning models M1 are constructed using different methods A, B, ..., but the same idea can be applied even if the construction method is the same and the explanatory variables and coefficients are different in this way.

[0029] The first learning model M1 having such explanatory variables and objective variables is constructed by performing machine learning in advance using training data. The first estimation unit 170 can estimate the damage assessment parameter P by inputting the state quantities acquired by the state quantity acquisition unit 150 to the first learning model M1 constructed in this manner. Furthermore, the input to the first learning model M1 may include, in addition to the state quantities acquired by the state quantity acquisition unit 150, unique information acquired by the unique information acquisition unit 160. In this case, the unique information can be converted into some kind of numerical value and used as an explanatory variable of the first learning model M1. By including the unique information in the explanatory variables of the first learning model M1 in this way, a first learning model M1 can be constructed that can accurately estimate the damage assessment parameter P even when the tool type, machine tool condition, etc. vary significantly.

[0030] Furthermore, when unique information is included in the explanatory variables of the first learning model M1, in the case of a drill tool, the unique information can include, for example, the source of the material, the rough machining process, the number of holes that can be machined, and the L / D (ratio of length to diameter) of the drill.

[0031] The second estimation unit 180 is configured to estimate the tool damage probability W or the tool damage risk using a second learning model M2. The damage probability W is an index that can be handled continuously between 0 and 100%, while the damage risk is an index that can be handled discontinuously, categorized into, for example, small, medium, and large (note that the number of discontinuously categorized categories for the damage risk may be arbitrary). The second learning model M2 is constructed as a computational model that uses at least one damage assessment parameter P as an explanatory variable and the damage probability W as a target variable. The construction method for the second learning model M2 is not limited, and examples that can be used include naive Bayes, decision trees, random forests, and neural networks.

[0032] The second learning model M2 having such explanatory variables and objective variables is constructed in advance by performing machine learning using training data. The second estimation unit 180 can estimate the damage probability W or the tool damage risk by inputting the damage assessment parameter P estimated by the first estimation unit 170 to the second learning model M2 constructed in this manner. In this embodiment, since the first estimation unit 170 estimates multiple damage assessment parameters P as described above, the second estimation unit 180 inputs these multiple damage assessment parameters P into the second learning model M2 to estimate the damage probability W or the tool damage risk.

[0033] Next, a description will be given of a tool damage probability estimation method implemented by the tool damage probability estimation device 100 having the above configuration. Fig. 3 is a flowchart showing a tool damage probability estimation method according to one embodiment.

[0034] First, the state quantity acquisition unit 150 acquires the state quantity (step S1), and the specific information acquisition unit 160 acquires the specific information (step S2). The acquisition of the state quantity in step S1 and the acquisition of the specific information in step S2 may be performed by an operator operating the input unit 110, or may be performed automatically at a predetermined timing. Furthermore, the execution of steps S1 and S2 is not limited to this order, and may be performed simultaneously or in reverse.

[0035] Next, the first estimation unit 170 selects at least one first learning model M1 based on the unique information acquired in step S2 (step S3). As described above, the unique information and the first learning model M1 are associated with each other, and in step S3, one corresponding to the unique information acquired in step S2 is selected from the multiple first learning models M1 (first learning model candidates). In this embodiment, any one first learning model M1 associated with the unique information is selected from the multiple first learning models M1 (first learning model candidates).

[0036] Next, the first estimation unit 170 estimates at least one damage assessment parameter P by inputting the state quantities acquired in step S1 to the first learning model M1 selected in step S3 (step S4). In this embodiment, multiple damage assessment parameters P (first damage assessment parameter P1, second damage assessment parameter P2, and third damage assessment parameter P3) are estimated.

[0037] Next, the second estimation unit 180 acquires at least one damage assessment parameter P estimated in step S4 and estimates the tool damage probability W or the tool damage risk using the second learning model M (step S5). In this embodiment, the first estimation unit 170 estimates a plurality of damage assessment parameters P (first damage assessment parameter P1, second damage assessment parameter P2, and third damage assessment parameter P3), and these damage assessment parameters P are input into the second learning model M2 to estimate the damage probability W or the tool damage risk.

[0038] In step S5, when estimating the damage probability W or the tool damage risk using the second learning model M2, the explanatory variables of the second learning model M2 may include the state quantities acquired by the state quantity acquisition unit 150. In this case, if the tool is a drill tool, the state quantities may include the material supplier, the roughing process, the number of holes that can be machined, the L / D (length to diameter ratio) of the drill, the equipment model number, the number of holes to be machined, the drill extension length, the feed rate, the number of days since the equipment accuracy inspection, the number of days since the jig, the tensile stress of the material, the drill diameter, and the employee's skills (such as years of work experience entered from a skill chart).

[0039] When estimating the damage risk in step S5, a corresponding category is identified from a plurality of categories (for example, "1. Small," "2. Medium," "3. Large," etc.) that are preset according to the magnitude of the damage risk.

[0040] In another embodiment, in step S3, the first estimation unit 170 may select multiple first learning models M1 based on the unique information acquired in step S2. In this case, multiple first learning models M1 are associated with the unique information, and the first estimation unit 170 selects two or more first learning models M1 associated with the unique information acquired by the unique information acquisition unit 160 from the multiple first learning models M1 (first learning model candidates).

[0041] Then, in step S4, the first estimation unit 170 comprehensively evaluates the calculation results of each of the multiple first learning models M1 to estimate the damage assessment parameter P. For example, by performing statistical processing such as averaging the calculation results of each of the multiple first learning models M1, it becomes possible to estimate the damage assessment parameter P with higher reliability than when a single first learning model M1 is used.

[0042] In this case, the first estimation unit 170 may calculate an index for evaluating the variation, such as the standard deviation σ, for each calculation result of the multiple first learning models M1. For example, assume that the first estimation unit 170 obtains a standard deviation of ±5 or ±15 for the damage assessment parameter P in addition to an average value of 90 from each calculation result of the multiple first learning models M1. Here, if a threshold value of 100 is set in advance for the damage assessment parameter P, when the standard deviation is ±5, the damage assessment parameter P becomes 90±5. Even when the standard deviation is taken into account, the damage risk can be determined to be low because the damage assessment parameter P does not exceed the threshold. On the other hand, when the standard deviation is ±15, the damage assessment parameter P becomes 90±15. When the standard deviation is taken into account, the damage risk can be determined to be high because the damage assessment parameter P exceeds the threshold in some cases.

[0043] Furthermore, the first estimation unit 170 may obtain an index indicating the variation in the maximum value of the damage assessment parameter P for each calculation result of the multiple first learning models M1. For example, when the number of selected first learning models M1 is small, such as two or three, it is difficult to perform evaluation with sufficient accuracy using the standard deviation σ described above. However, by calculating an index indicating the variation in the maximum value, effective evaluation becomes possible even when the number of selected first learning models M1 is small.

[0044] In this way, the first estimation unit 170 statistically processes the calculation results of the multiple first learning models M1, thereby enabling a detailed analysis of the damage assessment parameter P. By using such analysis results together with the damage probability W estimated by the second estimation unit 180 based on the damage assessment parameter P estimated by the first estimation unit 170, more reliable tool damage assessment is possible. As a result, even if there is no first learning model M1 that is perfectly (one-to-one) associated with the inherent parameters of the multiple first learning models M1, evaluation is possible using a relatively similar first learning model M1. Therefore, the present invention can be applied to a wide variety of tools, including types that have not been anticipated in advance.

[0045] As described above, according to the above embodiment, a damage assessment parameter P corresponding to a tool state quantity is estimated using the first learning model M1. The estimated damage assessment parameter P is then input to the second learning model M2, allowing for an appropriate estimation of the tool damage probability W. Furthermore, the results output by the first learning model M1 are merely estimates, and these estimates are obtained from statistics of data within an assumed range (population). Therefore, there is no guarantee that the estimated results will be completely correct, for example, due to subtle differences in assumptions. Therefore, in this embodiment, multiple first learning models M1 are prepared, which effectively allows for a range of assumptions, thereby enabling evaluation of the fluctuation (uncertainty) of the first learning model M1 and evaluation of the final risk from the results (in other words, making it possible to make decisions and operations on the safe side, assuming the worst case scenario).

[0046] In addition, within the scope of the present disclosure, the components in the above-described embodiments may be replaced with well-known components as appropriate, and the above-described embodiments may be combined as appropriate.

[0047] The contents described in each of the above embodiments can be understood, for example, as follows.

[0048] (1) A tool damage probability estimation device according to one aspect includes: A tool damage probability estimation device (100) for estimating a damage probability (W) or a damage risk of a tool used in machining a workpiece, comprising: a first estimation unit (170) for estimating at least one damage assessment parameter (P) for assessing a damage state occurring in the tool by inputting a state quantity of the tool into a first learning model (M1); a second estimation unit (180) for estimating the damage probability or the damage risk by inputting the at least one damage assessment parameter estimated by the first estimation unit into a second learning model (M2); Equipped with.

[0049] According to the above aspect (1), a damage assessment parameter corresponding to a state quantity of a tool is estimated using a first learning model. Then, the estimated damage assessment parameter is further input to a second learning model, whereby a damage probability or a damage risk of the tool can be estimated.

[0050] (2) In another embodiment, in the above embodiment (1), The first estimation unit selects the first learning model from a plurality of first learning model candidates constructed using the same or different methods so that at least one of the explanatory variables or coefficients is different from each other, based on the tool's specific information.

[0051] According to the above aspect (2), the first learning model for estimating the damage assessment parameter is selected from a plurality of first learning model candidates prepared in advance. The plurality of first learning model candidates are constructed using the same or different methods so that at least one of the explanatory variables or coefficients is different from each other, and an appropriate one is selected based on the tool-specific information, thereby enabling the damage assessment parameter to be suitably estimated.

[0052] (3) In another embodiment, in the above embodiment (1), The first estimation unit selects multiple first learning models from multiple candidate first learning models constructed using the same or different methods so that at least one of the explanatory variables or coefficients is different from each other based on the tool's specific information, and estimates the at least one damage assessment parameter by comprehensively evaluating the estimation results using each of the multiple first learning models.

[0053] According to the above aspect (3), a plurality of first learning models for estimating the damage assessment parameter are selected from a plurality of first learning model candidates prepared in advance. Then, estimation results are obtained using the selected plurality of first learning models, and the results are comprehensively evaluated, thereby enabling more reliable estimation of the damage assessment parameter.

[0054] (4) In another embodiment, in the above embodiment (3), The second estimation unit evaluates uncertainty about the damage probability based on estimation results using each of the plurality of first learning models.

[0055] According to the above aspect (4), the uncertainty of the estimated damage probability can be evaluated based on the estimation results using a plurality of first learning models, thereby enabling a quantitative evaluation of the reliability of the tool damage probability.

[0056] (5) In another embodiment, in any one of the above (2) to (4), At least one of the first learning model and the second learning model is constructed using a technique including at least one of regression analysis, neural network, random forest, or support vector machine.

[0057] According to the above aspect (5), by using these techniques to construct a first learning model used in the first estimation unit and a second learning model used in the second estimation unit, it becomes possible to estimate suitable damage assessment parameters, damage probability, or damage risk.

[0058] (6) In another embodiment, in any one of the above (1) to (5), The state quantity of the tool includes at least one of information about the tool, information about the workpiece, and information about machining conditions under which the machining is performed.

[0059] According to the above aspect (6), by including information about the workpiece or information about the machining conditions under which machining is performed in the state quantities of the tool input to the first learning model, it becomes possible to estimate suitable damage assessment parameters.

[0060] (7) In another embodiment, in any one of the above (1) to (6), The state quantity of the tool includes specific information of the tool.

[0061] According to the above aspect (7), by including the inherent information of the tool in the state quantities of the tool input to the first learning model, it becomes possible to estimate suitable damage assessment parameters.

[0062] (8) In another embodiment, in any one of the above (1) to (7), The at least one damage assessment parameter includes a parameter corresponding to at least one of a parameter related to chipping of the tool, a parameter related to the influence of cutting heat on the tool on defects, or a parameter related to the induction of cutting loads on the tool and damage resulting therefrom.

[0063] According to the above aspect (8), damage evaluation can be performed suitably by using these parameters as damage evaluation parameters.

[0064] (9) A tool damage probability estimation method according to one aspect includes: A tool damage probability estimation method for estimating a damage probability (W) or damage risk of a tool used in machining a workpiece, comprising: a step of estimating at least one damage assessment parameter for assessing a damage state occurring in the tool by inputting a state quantity of the tool into a first learning model (M1); a step of estimating the damage probability or the damage risk by inputting the at least one damage assessment parameter (P) into a second learning model (M2); Equipped with.

[0065] According to the above aspect (9), a damage assessment parameter corresponding to a state quantity of a tool is estimated using the first learning model. Then, the estimated damage assessment parameter is further input to the second learning model, whereby the damage probability or damage risk of the tool can be estimated.

[0066] (10) A program according to one aspect includes: A program for estimating the damage probability (W) or damage risk of a tool used in machining a workpiece using a computer, a step of estimating at least one damage assessment parameter for assessing a damage state occurring in the tool by inputting a state quantity of the tool into a first learning model (M1); a step of estimating the damage probability or the damage risk by inputting the at least one damage assessment parameter (P) into a second learning model (M2); is possible.

[0067] According to the above aspect (10), a damage assessment parameter corresponding to a state quantity of a tool is estimated using a first learning model. Then, the estimated damage assessment parameter is further input to a second learning model, whereby a damage probability or a damage risk of the tool can be estimated. [Explanation of symbols]

[0068] 100 Tool damage probability estimation device 110 Input section 120 Storage section 130 Arithmetic section 140 Output section 150 State quantity acquisition unit 160 Unique information acquisition unit 170 1st estimation part 180 Second estimation part M1 First learning model M2 Second learning model P damage assessment parameter W Damage Probability

Claims

1. A tool damage probability estimation device for estimating a damage probability or a damage risk of a tool used in machining a workpiece, comprising: a first estimation unit that estimates a plurality of damage assessment parameters for evaluating a damage state occurring in the tool from different viewpoints by inputting a state quantity of the tool into a first learning model; a second estimation unit that estimates the damage probability or the damage risk by inputting the plurality of damage assessment parameters estimated by the first estimation unit into a second learning model; A tool damage probability estimation device comprising:

2. 2. The tool damage probability estimation device according to claim 1, wherein the first estimation unit selects the first learning model from a plurality of first learning model candidates that differ in the constructed method or in at least one of the explanatory variables or coefficients, based on the tool's specific information.

3. 2. The tool damage probability estimation device according to claim 1, wherein the first estimation unit selects a plurality of first learning models from a plurality of first learning model candidates constructed using the same or different methods so that at least one of the explanatory variables or coefficients is different from each other, based on the specific information of the tool, and estimates the at least one damage assessment parameter by comprehensively evaluating the estimation results using each of the plurality of first learning models.

4. 4. The tool damage probability estimation device according to claim 3, wherein, when the damage probability is estimated by the second estimation unit, the second estimation unit evaluates uncertainty about the damage probability based on estimation results using each of the plurality of first learning models.

5. 5. The tool damage probability estimation device according to claim 1, wherein at least one of the first learning model and the second learning model is constructed using a technique including at least one of regression analysis, a neural network, a random forest, or a support vector machine.

6. 6. The tool damage probability estimation device according to claim 1, wherein the state quantity of the tool includes at least one of information about the tool, information about the workpiece, or information about machining conditions under which the machining is performed.

7. The tool damage probability estimation device according to claim 1 , wherein the state quantity of the tool includes information specific to the tool.

8. 8. The tool damage probability estimation device according to claim 1, wherein the at least one damage assessment parameter includes at least one of a parameter that has an effect on chipping of the tool, a parameter that has an effect on defect due to cutting heat of the tool, or a parameter related to induction of cutting load on the tool and damage resulting therefrom.

9. A tool damage probability estimation device as described in claim 1 or 2, wherein the multiple damage assessment parameters include a first damage assessment parameter corresponding to chipping of the tool, a second damage assessment parameter corresponding to burning of the tool, and a third damage assessment parameter corresponding to the amount of wear of the tool.

10. A tool damage probability estimation device for estimating the damage probability or damage risk of a tool used in machining a workpiece, comprising: a first estimation unit that estimates at least one damage assessment parameter for assessing a damage state occurring in the tool by inputting a state quantity of the tool into a first learning model; a second estimation unit that estimates the damage probability or the damage risk by inputting the at least one damage assessment parameter estimated by the first estimation unit into a second learning model; Equipped with the first estimation unit selects a plurality of first learning models from a plurality of first learning model candidates constructed using the same or different methods so that at least one of explanatory variables or coefficients is different from each other, based on the inherent information of the tool, and estimates the at least one damage assessment parameter by comprehensively evaluating estimation results using each of the plurality of first learning models; A tool damage probability estimation device, wherein when the damage probability is estimated by the second estimation unit, the second estimation unit evaluates uncertainty about the damage probability based on estimation results using each of the plurality of first learning models.

11. A tool damage probability estimation method for estimating a damage probability or a damage risk of a tool used in machining a workpiece, comprising: a step of estimating a plurality of damage assessment parameters for assessing a damage state occurring in the tool from different viewpoints by inputting the state quantities of the tool into a first learning model; a step of estimating the damage probability or the damage risk by inputting the plurality of damage assessment parameters into a second learning model; A tool damage probability estimation method comprising:

12. A program for estimating, using a computer, a damage probability or a damage risk of a tool used in machining a workpiece, a step of estimating a plurality of damage assessment parameters for assessing a damage state occurring in the tool from different viewpoints by inputting the state quantities of the tool into a first learning model; a step of estimating the damage probability or the damage risk by inputting the plurality of damage assessment parameters into a second learning model; A program that can be executed.

Images