Machining defect factor analysis device, machining system, and machining defect factor analysis method

Abstract

Provided is a machining defect factor analysis device (100) that analyzes a factor of a machining defect occurring to a machined surface machined by a machining device according to a machining program and is characterized by comprising: a machining simulation unit (11) that carries out a simulation based on the machining program to generate an ideal machined surface; a first state observation unit (12) that observes a first state variable indicating the state of an actual machined surface machined according to the machining program; a second state observation unit (13) that observes a second state variable indicating the state of the machined surface generated by the machining simulation unit (11); a state variable comparison unit (14) that outputs state variable comparison information indicating a result of comparison between the first state variable and the second state variable; and a learning unit (16) that uses learning data including the state variable comparison information and information indicating a factor of a machining defect that has occurred on the actual machining surface to generate a learned model for inferring the factor of the machining defect from the state variable comparison information.

Description

Machining defect factor analysis device, machining system, and machining defect factor analysis method

[0001] The present disclosure relates to a machining defect factor analysis device, a machining system, and a machining defect factor analysis method that analyze the factors of machining defects that occur in a machining device.

[0002] In a machining device that processes a workpiece using a machining means such as a tool, the operation is controlled using a machining program. For example, in a numerically controlled machine tool, when driving an actuator such as a motor to move the workpiece or the tool to perform machining, a numerically controlled machining program is used that describes movement commands that command the tool path. The machining program is input to the numerical control device of the numerically controlled machine tool, and the numerical control device analyzes the machining program and generates command positions for each drive axis by interpolating the tool path calculated from the movement path for each predetermined control period. The numerical control device controls the drive of the actuator while precisely managing the position based on the generated command positions.

[0003] Furthermore, in numerically controlled machine tools, machining conditions such as the type of tool to be used, the movement speed of each drive axis, and the rotation speed of the main spindle are determined in advance when the machining program is created. The machining conditions are determined by the operator based on past knowledge and experience, using information obtained from materials such as tool catalogs. If the machining conditions are not set appropriately, machining defects may occur. When machining defects occur, the operator must review the machining conditions and perform reprocessing, which results in rework and reduces production efficiency.

[0004] In order to take measures to address and eliminate machining defects, it is important to identify the causes of the defects. Because there are many different causes of machining defects, this is a very difficult task for workers. Furthermore, if the identified causes are incorrect, the measures taken will be incorrect, the machining defect will not be resolved, and further rework will be required. To address this issue, efforts are being made to estimate the causes of machining defects and support measures to address them.

[0005] For example, Patent Literature 1 discloses a machine learning device that observes inspection results of a machined surface of a workpiece by an inspection device as state variables and learns by associating label data indicating the causes of machining defects with the state variables. Patent Literature 1 also discloses a machining defect cause estimation device that uses the learning results of the above-mentioned machine learning device to estimate the causes of machining defects that occur on the machined surface of a workpiece.

[0006] Japanese Patent Application Laid-Open No. 2018-181216

[0007] However, according to the above-mentioned conventional technology, since the relationship between the inspection results of the machined surface and the causes of the machining defects is learned, there is a problem that if the correlation between the inspection results of the machined surface itself and the causes of the machining defects is small, it may not be possible to accurately estimate the causes of the machining defects.

[0008] The present disclosure has been made in view of the above, and aims to provide a machining defect factor analysis device that can improve the accuracy of estimating the cause of a machining defect.

[0009] In order to solve the above-mentioned problems and achieve the object, the machining defect factor analysis device disclosed herein is a machining defect factor analysis device that analyzes factors of machining defects that occur on a machined surface machined by a machining device in accordance with a machining program, and is characterized by comprising: a machining simulation unit that performs a simulation based on the machining program to generate an ideal machined surface that is an ideal machined surface; a first state observation unit that observes a first state variable that indicates the state of an actual machined surface that is the machined surface machined in accordance with the machining program; a second state observation unit that observes a second state variable that indicates the state of the machined surface generated by the machining simulation unit; a state variable comparison unit that outputs state variable comparison information that indicates a result of comparing the first state variable and the second state variable; and a learning unit that generates a trained model for inferring the factors of the machining defect from the state variable comparison information using learning data including the state variable comparison information and information that indicates the factors of the machining defect that occurred on the actual machined surface.

[0010] According to the present disclosure, it is possible to improve the accuracy of estimating the cause of a processing defect.

[0011] FIG. 1 shows an example of the configuration of a machining defect factor analysis device according to a first embodiment. FIG. 2 shows an example of a machined surface generated by a machining simulation unit. FIG. 3 shows an example of the configuration of a machining device. FIG. 4 shows a ball end mill, which is an example of a tool used in machining. FIG. 5 shows an example of a tool trajectory when machining a workpiece with a ball end mill. FIG. 6 shows an example of the detailed configuration of a learning unit. FIG. 7 shows an example of a neural network used by a model generation unit. FIG. 8 shows an example of a graph structure used by a Bayesian network.

[0012] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A machining defect factor analysis device, a machining system, and a machining defect factor analysis method according to embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings.

[0013] 1 is a diagram showing an example of the configuration of a machining defect factor analysis device 100 according to embodiment 1. The machining defect factor analysis device 100 includes a machining simulation unit 11, a first state observation unit 12, a second state observation unit 13, a state variable comparison unit 14, a label generation unit 15, a learning unit 16, and a machining defect factor determination unit 17.

[0014] The machining simulation unit 11 performs a simulation based on a machining program for controlling a machining device and outputs a machined surface generated by machining. The machining simulation unit 11 performs a machining simulation using a machining program in which movement commands for commanding a tool path are written, tool information indicating the characteristics of the tool used in machining, and workpiece information indicating the characteristics of the workpiece. The machining program used here is written, for example, using a description language commonly known as G-code, and describes the tool path, tool movement speed, and rotation speed of the spindle to which the tool is attached. The tool information includes information defining the shape of the tool. For example, the tool information may include information such as the type of tool, the number of tool blades, the tool diameter, and the tool length. Examples of tool types include square end mills, ball end mills, and face mills. The workpiece information includes information for defining the shape of the workpiece. The workpiece information may include information such as the material, size, shape (e.g., rectangular parallelepiped or cylindrical), and coordinate origin of the workpiece.

[0015] 2 is a diagram showing an example of a machined surface generated by the machining simulation unit 11. Here, an example is shown in which a flat surface is machined using a ball end mill.

[0016] An example of a machining simulation method in the machining simulation unit 11 will be described.

[0017] FIG. 3 is a diagram showing an example of the configuration of a machining device 200. Here, an example in which the machining device 200 is a machining center is shown. In the machining device 200, a tool 201 is attached to a spindle 202 via a holder (not shown), and a workpiece W is fixed on a table 203. By moving the tool 201 or the table 203, the relative position between the tool 201 and the workpiece W is changed, and the workpiece W is machined with the tool 201 rotating at high speed. The tool 201 moves along a point on a tool trajectory described in a machining program. At this time, the area passed by the blade of the tool 201 is removed by cutting, and the remaining portion forms the machined surface. The machining simulation unit 11 can predict the machined surface by calculating the envelope of the area passed by the tool 201.

[0018] 4 is a diagram showing a ball end mill, which is an example of a tool 201 used for machining. When the tool 201 is a ball end mill, the tip of the tool 201 is spherical, and the shape of the tip of the tool 201 can be expressed by the following mathematical formula (1). Here, the radius of the tool 201 is R.

[0019]

[0020] Here, (x c , y c , z c ) is the center position of the sphere at the tip of the tool 201, and means the coordinates of the tool 201 on the processing device 200, and will be referred to as tool coordinates hereinafter. The tool coordinates can be calculated from the path and tool length of the tool 201, and are expressed by the following formulas (2) to (4). Here, a point on the tool path at a certain time is defined as (x t , y t , z t ) and the tool length to the tip of the tool 201 is T.

[0021]

[0022]

[0023]

[0024] The machining simulation unit 11 uses the formulas (1) to (4) and the tool trajectory to calculate the passing area of ​​the tool 201. The machining simulation unit 11 can calculate the machining surface by removing the area where the passing area of ​​the tool 201 interferes with the workpiece W.

[0025] FIG. 5 is a diagram showing an example of a tool path when machining a workpiece W with a ball end mill. In FIG. 5, the tool path is indicated by an arrow. FIG. 2 shows a machined surface resulting from machining a rectangular parallelepiped workpiece W with a tool 201 that is a ball end mill using the tool path shown in FIG. 5. In the above-described manner, the machining simulation unit 11 can acquire an ideal machined surface, which is an ideal machined surface. The ideal machined surface is a machined surface that is formed when the tool 201 moves without error according to the machining program and cuts the workpiece W according to the shape of the tool 201 without error. Note that the method for simulating the machined surface in the machining simulation unit 11 is just one example, and simulation may be performed using other methods.

[0026] As shown in Figure 2, even ideal machined surfaces produced by cutting with a ball end mill have irregularities. These irregularities are called cusps, and are geometrically determined by machining conditions such as spindle speed and feed rate, as well as the shape, radius, and number of flutes of the tool 201. The actual machined surface, which is the machined surface formed by actual machining, has machining errors superimposed on the ideal machined surface that is geometrically generated as described above.

[0027] The machining simulation unit 11 outputs three-dimensional shape information of the machining surface generated as a result of the simulation. The three-dimensional shape information is, for example, information on the height in the Z-axis direction at a grid position on an XY plane, recorded together with position information on the X-axis and Y-axis, with a certain position of the machining surface as a reference, as shown in FIG. 2 . The three-dimensional shape information may be data of any structure as long as it is data from which the three-dimensional shape of the machining surface can be restored. The machining simulation unit 11 outputs three-dimensional shape information of the ideal machining surface generated as a result of the simulation to the second state observation unit 13.

[0028] Returning to the explanation of Fig. 1, the first state observation unit 12 observes the state of the actual machined surface, which is the machined surface generated as a result of the machining device 200 actually machining the workpiece W, and outputs a first state variable, which is a state variable indicating the state of the actual machined surface. The output state variable is, for example, the arithmetic mean roughness S, which is the average height of the machined surface. a , the maximum peak height S, which is the height from the reference plane at the highest point of the unevenness of the processed surface p , the maximum valley height S, which is the height from the reference surface at the lowest point of the unevenness of the processed surface v , the maximum height S, which is the sum of the height of the highest peak and the depth of the deepest valley of the processed surface z , the root mean square height S, which means the standard deviation of the surface roughness q , bias S sk , Kurtosis S ku , brightness, color, or three-dimensional shape information of the machined surface. The first state observing unit 12 can, for example, observe the three-dimensional shape of the actual machined surface and calculate the state variables from the observation results. The first state observing unit 12 can measure the three-dimensional shape of the actual machined surface using, for example, a white light interferometer, a contact-type surface roughness measuring instrument, or the like. Alternatively, the first state observing unit 12 may directly observe other state variables instead of observing the three-dimensional shape.

[0029] Among the state variables, for example, the arithmetic mean roughness S a can be calculated using the following formula (5): n is the height of the machined surface at the n-th measurement point, where n is an integer from 1 to N. N is the number of measurement points for the height of the machined surface. The first state observing unit 12 outputs a first state variable indicating the state of the actual machined surface to each of the state variable comparing unit 14 and the learning unit 16.

[0030]

[0031] The second state observing unit 13 observes the state of the ideal machined surface output by the machining simulation unit 11. Since the machining simulation unit 11 outputs three-dimensional shape information of the ideal machined surface as a simulation result, the second state observing unit 13 calculates second state variables, which are state variables indicating the state of the ideal machined surface, from the three-dimensional shape information of the ideal machined surface. The second state observing unit 13 calculates second state variables containing the same type of information as the information contained in the first state variables output by the first state observing unit 12. The second state observing unit 13 outputs the calculated second state variables to the state variable comparing unit 14 and the learning unit 16, respectively.

[0032] The state variable comparison unit 14 compares the first state variable output by the first state observation unit 12 with the second state variable output by the second state observation unit 13, generates state variable comparison information indicating the comparison result, and outputs the state variable comparison information. For example, the state variable comparison unit 14 can calculate the difference between the first state variable and the second state variable, and use the difference between each variable indicating the state of the machined surface as the state variable comparison information. The state variable comparison unit 14 can also calculate the ratio of the magnitude of the second state variable to the magnitude of the first state variable, and use the calculated ratio for each variable indicating the state of the machined surface as the state variable comparison information. The calculation method of the state variable comparison information is just an example, and the state variable comparison unit 14 can use information indicating the comparison result between the first state variable and the second state variable as the state variable comparison information. If the first state variable and the second state variable each include multiple types of state variables, the state variables are compared between state variables of the same type. The state variable comparison unit 14 outputs the state variable comparison information to the learning unit 16.

[0033] The label generation unit 15 generates labels to be used when constructing a trained model for determining machining defect factors in the learning unit 16. The labels are an example of information indicating the factors of machining defects, and are generated based on machining defect factors input by the operator using an input unit (not shown) provided in the label generation unit 15. Machining defect factors refer to factors that cause differences between an actual machined surface and an ideal machined surface. The label generation unit 15 can, for example, display the actual machined surface on a display screen and accept from the operator the input designation of areas that will cause machining defects on the displayed actual machined surface. The machining defect factors to be input may be selected from those prepared in advance, or may be input by the operator using a keyboard or the like. It may be possible to input multiple machining defect factors for one machined surface.

[0034] Machining errors are caused by a variety of factors. Typical examples of machining errors include errors during the unloaded motion of the machining device 200, errors due to the machining process, and errors due to environmental fluctuations. The machining results are influenced by a combination of these factors. For example, errors during the unloaded motion of the machining device 200 include positioning errors of the machining device 200, errors due to vibration, and errors due to friction. Vibration-related errors are caused by vibrations of the machine structure when the machine accelerates or decelerates, and vibrations due to the cutting force generated during cutting, resulting in errors corresponding to the vibration period. Friction-related errors occur when the direction of motion of each axis of the machining device 200 is reversed. Friction-related errors cause scratches called quadrant protrusions on the machined surface.

[0035] The errors due to the machining process include errors due to deformation and vibration of the tool 201 caused by the cutting force generated during cutting, and errors due to deformation and vibration of the workpiece W. Further errors due to the machining process include errors due to the effects of wear on the tool 201 and chip discharge, and errors due to attachment errors of the tool 201. Examples of errors due to environmental fluctuations include errors due to thermal deformation of the machining device 200 caused by temperature changes. When a machining error exceeds a certain level, the operator will determine that the machining is defective, and the causes of the machining error listed here will become the causes of the machining defect. Note that the causes of the machining defect are not limited to those listed here.

[0036] The label generator 15 may also acquire information during processing by the processing device 200 and determine the cause of the processing defect from the acquired information. For example, an acceleration sensor may be installed in the processing device 200, and if data from the acceleration sensor during processing exceeds a predetermined threshold, the processing defect may be determined. In this case, the label generator 15 analyzes the vibration frequency from the acceleration sensor data and identifies the cause of the processing defect from the frequency at which the vibration is large. The label generator 15 may also acquire acceleration data using an encoder, linear encoder, or the like instead of an acceleration sensor. In this case, the label generator 15 may obtain acceleration data by second-order differentiation of position information from the encoder or linear encoder. As another method, a cutting dynamometer may be installed in the processing device 200 to measure the cutting force during processing, and if the cutting force exceeds a predetermined threshold, the processing defect may be determined. The label generator 15 may also analyze the frequency of the cutting force and identify the cause of the processing defect from the frequency at which the cutting force is large. The label generator 15 may also determine the processing defect using sensor information other than the above, or may use other means.

[0037] The label generation unit 15 generates a label corresponding to the actual machining surface based on the input machining defect factors. For example, the label generation unit 15 can assign a label value of "0" when there is no machining defect and assign a unique value to each machining defect factor to quantify the machining defect factors. The label generation unit 15 can also generate a matrix as a label, which takes the value "1" when a specific machining defect factor is present and the value "0" when it is not present. In this case, the generated matrix corresponds to the number of options for the machining defect factors. The method by which the label generation unit 15 generates labels and the data format of the labels are not limited to those described above.

[0038] The learning unit 16 performs learning by associating the first state variable observed by the first state observation unit 12, the second state variable observed by the second state observation unit 13, the state variable comparison information calculated by the state variable comparison unit 14, and the label output by the label generation unit 15.

[0039] 6 is a diagram illustrating an example of a detailed configuration of the learning unit 16. The learning unit 16 includes a learning data acquisition unit 161, a model generation unit 162, and a trained model storage unit 163.

[0040] The learning data acquisition unit 161 associates the first state variable observed by the first state observation unit 12, the second state variable observed by the second state observation unit 13, the state variable comparison information calculated by the state variable comparison unit 14, and the label output by the label generation unit 15, and acquires them as learning data.

[0041] The model generation unit 162 learns a label indicating the presence or absence of a machining defect and, if a machining defect exists, the cause of the machining defect, based on the learning data created based on the first state variable, the second state variable, the state variable comparison information, and the label output from the learning data acquisition unit 161. That is, the model generation unit 162 generates a learned model that infers an optimal label from the first state variable indicating the state of the actual machined surface, the second state variable indicating the state of the machined surface generated by simulation, and the state variable comparison information indicating the comparison result between the first state variable and the second state variable.

[0042] The learning algorithm used by the model generation unit 162 can be a known algorithm such as supervised learning, unsupervised learning, reinforcement learning, etc. As an example, a case where a neural network, which is an example of supervised learning, is applied will be described here.

[0043] The model generation unit 162 learns labels, for example, by supervised learning according to a neural network model. Here, supervised learning refers to a technique in which a learning device is provided with pairs of input and result data, and the device learns features of the learning data and infers results from the input.

[0044] A neural network consists of an input layer consisting of multiple neurons, an intermediate layer consisting of multiple neurons, and an output layer consisting of multiple neurons. The intermediate layer, also called a hidden layer, may be one layer or two or more layers.

[0045] 7 is a diagram showing an example of a neural network used by the model generation unit 162. For example, in the case of a three-layer neural network as shown in FIG. 7, when multiple inputs are input to the input layer (X1-X3), the values ​​are multiplied by weights W1 (w11-w16) and input to the intermediate layer (Y1-Y2), and the results are further multiplied by weights W2 (w21-w26) and output from the output layer (Z1-Z3). This output result varies depending on the values ​​of the weights W1 and W2.

[0046] In the present application, the neural network learns the labels by so-called supervised learning in accordance with the learning data created based on a combination of the first state variable, the second state variable, the state variable comparison information, and the label acquired by the learning data acquisition unit 161.

[0047] That is, the neural network learns by inputting the first state variable, the second state variable, and state variable comparison information into the input layer and adjusting the weights W1 and W2 so that the result output from the output layer approaches the label.

[0048] The configuration of the neural network is not limited to the example shown in Figure 7. The number of neurons included in each of the input layer, intermediate layer, and output layer is not limited to the example shown, and the intermediate layer may have two or more layers. Furthermore, although the input data here includes all of the first state variable, second state variable, and state variable comparison information, it may include only some of the first state variable, second state variable, and state variable comparison information.

[0049] The model generation unit 162 stores the model for which learning has converged as a trained model in the trained model storage unit 163. Whether learning has converged can be determined using a known determination method.

[0050] The above describes a method for generating a trained model using a neural network. Next, we will explain an example using a Bayesian network. A Bayesian network is a method for learning the causal relationships between various events. FIG. 8 is a diagram showing an example of a graph structure used by a Bayesian network. As shown in FIG. 8, a Bayesian network is a model that shows the relationships between events A to E in a graph structure and expresses the relationships between each event using conditional probabilities. FIG. 8 shows that events A and B affect event C, and event C affects events D and E. Events not connected by arrows do not affect each other. These connections are further expressed using conditional probabilities. For example, a conditional probability is the probability that event C will occur if event A occurs. When the model generation unit 162 uses a Bayesian network, the information input to the Bayesian network may be all or some of the first state variable, second state variable, state variable comparison information, and label. When a Bayesian network is used, when inference data is input to the trained model, a probability corresponding to each label is output.

[0051] In this embodiment, the case where supervised learning is applied to the learning algorithm used by the model generation unit 162 has been described, but the present invention is not limited to this. As for the learning algorithm, reinforcement learning, unsupervised learning, semi-supervised learning, or the like can also be applied in addition to supervised learning.

[0052] Furthermore, the learning algorithm used in the model generation unit 162 may be deep learning, which learns to extract the features themselves, or machine learning may be performed according to other known methods, such as genetic programming, functional logic programming, support vector machines, decision trees, gradient boosting, etc.

[0053] The model generation unit 162 may also learn labels according to learning data created for multiple processing devices 200. The model generation unit 162 may acquire first state variables included in the learning data from actual processing surfaces by multiple processing devices 200 used in the same area, or may learn labels using learning data collected from multiple processing devices 200 operating independently in different areas. It is also possible to add or remove processing devices 200 that collect learning data from the target during the process. Furthermore, a learning device that has learned labels indicating causes of processing defects for a certain processing device 200 may be applied to another processing device 200, and the labels for the other processing device 200 may be re-learned and updated.

[0054] Next, the flow of the learning process performed by the learning unit 16 will be described with reference to Fig. 9. Fig. 9 is a flowchart for explaining the learning process performed by the learning unit 16.

[0055] The learning data acquisition unit 161 of the learning unit 16 acquires the first state variable, the second state variable, the state variable comparison information, and the label as learning data (step S101). Note that, although the first state variable, the second state variable, the state variable comparison information, and the label are acquired simultaneously here, it is sufficient if the first state variable, the second state variable, the state variable comparison information, and the label can be acquired in association with each other, and the first state variable, the second state variable, the state variable comparison information, and the label may be acquired at different times.

[0056] The model generation unit 162 performs a learning process to learn the labels and generate a learned model by so-called supervised learning in accordance with the learning data created based on a combination of the first state variable, the second state variable, the state variable comparison information, and the label acquired by the learning data acquisition unit 161 (step S102).

[0057] The trained model storage unit 163 stores the trained model generated by the model generation unit 162 (step S103).

[0058] Returning to the explanation of FIG. 1 , the learning unit 16 can output the learned model stored in the learned model storage unit 163 to the machining defect factor determination unit 17. The machining defect factor determination unit 17 determines the factor of the machining defect using the learned model. Note that in FIG. 1 , the flow of data during learning is indicated by arrows, and when determining the factor of the machining defect, the first state variable output by the first state observation unit 12, the second state variable output by the second state observation unit 13, state variable comparison information output by the state variable comparison unit 14, and the learned model stored in the learning unit 16 are input to the machining defect factor determination unit 17.

[0059] 10 is a diagram showing an example of a detailed configuration of the machining defect factor determination unit 17. The machining defect factor determination unit 17 includes an inference data acquisition unit 171, an inference unit 172, and a determination result output unit 173.

[0060] The inference data acquisition unit 171 acquires inference data including a first state variable, a second state variable, and state variable comparison information. Note that, although the inference data includes all of the first state variable, the second state variable, and the state variable comparison information, it may be a part of the above information, and is the same type of information as the input data used when the learning unit 16 generates the trained model.

[0061] The inference unit 172 infers a label obtained by using the trained model. That is, the inference unit 172 can output a label inferred from the first state variable, the second state variable, and the state variable comparison information by inputting the inference data acquired by the inference data acquisition unit 171 to the trained model stored in the trained model storage unit 163. The inference unit 172 outputs the label, which is the inference result, to the determination result output unit 173.

[0062] Here, the machining defect factor determination unit 17 outputs a label using a trained model learned by the learning unit 16 using the actual machining surface, which is the machining result of the same machining device 200 as at the time of inference, but it is also possible to obtain a trained model from outside, such as from another device of the same type as the machining device 200, and infer a label based on this trained model.

[0063] When a trained model generated using a neural network is used, the inference unit 172 receives the first state variable, the second state variable, and the state variable comparison information as input data, and outputs a label from the trained model. When a trained model generated using a Bayesian network is used, the inference unit 172 receives the first state variable, the second state variable, and the state variable comparison information as input data, and outputs a probability corresponding to each label from the trained model. The inference unit 172 outputs the output from the trained model to the determination result output unit 173 as an inference result.

[0064] The determination result output unit 173 outputs a determination result of the cause of the machining defect on the actual machined surface of the target based on the inference result output by the inference unit 172. When a trained model generated using a neural network is used, the determination result output unit 173 outputs the cause of the machining defect corresponding to the label output as the inference result as the cause of the machining defect on the actual machined surface of the target. Furthermore, when a trained model generated using a Bayesian network is used, the determination result output unit 173 can output the cause of the machining defect corresponding to the label with the highest probability output as the inference result as the cause of the machining defect on the actual machined surface of the target.

[0065] The judgment result output unit 173 can output multiple factors of machining defects. For example, when a trained model generated using a Bayesian network is used, a threshold value may be set in advance, and the judgment result output unit 173 may extract multiple factors of machining defects with a probability equal to or greater than the threshold value, and display the output probability for each factor as a judgment result to the worker. Also, even when a trained model generated using a neural network or other method is used, it is possible to display multiple factors of machining defects.

[0066] 11 is a flowchart for explaining the flow of the processing defect cause determination process. The inference data acquisition unit 171 acquires inference data including a first state variable, a second state variable, and state variable comparison information (step S111).

[0067] The inference unit 172 inputs the inference data acquired by the inference data acquisition unit 171 into the trained model (step S112).

[0068] When the inference unit 172 obtains an inference result from the trained model, it outputs the inference result to the judgment result output unit 173 (step S113).

[0069] The judgment result output unit 173 judges the cause of the processing defect based on the inference result output by the inference unit 172, and outputs the judgment result of the cause of the processing defect to the worker, for example, using a display screen (step S114).

[0070] The machining defect factor analysis device 100 may be built into the machining device 200 or may be a device separate from the machining device 200. Furthermore, the functions of the machining defect factor analysis device 100 may be implemented in separate devices. Furthermore, the machining defect factor analysis device 100 may exist on a cloud server.

[0071] On the actual machined surface, machining errors are superimposed on the geometric shape determined by the machining conditions and the tool 201. For example, the arithmetic mean roughness S a , maximum peak height S p Therefore, it may be difficult to accurately analyze the cause of the machining defect using only the first state variable, which is a state variable indicating the state of the actual machined surface, because it is affected by the geometric shape determined by the machining conditions and the tool 201.

[0072] For example, if a machining defect occurs due to vibration of the machining device 200, the vibration of the machining device 200 is transferred to the actual machined surface, causing the actual machined surface to deviate from the ideal machined surface. Figure 12 shows an example of a machined surface affected by the vibration of the machining device 200. The effect of the vibration of the machining device 200 is transferred to the actual machined surface as a periodic error. Analysis of the actual machined surface results in values ​​that include both geometric shapes and vibration-induced errors, making it difficult to analyze only the effect of machine vibration. Therefore, by using state variable comparison information indicating the difference between the actual machined surface and the ideal machined surface, the geometric shapes are canceled out because they are identical between the ideal machined surface and the actual machined surface, making it possible to extract only the effect of periodic errors due to machine vibration. Therefore, by using not only information on the actual machined surface but also the ideal machined surface obtained by simulation and state variable comparison information indicating the difference between the actual machined surface and the ideal machined surface, it becomes easier to analyze the causes of machining defects.

[0073] As another example, in the case of machining defects due to wear of the tool 201 caused by continuously performing machining with the same tool 201, normal cutting may no longer be possible, resulting in adverse effects such as a deterioration in surface roughness. Since surface roughness is affected not only by wear of the tool 201 but also by machining conditions such as the feed rate and spindle rotation speed of the tool 201, it is often difficult to analyze the cause of machining defects by measuring only the surface roughness of the actual machined surface. Therefore, the effect of wear of the tool 201 can be extracted by comparing the surface roughness of the actual machined surface with the surface roughness of the ideal machined surface.

[0074] As described above, the machining defect factor analysis device 100 according to the first embodiment is a machining defect factor analysis device 100 that analyzes factors of machining defects that occur on a machined surface machined by a machining device 200 in accordance with a machining program, and includes: a machining simulation unit 11 that performs a simulation based on the machining program to generate an ideal machined surface that is an ideal machined surface; a first state observation unit 12 that observes a first state variable that indicates the state of an actual machined surface that is a machined surface machined in accordance with the machining program; a second state observation unit 13 that observes a second state variable that indicates the state of the machined surface generated by the machining simulation unit 11; a state variable comparison unit 14 that outputs state variable comparison information that indicates a result of comparing the first state variable with the second state variable; and a learning unit 16 that generates a trained model for inferring the factors of the machining defects from the state variable comparison result, using learning data including the state variable comparison result and the factors of the machining defects that occur on the actual machined surface.

[0075] In addition, the learning data further includes a first state variable and a second state variable, and the learning unit 16 can generate a learned model for inferring the cause of the processing defect from the first state variable, the second state variable, and the state variable comparison result.

[0076] As described above, the machining defect factor analysis device 100 generates a trained model for inferring the cause of a machining defect from state variable comparison information indicating the result of comparing a first state variable indicating the state of the actual machined surface with a second state variable indicating the state of the ideal machined surface. Therefore, it becomes possible to estimate the cause of a machining defect based on the comparison result between the state of the actual machined surface and the state of the ideal machined surface, thereby improving the accuracy of estimating the cause of the machining defect. This makes it easier for an operator to identify the cause of each machining defect and implement appropriate countermeasures for the machining defect. Since it becomes possible to implement appropriate countermeasures for the machining defect, it is possible to reduce rework such as reprocessing and improve machining productivity.

[0077] Here, it is important to perform learning using state variable comparison information, which indicates the results of a comparison between the actual machined surface and the ideal machined surface, as input data. Since even ideal machined surfaces have irregularities, using state variable comparison information, which indicates the results of a comparison between the actual machined surface and the ideal machined surface, makes it possible to remove the influence of geometric shapes that appear in common between the ideal machined surface and the actual machined surface and extract the influence of machining errors. It is believed that, in many cases, when estimating the causes of machining defects, skilled workers estimate the ideal machined surface based on their experience and then determine the causes of the machining defects by taking into account the difference from the actual machined surface. Therefore, using state variable comparison information makes it possible to perform learning that is closer to the judgment of skilled workers.

[0078] Second Embodiment Fig. 13 is a diagram showing the configuration of a machining defect factor analysis device 100A according to a second embodiment. The machining defect factor analysis device 100A has a machining simulation unit 11, a first state observation unit 12, a second state observation unit 13, a state variable comparison unit 14, a label generation unit 15, a learning unit 16, a machining defect factor determination unit 17, and a machined surface property input unit 18. The following mainly describes the parts that are different from the machining defect factor analysis device 100 according to the first embodiment.

[0079] In the machining defect factor analysis device 100, the machining simulation unit 11 can obtain an ideal machining surface without error through simulation using a machining program, tool information, and information on the workpiece as input. However, the ideal machining surface obtained through simulation may differ from the machining surface envisioned by the operator. This is because the phenomena that occur during machining are complex, and it may not be possible to reflect all factors that affect the machining surface in the simulation. The operator empirically envisions, based on his or her own experience, what kind of machining surface will be obtained with a specific tool 201 and machining conditions. Here, the "machined surface envisioned by the operator" refers to a machining surface that the operator determines is not a machining defect.

[0080] Therefore, the machining defect factor analysis device 100A according to the second embodiment has a machined surface property input unit 18 that receives input of surface property information indicating the properties of the machined surface. The machining simulation unit 11 generates an ideal machined surface based on the input of the machined surface property input unit 18.

[0081] FIG. 14 is a diagram showing an example of a screen displayed by the machined surface property input unit 18. In the example shown here, the machined surface property input unit 18 can accept input of surface property information such as "cusp magnification," "pick-up size," and "pick-up period." In the example of FIG. 14, an interface is used in which each item is input by operating a slider, but an interface in which each item is input numerically or an interface in which pre-prepared options are selected may also be used. Furthermore, the number of items to be input may be one or more, and the contents of the items to be input are not limited to those exemplified in FIG. 14.

[0082] The "cusp magnification" in FIG. 14 is an item for changing the height of cusps formed on the machined surface output by the simulation. The simulation assumes ideal cutting. However, depending on the shape of the tool 201, etc., there are cases where the cusp shape period is consistent but the height is different. The operator estimates the cusp height under specific machining conditions based on experience. The operator can change the cusp height by operating the screen displayed by the machined surface property input unit 18. The cusp magnification is a magnification of the cusp height of the ideal machined surface generated by the machining simulation unit 11 when there is no input from the machined surface property input unit 18. A magnification of "1" is the cusp height of the ideal machined surface generated by the machining simulation unit 11 when there is no input from the machined surface property input unit 18, and a magnification of "2" is twice the cusp height of the ideal machined surface generated by the machining simulation unit 11 when there is no input from the machined surface property input unit 18.

[0083] The "Pull Size" is an item for changing the size of pull on the machined surface output by the simulation, and the "Pull Period" is an item for changing the period at which pull occurs on the machined surface output by the simulation. "Pull" refers to a phenomenon in which the tool 201 is unable to maintain a certain cutting ability due to wear of the tool 201 during machining or the shape of the cutting edge of the tool 201, resulting in machining that pull away the workpiece W rather than cutting it. When pull occurs, the shape changes at a higher period compared to the shape of the cusp. Operators empirically anticipate whether pull will occur under specific machining conditions. In other words, even if pull occurs, the operator may or may not determine that the machining is defective. Therefore, if the operator anticipates the occurrence of pull and does not determine that the occurrence of pull is defective, pull can be generated on the machined surface output by the simulation by operating the screen displayed by the machined surface property input unit 18.

[0084] Fig. 15 is a diagram showing the machined surface properties changed based on input from the machined surface property input unit 18. "Magnification 1" in Fig. 15 shows the machined surface properties when the cusp magnification value is set to "1", "Magnification 2" shows the machined surface properties when the cusp magnification value is set to "2", and "Plough" shows the machined surface properties when ploughing is taken into consideration. The vertical axis in Fig. 15 represents the machined surface height.

[0085] The machining simulation unit 11 simulates machining based on the surface quality information output by the machined surface quality input unit 18, and outputs an ideal machined surface that takes into account differences in cusp height and gouges. The second state observation unit 13 observes the state variables of the ideal machined surface generated based on the surface quality information output by the machined surface quality input unit 18. The subsequent learning process and judgment process of the causes of machining defects are the same as those in the first embodiment.

[0086] As described above, the machining defect factor analysis device 100A according to the second embodiment, like the first embodiment, can estimate the cause of a machining defect based on the comparison between the state of the actual machined surface and the state of the ideal machined surface, thereby improving the accuracy of estimating the cause of the machining defect. Furthermore, the machining defect factor analysis device 100A includes the machined surface property input unit 18, and the machining simulation unit 11 generates the ideal machined surface based on input from the machined surface property input unit 18. This allows the operator's anticipated machined surface property to be reflected in the ideal machined surface, which is the simulation result. Therefore, it is possible to generate an ideal machined surface that takes into account phenomena anticipated by the operator that cannot be reflected in a machining simulation that uses a machining program, tool information, and workpiece information as input, such as cusp height changes and the occurrence of gouges. This allows the operator to determine the cause of a machining defect based on an ideal machined surface that is closer to the operator's expectations. Therefore, it is possible to further improve the accuracy of estimating the cause of a machining defect.

[0087] Third Embodiment Fig. 16 is a diagram showing an example of the configuration of a machining defect factor analysis device 100B according to a third embodiment. The machining defect factor analysis device 100B includes a machining simulation unit 11, a first state observation unit 12, a second state observation unit 13, a state variable comparison unit 14, a label generation unit 15, a learning unit 16, a machining defect factor determination unit 17, and a machining defect information input unit 19. The following mainly describes the differences from the machining defect factor analysis device 100 according to the first embodiment.

[0088] In the third embodiment, the machining simulation unit 11 performs a simulation under conditions where at least one factor of machining defect exists in addition to an ideal machined surface without machining defects, and generates an assumed machining defect surface, which is a machined surface when a factor of machining defect exists. The machining defect information input unit 19 accepts input of machining defect information from an operator, which is information used by the machining simulation unit 11 to simulate machining defects. There are various factors of machining defects, and in some cases, machining defects may be caused by multiple factors. Some of the factors of machining defects can be incorporated into the simulation and simulated by formulating them, for example.

[0089] FIG. 17 is an explanatory diagram of an example of a simulation that takes into account the contact of a tool 201 as a factor in machining defects. Contact of the tool 201 is one example of a factor in machining defects. In machining using a tool 201 such as an end mill or a drill in a machining center, the workpiece W is fixed, and the tool 201 is rotated at high speed to bring it into contact with the workpiece W, thereby performing machining. The tool 201 is fixed to a spindle 202 via a holder (not shown) or the like, and rotates at high speed in conjunction with the rotation of the spindle 202. Ideally, it is desirable that the rotational axis 202a of the spindle 202 and the central axis 201a of the tool 201 coincide with each other. However, due to mounting errors and the like, the tool 201 may be mounted misaligned from the rotational axis 202a of the spindle 202, as shown in FIG. 17 . It is known that this misalignment reduces the accuracy and quality of the machined surface.

[0090] When the machining simulation unit 11 takes into account the movement of the tool 201, it simulates the central axis 201a of the tool 201 moving while rotating at a position offset from the central axis 202a of rotation of the spindle 202, and calculates the passing area of ​​the tool 201.

[0091] Fig. 18 is a diagram showing an example of the simulation results of the machined surface taking into account the contact of the tool 201. The example of Fig. 18 shows the simulation results when a rectangular parallelepiped workpiece W is machined with a ball end mill according to the tool trajectory shown in Fig. 5. Compared to Fig. 2, the unevenness of the machined surface is no longer uniform, resulting in a result that differs from the ideal machined surface. In other words, when contact of the tool 201 exists in actual machining, the shape of the machined surface is closer to the machined surface simulated in consideration of the contact of the tool shown in Fig. 18 than to the ideal machined surface shown in Fig. 2.

[0092] In addition, when the machining simulation unit 11 takes into consideration the vibration of the machining device 200, the position of the tip of the tool 201 (x t , y t , z t ) can be used to represent a change in the position of the tool 201 due to vibration of the processing device 200.

[0093] 19 is a diagram showing an example of the simulation results of the machined surface taking into consideration the vibration of the machining device 200. Compared with FIG. 2, the convex portions of the machined surface are scraped away more than the ideal machined surface of FIG.

[0094] The machining defect information input unit 19 accepts input of the cause of the machining defect to be simulated and information for reproducing the machining defect. The information required to reproduce the machining defect varies depending on the type of machining defect. In the case of contact of the tool 201, the machining defect information input unit 19 accepts input of the deviation b between the central axis 201a of the tool 201 and the rotational center axis 202a of the spindle 202, as shown in FIG. 17 . In addition, in the case of vibration of the machining device 200, the machining defect information input unit 19 can accept input of the natural frequency and amplitude of the vibration. For one simulation, the machining defect information input unit 19 may accept input for one type of machining defect or for multiple types of machining defects. In addition, the information for reproducing the machining defect may be measured in advance using an actual machine or calculated by simulation, etc.

[0095] The machining simulation unit 11 generates an assumed machining defect surface, which is a machining surface when a factor of machining defect exists, by performing a simulation under conditions where at least one factor of machining defect exists, in addition to the ideal machining surface similar to that in embodiment 1. The machining simulation unit 11 can output three-dimensional shape information of the ideal machining surface and the assumed machining defect surface.

[0096] The second state observation unit 13 observes the states of the ideal machined surface and the assumed machining defect surface output by the machining simulation unit 11, and outputs second state variables for each machined surface. The state variable comparison unit 14 compares the first state variables with the second state variables calculated for each simulated machined surface, and generates state variable comparison information. That is, the state variable comparison unit 14 outputs state variable comparison information for each machined surface calculated by the machining simulation unit 11.

[0097] As in the first embodiment, the learning unit 16 performs learning by associating the first state variable observed by the first state observation unit 12, the second state variable observed by the second state observation unit 13, the state variable comparison information calculated by the state variable comparison unit 14, and the label output from the label generation unit 15.

[0098] When a machining defect occurs on a machined surface due to a certain machining defect factor, the assumed machining defect surface simulated by taking into account the corresponding machining defect factor is similar to the actual machined surface. In other words, in this case, the first state variable of the actual machined surface and the second state variable of the assumed machining defect surface have similar values, and the comparison information of these state variables has a value close to 0. In this way, information on which machined surface the actual machined surface is similar to takes into account which machining defect factor is used as input data for learning and inference, enabling more accurate learning and inference.

[0099] As described above, according to the machining defect factor analysis device 100B of the third embodiment, the machining simulation unit 11 performs a simulation under conditions in which at least one factor of machining defect exists in addition to an ideal machined surface to generate an assumed machining defect surface, which is a machined surface when a factor of machining defect exists, and the second state observation unit 13 observes a second state variable indicating the state of the assumed machining defect surface in addition to a second state variable indicating the state of the ideal machined surface. As a result, a trained model for determining the factor of machining defect is generated based on a comparison result between the assumed machining defect surface generated by performing a simulation under conditions in which a factor of machining defect exists in addition to the ideal machined surface and the actual machined surface, thereby making it possible to further improve the accuracy of estimating the factor of machining defect.

[0100] Fourth Embodiment Fig. 20 is a diagram showing an example of the configuration of a machining defect factor analysis device 100C according to a fourth embodiment. The machining defect factor analysis device 100C includes a machining simulation unit 11, a first state observation unit 12, a second state observation unit 13, a state variable comparison unit 14, a label generation unit 15, a learning unit 16, a machining defect factor determination unit 17, a third state observation unit 20, a condition input unit 21, and a data selection unit 22. The following mainly describes the parts that are different from the first embodiment.

[0101] The third state observation unit 20 observes the state of the machining device 200 during machining. The state of the machining device 200 during machining includes, for example, vibration during machining and motor output. Therefore, it is useful to observe the state of the machining device 200 during machining and use it as input to determine the cause of machining defects. For example, vibration during machining can be measured by sensors such as an encoder and an acceleration sensor attached to the motor used to drive the machine. Furthermore, motor output can be obtained from the motor current value, etc. Other states of the machining device 200 during machining include position, speed, acceleration, etc.

[0102] The third state observing unit 20 observes the state of the machining apparatus 200 during machining as described above and outputs the observed state as a third state variable. The output data may be a value calculated using the acquired data, such as the average, standard deviation, maximum value, and minimum value of the acquired data. The third state observing unit 20 may also add information, such as the position of the machining surface, for correlating the third state variable with the first state variable and the second state variable and output the third state variable.

[0103] The condition input unit 21 accepts input of information that is determined in advance before machining, such as machining conditions, tool information, and information on the workpiece to be machined. Machining conditions include, for example, a command speed, a spindle rotation speed, and a cutting speed. The condition input unit 21 accepts this information input by an operator and outputs it as machining condition information.

[0104] The learning unit 16 and the machining defect factor determination unit 17 can use a third state variable and machining condition information as input data in addition to the first state variable, the second state variable, and the state variable comparison information. The first state variable, the second state variable, and the state variable comparison information are information indicating the properties of the machined surface. However, using the third state variable and the machining condition information can improve the accuracy of learning and determination. The implementation methods of the learning unit 16 and the machining defect factor determination unit 17 are the same as those of the first embodiment except for the input data. Note that, as in FIG. 1 , FIG. 20 indicates the flow of data during learning with arrows, and the input data to the machining defect factor determination unit 17 is omitted. The input data to the machining defect factor determination unit 17 is the same type of information as the information used as input data when generating the trained model used by the machining defect factor determination unit 17.

[0105] The properties of the machined surface are correlated with machining conditions, such as the feed rate and spindle rotation speed, which are determined before machining, and the state of the machining device 200 during machining. Therefore, it may be possible to accurately determine the cause of machining defects by using this information in addition to the information indicating the properties of the machined surface.

[0106] The data selection unit 22 selects data from the input data that the learning unit 16 will use to generate a trained model. If a large amount of information is input, unnecessary information may be included in determining the cause of processing defects, which may actually reduce the accuracy of learning and determination. Therefore, the data selection unit 22 selects information useful for learning from the input data and outputs the selected data to the learning unit 16.

[0107] Specifically, the data selection unit 22 calculates the correlation coefficient r of each data with respect to the label, and selects data for which the absolute value of the correlation coefficient r is equal to or greater than a threshold. This makes it possible to use only data that is correlated with the label, i.e., correlated with the cause of the processing defect. Alternatively, the correlation coefficient r may be calculated between input data, and if the absolute value of the correlation coefficient r is equal to or greater than a threshold, the correlation is deemed high, and one of the data may be removed.

[0108] The correlation coefficient r is an index that indicates the strength of the correlation between two data, and can be expressed, for example, by the following formula (6): i , y i indicates each data, n is the number of data, and i is an integer from 1 to n.

[0109]

[0110] The method of selecting data in the data selection unit 22 is an example, and other means may be used.

[0111] As described above, the machining defect factor analysis device 100C according to the fourth embodiment may further include a third state observation unit 20 that observes the state of the machining device 200 during machining, a first state observation unit 12, a second state observation unit 13, a third state observation unit 20, a state variable comparison unit 14, and a data selection unit 22 that selects data that the learning unit 16 uses to generate a trained model from data output by the condition input unit 21. Note that the data selection unit 22 may calculate a correlation with a label for each piece of data and select data whose correlation is equal to or greater than a threshold, or may calculate a correlation between input data and reduce the number of input data by deleting one of the data whose correlation is high.

[0112] In addition, the processing defect cause analysis device 100C further includes a condition input unit 21 that accepts input of processing condition information, which is information that is determined before processing, and the learning unit 16 can further use the processing condition information as input data to generate a learned model.

[0113] Next, the hardware configuration of the machining defect factor analysis devices 100, 100A, 100B, and 100C will be described. Each of the machining defect factor analysis devices 100, 100A, 100B, and 100C is implemented, for example, by a computer system. Each of the machining defect factor analysis devices 100, 100A, 100B, and 100C may be implemented by a single computer system or multiple computer systems. For example, the machining defect factor analysis devices 100, 100A, 100B, and 100C may be implemented by a cloud system. In a cloud system, the hardware of the computer system and devices such as servers for each function can be arbitrarily separated. For example, a single computer system may function as multiple devices, or multiple computer systems may function as a single device.

[0114] An example of the configuration of a computer system that realizes the machining defect factor analysis devices 100, 100A, 100B, and 100C will be described. Fig. 21 is a diagram showing an example of the configuration of a computer system that realizes the machining defect factor analysis devices 100, 100A, 100B, and 100C. As shown in Fig. 21, this computer system includes a control unit 101, an input unit 102, a storage unit 103, a display unit 104, a communication unit 105, and an output unit 106, which are connected via a system bus 107.

[0115] In FIG. 21 , the control unit 101 is, for example, a central processing unit (CPU). The control unit 101 executes a machining defect factor analysis program that describes the processes to be performed by the machining defect factor analysis devices 100, 100A, 100B, and 100C of this embodiment. The input unit 102 is composed of, for example, a keyboard, a mouse, and the like, and is used by a user of the computer system to input various information. The memory unit 103 includes various memories such as a random access memory (RAM) and a read-only memory (ROM), as well as a storage device such as a hard disk, and stores programs to be executed by the control unit 101, necessary data obtained during processing, and the like. The memory unit 103 is also used as a temporary storage area for programs. The display unit 104 is composed of, for example, a liquid crystal display (LCD) panel, and displays various screens to the user of the computer system. The communication unit 105 is, for example, a communication circuit that performs communication processing. The communication unit 105 may be composed of multiple communication circuits corresponding to multiple communication methods. The output unit 106 is an output interface that outputs data to an external device such as a printer or an external storage device.

[0116] Note that Fig. 21 is merely an example, and the configuration of the computer system is not limited to the example of Fig. 21. For example, the computer system may not include the output unit 106. Furthermore, when the machining defect factor analysis devices 100, 100A, 100B, and 100C are realized by a plurality of computer systems, not all of these computer systems may be the computer systems shown in Fig. 21. For example, some of the computer systems may not include at least one of the display unit 104, the output unit 106, and the input unit 102 shown in Fig. 21.

[0117] Here, an example of the operation of a computer system until a machining defect factor analysis program describing the processing of the machining defect factor analysis devices 100, 100A, 100B, and 100C is ready to be executed will be described. In the computer system having the above configuration, the machining defect factor analysis program is installed in the memory unit 103 from, for example, a CD-ROM or DVD-ROM inserted in a CD (Compact Disc)-ROM drive or DVD (Digital Versatile Disc)-ROM drive (not shown). When the machining defect factor analysis program is executed, the machining defect factor analysis program is read from the memory unit 103 and stored in an area that serves as the main memory of the memory unit 103. In this state, the control unit 101 executes the processing of the machining defect factor analysis devices 100, 100A, 100B, and 100C of this embodiment in accordance with the machining defect factor analysis program stored in the memory unit 103.

[0118] In the above description, a program describing the processing in the processing defect factor analysis devices 100, 100A, 100B, and 100C is provided using a CD-ROM or a DVD-ROM as a recording medium. However, the present invention is not limited to this, and a program provided by a transmission medium such as the Internet via the communication unit 105 may be used depending on the configuration of the computer system, the capacity of the program to be provided, and the like.

[0119] The machining defect factor analysis program of this embodiment causes a computer to execute a process of analyzing factors of machining defects that occur on a machined surface machined by the machining device 200, specifically, a step of performing a simulation based on the machining program to generate an ideal machined surface that is an ideal machined surface, a step of observing a first state variable that indicates the state of an actual machined surface that is a machined surface machined by the machining device 200 in accordance with the machining program, a step of observing a second state variable that indicates the state of the machined surface generated by the simulation, a step of generating state variable comparison information that indicates the result of comparing the first state variable with the second state variable, and a step of generating a trained model for inferring the factors of the machining defects from the state variable comparison information using learning data that includes the state variable comparison information and information that indicates the factors of the machining defects that occur on the actual machined surface.

[0120] The machining simulation unit 11, the first state observation unit 12, the second state observation unit 13, the state variable comparison unit 14, the label generation unit 15, the learning unit 16, the machining defect factor determination unit 17, the third state observation unit 20 and the data selection unit 22 are realized by the control unit 101 shown in FIG. 21, and the machined surface property input unit 18, the machining defect information input unit 19 and the condition input unit 21 are realized by the control unit 101 shown in FIG. 21 causing the display unit 104 to display a display screen and receiving input operations on the display screen by the input unit 102.

[0121] The division of functions shown in FIG. 1 and the like is an example, and as long as the processing defect cause analysis devices 100, 100A, 100B, and 100C can perform the above-described operations, the division of each functional unit is not limited to the example shown in FIG. 1.

[0122] The configurations shown in the above embodiments are merely examples, and may be combined with other known technologies, or different embodiments may be combined with each other. It is also possible to omit or modify parts of the configurations as long as they do not deviate from the gist of the invention.

[0123] REFERENCE SIGNS LIST 11 Machining simulation unit, 12 First state observation unit, 13 Second state observation unit, 14 State variable comparison unit, 15 Label generation unit, 16 Learning unit, 17 Machining defect factor judgment unit, 18 Machined surface property input unit, 19 Machining defect information input unit, 20 Third state observation unit, 21 Condition input unit, 22 Data selection unit, 100, 100A, 100B, 100C Machining defect factor analysis device, 101 Control unit, 102 Input unit, 103 Memory unit, 104 Display unit, 105 Communication unit, 106 Output unit, 107 System bus, 161 Learning data acquisition unit, 162 Model generation unit, 163 Learned model storage unit, 171 Inference data acquisition unit, 172 Inference unit, 173 Judgment result output unit, 200 Machining device, 201 Tool, 201a Center axis, 202 Spindle, 202a: Rotational center axis, 203: Table, W: Workpiece.

Claims

1. A machining defect cause analysis device that analyzes causes of machining defects that occur on a machined surface machined by a machining device in accordance with a machining program, comprising: a machining simulation unit that performs a simulation based on the machining program to generate an ideal machined surface that is an ideal machined surface; a first state observation unit that observes a first state variable that indicates a state of an actual machined surface that is a machined surface machined in accordance with the machining program; a second state observation unit that observes a second state variable that indicates the state of the machined surface generated by the machining simulation unit; a state variable comparison unit that outputs state variable comparison information that indicates a result of comparing the first state variable with the second state variable; and a learning unit that uses learning data including the state variable comparison information and information that indicates causes of machining defects that occur on the actual machined surface to generate a learned model for inferring the causes of the machining defects from the state variable comparison information.

2. The machining defect cause analysis device according to claim 1, further comprising a machined surface property input unit that accepts input of the properties of the machined surface, wherein the machining simulation unit generates the ideal machined surface based on the input from the machined surface property input unit.

3. The machining defect cause analysis device according to claim 1 or 2, characterized in that the machining simulation unit generates a machining defect assumed surface, which is the machining surface when the cause of the machining defect exists, by performing a simulation under conditions in which at least one factor of the machining defect exists in addition to the ideal machining surface, and the second state observation unit observes the second state variable indicating the state of the ideal machining surface, as well as the second state variable indicating the state of the machining defect assumed surface.

4. The machining defect cause analysis device according to any one of claims 1 to 3, further comprising: an inference data acquisition unit that acquires inference data including the state variable comparison information; an inference unit that outputs the cause of the machining defect from the state variable comparison information input from the inference data acquisition unit using a learned model for inferring the cause of the machining defect from the state variable comparison information; and a judgment result output unit that outputs a judgment result indicating whether or not the machining defect has occurred, and which further indicates the cause of the machining defect if the machining defect has occurred.

5. The processing defect cause analysis device according to claim 4, characterized in that, when the processing defect occurs, the judgment result output unit outputs at least one cause of the processing defect, and for each cause, outputs the probability that the cause will cause the processing defect.

6. A machining defect cause analysis device as described in any one of claims 1 to 5, further comprising: a third state observation unit that observes the state of the machining device during machining; and a data selection unit that selects data to be used by the learning unit to generate the trained model from data output by the first state observation unit, the second state observation unit, the third state observation unit, and the state variable comparison unit.

7. A machining defect cause analysis device as described in any one of claims 1 to 6, further comprising a condition input unit that accepts input of machining condition information, which is information that is determined before machining, wherein the learning unit further uses the machining condition information as input data to generate the trained model.

8. A machining defect cause analysis device as described in any one of claims 1 to 7, characterized in that the learning data further includes the first state variable and the second state variable, and the learning unit generates a learned model for inferring the cause of the machining defect from the first state variable, the second state variable, and the state variable comparison information.

9. A machining system comprising: a machining device which machines an object to be machined in accordance with a machining program; and a machining defect factor analysis device which analyzes factors of machining defects occurring on a machined surface machined by the machining device, wherein the machining defect factor analysis device comprises: a machining simulation unit which performs a simulation based on the machining program to generate an ideal machined surface which is an ideal machined surface; a first state observation unit which observes a first state variable which indicates a state of an actual machined surface which is a machined surface machined in accordance with the machining program; a second state observation unit which observes a second state variable which indicates the state of the machined surface generated by the machining simulation unit; a state variable comparison unit which outputs state variable comparison information which indicates a result of comparing the first state variable with the second state variable; and a learning unit which uses learning data including the state variable comparison information and factors of machining defects occurring on the actual machined surface to generate a learned model for inferring the factors of the machining defects from the state variable comparison information.

10. A machining defect cause analysis method comprising: a step of: a machining device machining an object in accordance with a machining program; and a step of a machining defect cause analysis device analyzing causes of machining defects occurring on a machining surface machined by the machining device, wherein the step of analyzing the causes of the machining defects comprises: a step of performing a simulation based on the machining program to generate an ideal machined surface which is an ideal machined surface; a step of observing a first state variable indicating a state of an actual machined surface which is a machined surface machined by the machining device; a step of observing a second state variable indicating the state of the machined surface generated by the simulation; a step of generating state variable comparison information indicating a result of comparing the first state variable with the second state variable; and a step of generating a trained model for inferring the cause of the machining defect from the state variable comparison information using learning data including the state variable comparison information and information indicating the cause of the machining defect occurring on the actual machined surface.

Images