Learning device, electrical discharge machine, learning method, and storage medium

Abstract

The learning device (23) includes: a data acquisition unit (231) that acquires learning data including workpiece information, machining conditions, and values of evaluation indexes, where the workpiece information includes information related to a target workpiece and information related to an electrode used in an electric discharge machine, the machining conditions are machining conditions used when shaping a workpiece into the target workpiece by electric discharge machining, and the values of the evaluation indexes indicate the machining surface quality of the machined surface of the workpiece after shaping electric discharge machining using the machining conditions; and a model generation unit (232) that uses the learning data to generate a trained model for inferring machining conditions according to the workpiece information, the machining conditions being those that improve the machining surface quality of the machined surface.

Description

Technical Field

[0001] The present invention relates to a learning device, an electric discharge machine, a learning method, and a storage medium applied to profile electric discharge machining. Background Art

[0002] An electric discharge machine performs machining on a workpiece by generating an electric discharge in a gap between the workpiece and an electrode under machining conditions such as machining voltage, machining pulses, and chip discharge operation. As a result of the electric discharge machining, the occurrence state of the discharge marks formed on the machining surface changes according to the machining conditions. A discharge mark is a pit-shaped dent formed on the machining surface when the machining surface is removed by one pulse discharge. The shape of the discharge mark varies depending on the electrical conditions of the pulse discharge, the electrodes, the material of the workpiece, and the concentration of chips in the machining fluid in which the workpiece is immersed.

[0003] Patent Document 1 describes an electric discharge machine that estimates the volume resistivity of the entire machining fluid, thereby obtaining a discharge current for making the average discharge voltage during electric discharge machining the same as the discharge voltage using a new machining fluid, and obtaining machining conditions corresponding to the obtained discharge current.

[0004] The electric discharge machine described in Patent Document 1 can suppress the overall amount of occurrence of pits, which are discharge marks larger than the desired discharge marks formed on the machining surface. Here, the "desired discharge marks" refer to the distribution of discharge marks having a discharge mark diameter within a certain range centered on the target discharge mark diameter. In addition, "larger than the desired discharge marks" means a shift in the direction of a larger number of discharge marks with a relatively large discharge mark diameter in the distribution of the ideal discharge mark diameter and a shift from the distribution of the ideal discharge mark diameter to the direction of a larger discharge mark diameter. The evaluation index is machining surface quality data obtained by evaluating the machining surface quality based on the distribution of discharge marks larger than a predetermined size on the machining surface obtained from an image of the machining surface after profile electric discharge machining.

[0005] Patent Document 1: Japanese Patent Application Laid-Open No. 2005-205574 Summary of the Invention

[0006] However, in electrical discharge machining, not only the occurrence number of pits is suppressed, but also the machining surface quality indicating the uniformity of the surface roughness of the machining surface and the uniformity of the size of the discharge marks formed on the machining surface is important. The machining surface quality can be represented by the fluctuation of the size of the discharge marks occurring on the machining surface or the aggregation of the occurrence positions of the discharge marks on the machining surface. However, even if the electrical discharge machining machine described in Patent Document 1 can suppress the overall amount of pit occurrence in the machining surface, it cannot suppress the fluctuation of the size of the discharge marks in the machining surface. In addition, the electrical discharge machining machine described in Patent Document 1 cannot suppress the aggregation of the occurrence positions of the discharge marks in the machining surface. That is, in the electrical discharge machining machine described in Patent Document 1, appropriate machining conditions for electrical discharge machining that can improve the machining surface quality cannot be obtained.

[0007] The present invention has been made in view of the above circumstances, and an object thereof is to obtain a learning device that can learn appropriate machining conditions for electrical discharge machining that can improve the machining surface quality.

[0008] To solve the above problems and achieve the object, the learning device according to the present invention learns the machining conditions used in the profile electric discharge machining of a workpiece in an electric discharge machining machine. The learning device includes: a data acquisition unit that acquires learning data including workpiece information, machining conditions, and values of evaluation indices, where the workpiece information includes information related to the target workpiece and information related to the electrode used in the electric discharge machining machine, the machining conditions are the machining conditions used when profile electric discharge machining the workpiece into the target workpiece, and the value of the evaluation index represents the machining surface quality of the workpiece after profile electric discharge machining using the machining conditions; and a model generation unit that uses the learning data to generate a trained model for inferring machining conditions according to the workpiece information, the machining conditions being such that the machining surface quality is improved.

[0009] Effects of the Invention

[0010] The learning device according to the present invention has the following effect, that is, it can learn appropriate machining conditions for electrical discharge machining that can improve the machining surface quality. Description of the Drawings

[0011] Figure 1 It is a diagram showing the structure of the electric discharge machining machine according to Embodiment 1.

[0012] Figure 2 It is a diagram showing the structure of the control device according to Embodiment 1.

[0013] Figure 3 It is a diagram showing the structure of the learning device according to Embodiment 1.

[0014] Figure 4 This is a diagram showing the structure of the inference device according to Embodiment 1.

[0015] Figure 5 This is a distribution diagram showing an example of the distribution of the discharge mark diameters in the machined surface of the workpiece after electrical discharge machining by the electrical discharge machine according to Embodiment 1 and the ideal distribution of the discharge mark diameters.

[0016] Figure 6 This is a flowchart showing the processing sequence of the learning process performed by the learning device according to Embodiment 1.

[0017] Figure 7 This is a flowchart showing the processing sequence of the inference process performed by the inference device according to Embodiment 1 and the control process performed by the control device according to Embodiment 1.

[0018] Figure 8 This is a diagram showing the structure of the learning device of the electrical discharge machine in Embodiment 2.

[0019] Figure 9 This is a diagram for explaining the control of the interelectrode distance and the lifting operation control of the shaped electrode discharge machining in the machining fluid by the electrical discharge machine in Embodiment 2.

[0020] Figure 10 This is a characteristic diagram showing the relationship between the non-uniformity of the distribution of the machining chips between the machining surface and the electrode in the in-plane direction of the machining surface of the shaped electrode discharge machining in the machining fluid by the electrical discharge machine in Embodiment 2 and the machining chip discharge effect of the lifting operation. [[ID=2)]]

[0021] Figure 11 This is a diagram showing an example of the machining surface non-uniformity pattern in Embodiment 2.

[0022] Figure 12 This is a diagram showing an example of the machining surface non-uniformity pattern in Embodiment 2.

[0023] Figure 13 This is a diagram showing the machining surface quality data calculation unit that stores the machining surface non-uniformity pattern in Embodiment 2.

[0024] Figure 14 This is a diagram showing the first machining surface quality data in the region of the machining surface image segmented in Embodiment 2.

[0025] Figure 15 This is a diagram showing the state after overlapping the first machining surface quality data in the segmented region of the machining surface image and the image of the machining surface non-uniformity pattern in Embodiment 2.

[0026] Figure 16This is a flowchart showing the processing sequence of the learning process performed by the learning device in Embodiment 2.

[0027] Figure 17 This is a flowchart showing the processing sequence of other learning processes performed by the learning device in Embodiment 2.

[0028] Figure 18 This is a diagram showing an example of the hardware structure of the machining condition setting unit related to Embodiments 1 and 2. Detailed Embodiment

[0029] Hereinafter, the learning device, the electric discharge machine, the learning method, and the storage medium related to the embodiments will be described in detail with reference to the drawings.

[0030] Embodiment 1.

[0031] Figure 1 This is a diagram showing the structure of the electric discharge machine related to Embodiment 1. The electric discharge machine 1 is a device for performing profile electric discharge machining. The electric discharge machine 1 is a device that performs profile electric discharge machining by applying a high-frequency pulse voltage between the machining electrode, i.e., the electrode E, and the workpiece 17, thereby causing an electric discharge between the workpiece 17 to be machined and the electrode E. In the following description, the profile electric discharge machining of the workpiece 17 may sometimes be simply referred to as electric discharge machining.

[0032] The electric discharge machine 1 includes a base 19, a control device 2, a drive unit 12, a display unit 13, and a table 18. The electric discharge machine 1 machines the workpiece 17 placed on the table 18 on the base 19 using the electrode E mounted on the drive unit 12 under the control of the control device 2. The electrode E is disposed at a position opposite to the workpiece 17.

[0033] The control device 2 includes a machine control unit 14, a power supply control unit 15, and a machining condition setting unit 16. The control unit composed of the machine control unit 14 and the power supply control unit 15 controls the operation of the profile electric discharge machining performed by the electric discharge machine 1, i.e., the profile electric discharge machining operation.

[0034] The machine control unit 14 and the power supply control unit 15 are control units that control the controlled objects. The controlled object of the machine control unit 14 is the drive unit 12, etc., and the controlled object of the power supply control unit 15 is the electric power.

[0035] The machine control unit 14 controls the position, etc., of the drive unit 12 based on the machining conditions sent from the machining condition setting unit 16. The machine control unit 14 controls the drive unit 12, thereby controlling the gap between the electrode E and the workpiece 17 so that an electric discharge occurs between the electrode E and the workpiece 17.

[0036] The power supply control unit 15 controls the power supplied between the electrode E and the workpiece 17 based on the machining conditions sent from the machining condition setting unit 16. That is, the power supply control unit 15 controls the discharge between the electrode E and the workpiece 17.

[0037] The machining condition setting unit 16 sets the machining conditions based on the machining surface quality data described later. The machining condition setting unit 16 sends an instruction for instructing the machining conditions for controlling the drive unit 12 etc. during electric discharge machining to the machine control unit 14, and sends an instruction for instructing the machining conditions for controlling the power supplied between the electrode E and the workpiece 17 during electric discharge machining to the power supply control unit 15.

[0038] The machine control unit 14 controls the drive unit 12 etc. based on the instruction sent from the machining condition setting unit 16. The power supply control unit 15 controls the power supplied between the electrode E and the workpiece 17 based on the instruction sent from the machining condition setting unit 16.

[0039] The drive unit 12 moves in the X direction, Y direction, and Z direction according to the instruction sent from the machine control unit 14. In the present embodiment, the case where the Z direction is the vertical direction and the XY plane is the horizontal plane will be described.

[0040] The display unit 13 displays various information sent from the machining condition setting unit 16. An example of the display unit 13 is a liquid crystal monitor. The display unit 13 displays, for example, the machining surface quality data calculated by the machining condition setting unit 16 based on the image of the machining surface described later, the value of the evaluation index, the machining conditions corrected by the machining condition setting unit 16, etc. In addition, the display unit 13 can also display the information sent from the machine control unit 14 and the power supply control unit 15.

[0041] Figure 2 is a diagram showing the structure of the control device according to Embodiment 1. In Figure 2 it, on the basis of the structural elements of the control device 2, namely the machining condition setting unit 16, the machine control unit 14, and the power supply control unit 15, the display unit 13 is also shown. Figure 3 is a diagram showing the structure of the learning device according to Embodiment 1. Figure 4 is a diagram showing the structure of the inference device according to Embodiment 1.

[0042] The machining condition setting unit 16 includes an input unit 21, a machined surface quality data calculation unit 22, a learning device 23, a trained model storage unit 24, an inference device 25, a machining condition storage unit 26, and a machining condition output unit 27. In the electric discharge machine 1, a machining condition correction unit 28 for correcting machining conditions is constituted by the machined surface quality data calculation unit 22, the learning device 23, the trained model storage unit 24, and the inference device 25.

[0043] The input unit 21 receives machining object information, machining conditions, and image information of the machined surface used when the workpiece 17 is subjected to electric discharge machining. The image information of the machined surface is information of an image of the machined surface. The input unit 21 sends the machining object information and the machining conditions to the learning device 23 and the machining condition storage unit 26. In addition, the input unit 21 sends the machining object information to the inference device 25 and the machining condition output unit 27. In addition, the input unit 21 sends the image information of the machined surface to the machined surface quality data calculation unit 22.

[0044] The machining object information, the machining conditions, and the image information of the machined surface are input to the input unit 21 from the outside of the electric discharge machine 1. In addition, the electric discharge machine 1 may have an image information acquisition unit (not shown) that obtains the image information of the machined surface by photographing the machined surface of the workpiece 17 after electric discharge machining. In this case, the image information of the machined surface obtained by the image information acquisition unit is input to the input unit 21. A photographing device such as a camera is used in the image information acquisition unit. In addition, the image information of the machined surface may be obtained by an operator using a photographing device such as a camera to photograph the machined surface of the workpiece 17 removed from the electric discharge machine 1 after electric discharge machining. In this case, the image information of the machined surface obtained by the operator using the photographing device is input to the input unit 21.

[0045] The machining object information is information including information related to the target processed product obtained by electric discharge machining of the workpiece 17 and information related to the electrode E, which is the tool used in the electric discharge machine 1. In addition, the machining object information is information of parameters that cannot be directly controlled by the electric discharge machine 1 among the information used in the electric discharge machining of the workpiece 17. For example, the machining object information includes the material of the workpiece 17, that is, the machining material, the shape of the workpiece 17 before machining and the shape of the processed product, the shape of the electrode E, the material of the electrode E, the target value of the surface roughness of the machined surface, the shape of the discharge mark, and the target value of the discharge mark diameter.

[0046] The processing conditions are the conditions used when machining a workpiece 17 for the purpose of shape engraving electrical discharge machining. In addition, the processing conditions are the conditions used during the electrical discharge machining of the workpiece 17 among the information used in the electrical discharge machining, and are the setting information of parameters that can be directly controlled by the electrical discharge machining machine 1. That is, the processing conditions are the control conditions given to the mechanical control unit 14 and the power supply control unit 15, and are the power supply control conditions and mechanical control conditions generated by inputting the workpiece information to the inference device 25. The processing conditions include, for example, the electrical conditions of the high-frequency pulses used in the processing and the position conditions that define the position of the electrode E.

[0047] The electrical conditions include the pulse width, the rest time of the pulse, the voltage value, the current value, etc. The lift action conditions include the moving height of the electrode E, the moving speed of the electrode E, the moving acceleration of the electrode E, the lifting path of the electrode E, the return position of the electrode E, and the downtime, which is the time interval from the lift action of the electrode E to the next lift action.

[0048] The position conditions include the X coordinate, Y coordinate, and Z coordinate of the electrode E. The X coordinate and Y coordinate correspond to the machining position derived from the workpiece information, and the Z coordinate corresponds to the machining depth derived from the workpiece information. The position conditions include the machining position, the machining depth, the conditions for the approaching movement amount when machining by combining multiple processing conditions, the conditions for the machining chip discharge action, i.e., the lift action conditions, and the drive control conditions required for machining. The drive control conditions required for machining are the setting conditions of the axis of the electrode E during machining other than during the lift action, and are the setting conditions of the axis of the electrode E in the inter-electrode voltage control.

[0049] In addition, information other than the above used in the electrical discharge machining of the workpiece 17 can also be input to the input unit 21. Depending on the machining content, the number of electrode roots of the electrode E, the information of the electrode E, the information of the workpiece 17, the information for correcting the position of the electrode E, the moving method of the electrode E, the lift action method, etc. are different, so various information is input to the input unit 21 for each machining content.

[0050] The position information used in the processing conditions is specified by the positions in the three-axis directions of the X-axis, Y-axis, and Z-axis. The machining position is the position of the electrode E in the XY plane. The machining position corresponds to the position of the upper surface of the workpiece 17 and is represented by the center position of the electrode E corresponding to the center position of the workpiece 17. The machining depth is the depth of machining the workpiece 17. The machining depth is represented by the distance from the upper surface of the workpiece 17 before machining.

[0051] The machining condition storage unit 26 stores machining conditions associated with the workpiece information. The machining condition storage unit 26 updates the stored machining conditions with the machining conditions transmitted from the inference device 25 associated with the same workpiece information. The machining condition storage unit 26 updates the machining conditions associated with the same workpiece information each time machining conditions are transmitted from the inference device 25.

[0052] The machining condition output unit 27 reads out from the machining condition storage unit 26 the machining conditions corresponding to the workpiece information transmitted from the input unit 21. That is, the machining condition output unit 27 reads out from the machining condition storage unit 26 the latest updated machining conditions associated with the workpiece information. The machining condition output unit 27 outputs the position conditions among the machining conditions to the machine control unit 14 and outputs the electrical conditions among the machining conditions to the power supply control unit 15.

[0053] In profile EDM, while machining in the Z direction, machining is also performed with movement in the XY plane. In this case, the larger the reduction amount of the electrode E, the larger the movement amounts of the electrode E in the X and Y directions during the movement machining in the XY plane. The electrode E during the movement machining moves from the center position in the XY plane of the workpiece 17, that is, the center position at the initial position of the electrode E in the XY plane, in the positive X direction, positive Y direction, negative X direction, or negative Y direction. In profile EDM, for example, movement machining in the two-dimensional plane direction perpendicular to the depth direction can be performed simultaneously with the machining in the depth direction. In addition, in profile EDM, movement machining in any direction can be performed simultaneously with the machining in the depth direction. In EDM, for example, movement machining can be performed in a direction along the surface of a hemisphere or the like.

[0054] The machined surface quality data calculation unit 22 calculates the machined surface quality data based on an image obtained by photographing the machined surface 17a of the workpiece 17 after electrical discharge machining. The machined surface quality data calculation unit 22 sends the machined surface quality data to the learning device 23. Hereinafter, the image obtained by photographing the machined surface 17a of the workpiece 17 after electrical discharge machining is sometimes referred to as a machined surface image. The machined surface image is an image obtained by photographing the machined surface 17a of the workpiece 17 removed from the electrical discharge machine 1 after the electrical discharge machining is completed. The machined surface image is an image capable of discriminating the size of the discharge marks formed on the machined surface 17a of the workpiece 17 after electrical discharge machining, that is, the diameter of the discharge marks. In addition, the machined surface image does not necessarily have to photograph the entire machined surface 17a of the workpiece 17 after electrical discharge machining, and as long as the machined surface 17a is uniformly obtained without uneven aggregation, a part of the machined surface 17a can be uniformly spaced apart. Further, in the case of a region where the machined surface 17a has a symmetric shape, if one region among the symmetric shape regions is obtained, the other region among the symmetric shape regions can be omitted. Here, the machined surface image can be an image capable of discriminating the depth of the discharge marks formed on the machined surface 17a. The depth is the depth from the surface of the machined surface 17a. In addition, if the size of the discharge marks increases, the depth of the discharge marks also becomes deeper.

[0055] The machined surface quality is the uniformity of the discharge marks formed on the machined surface 17a of the workpiece 17. More specifically, it is the uniformity of the surface roughness of the machined surface 17a and the uniformity of the size of the discharge marks formed on the machined surface 17a. That is, a state where the machined surface quality of the machined surface 17a is high is a state where the uniformity of the surface roughness of the machined surface 17a and the uniformity of the size of the discharge marks formed on the machined surface 17a are high. In addition, a state where the machined surface quality of the machined surface 17a is low is a state where the uniformity of the surface roughness of the machined surface 17a and the uniformity of the size of the discharge marks formed on the machined surface 17a are low, and is a state where the surface roughness of the machined surface 17a and the size of the discharge marks formed on the machined surface 17a are non-uniform.

[0056] The machined surface quality data is the evaluation result obtained by quantitatively evaluating the machined surface quality of the machined object 17 after electrical discharge machining on the machined surface 17a. That is, the machined surface quality data calculation unit 22 calculates and quantifies the machined surface quality of the machined surface 17a of the machined object 17 after electrical discharge machining based on the image of the machined surface 17a. In the present Embodiment 1, data obtained by quantifying the difference in the distribution of discharge mark diameters calculated based on the size and number of discharge marks formed on the machined surface 17a of the machined object 17 after electrical discharge machining is used as the machined surface quality data. In addition, whether to treat the discharge marks as pits depends on the required machined surface quality of the machined surface 17a, that is, the uniformity of the machined surface 17a.

[0057] Hereinafter, data obtained by quantifying the difference in the distribution of discharge mark diameters calculated based on the size and number of discharge marks formed on the machined surface 17a of the machined object 17 after electrical discharge machining is referred to as the first machined surface quality data. Here, for the machined object 17 with the same machined object information, the distribution of discharge mark diameters of the machined surface 17a of the machined object 17 after electrical discharge machining in the case of machining without non-uniformity in the machined surface 17a of the machining chips generated during machining, that is, the ideal distribution of discharge mark diameters, and the distribution of discharge mark diameters of the machined surface 17a of the machined object 17 after electrical discharge machining under any machining conditions corresponding to the machined object information, that is, the actual distribution of discharge mark diameters, are used as the first machined surface quality data. That is, the first machined surface quality data is machined surface quality data obtained by evaluating the machined surface quality of the machined surface 17a based on the distribution of discharge marks larger than a predetermined size in the machined surface 17a obtained from the image of the machined surface 17a after profile electric discharge machining.

[0058] Figure 5 It is a distribution diagram showing an example of the distribution of discharge mark diameters in the machined surface of the machined object after electrical discharge machining by the electrical discharge machining apparatus according to Embodiment 1 and the ideal distribution of discharge mark diameters. Hereinafter, an example of the distribution of discharge mark diameters of the machined surface 17a of the machined object 17 after electrical discharge machining by the electrical discharge machining apparatus 1 may be referred to as the distribution g(r). In addition, the ideal distribution of discharge mark diameters of the machined surface 17a of the machined object 17 after electrical discharge machining by the electrical discharge machining apparatus 1 may be referred to as the distribution f(r). Figure 5 The distribution f(r) and the distribution g(r) in are calculated based on the image obtained by photographing the machined surface 17a of the machined object 17 after electrical discharge machining.

[0059] In Figure 5In FIG, the horizontal axis shows the diameter of the discharge scar formed on the machined surface 17a of the workpiece 17 obtained from the machined surface image of the workpiece 17 after the discharge machining. The discharge scar diameter here is the diameter of the discharge scar formed on the machined surface 17a of the workpiece 17. Figure 5 In FIG. 1 , the vertical axis indicates the number of discharge marks formed on the machined surface 17 a of the workpiece 17 obtained from the machined surface image of the workpiece 17 after the electrical discharge machining.

[0060] The distribution f(r) is the distribution of discharge scar diameters when machining the machined surface 17a of the workpiece 17 without unevenness in the machined surface 17a of machining chips generated during machining, for each electrical condition. Figure 5 In the figure, the distribution f(r) is shown by a solid line. Figure 5 In FIG, the distribution g(r) is shown by a dotted line. The discharge trace diameter of the distribution g(r) is less than or equal to g0 and has the same distribution as the distribution f(r).

[0061] When the machined surface 17a of the workpiece 17 is machined to a high quality, the distribution of the discharge scar diameter on the machined surface 17a becomes Figure 5 The normal distribution is shown by the solid line in the figure. Figure 5 The center of the discharge crater diameter in the distribution f(r) is the discharge crater diameter setting value c, which is set in advance as the discharge crater diameter generated during the electrical discharge machining of the workpiece 17. In other words, the discharge crater diameter setting value c is the ideal discharge crater diameter assumed to occur on the machined surface 17a under the machining conditions used.

[0062] On the other hand, when the machined surface 17a of the workpiece 17 is machined with poor surface quality, the distribution of the discharge mark diameter on the machined surface 17a is as follows: Figure 5 As shown by the middle dashed line distribution g(r), the number of discharge traces in the region with a large discharge trace diameter increases, and the discharge traces with a large discharge trace diameter are dispersed. Figure 5 In the region (i) of the distribution g(r), the number of discharge traces with large diameters in the distribution f(r) is relatively large, and the discharge traces with larger diameters than those in the distribution f(r) are present at the end of the distribution f(r). Figure 5 In the region (ii) of the distribution g(r), discharge traces having a larger diameter than those in the region (i) are locally concentrated and dispersed.

[0063] In profile grinding electrical discharge machining, when machining is performed under machining conditions where the heat energy generated by the discharge is weaker, the surface roughness of the machining surface 17a and the size of the discharge marks become smaller. However, when machining is performed under machining conditions where the heat energy generated by the discharge is weak, the machining time becomes longer. Therefore, generally, machining is started under machining conditions where the heat energy generated by the discharge is strong. Then, the machining conditions are gradually changed to machining conditions where the heat energy generated by the discharge is weak. Moreover, finally, the desired surface roughness can be obtained, and machining is performed under machining conditions where the heat energy generated by the discharge is weak.

[0064] In addition, in profile grinding electrical discharge machining, the stronger the heat energy generated by the discharge, the greater the gap between the electrode E and the workpiece 17. Therefore, when switching from machining conditions where the heat energy generated by the discharge is strong to machining conditions where the heat energy generated by the discharge is weak, it is necessary to make the distance between the electrode E and the workpiece 17 close to a value greater than or equal to the difference between the gap between the electrode E and the workpiece 17 under machining conditions where the heat energy generated by the discharge is strong and the gap between the electrode E and the workpiece 17 under machining conditions where the heat energy generated by the discharge is weak.

[0065] As described above, when switching from the gap between the electrode E and the workpiece 17 under machining conditions where the heat energy generated by the discharge is strong to the gap between the electrode E and the workpiece 17 under machining conditions where the heat energy generated by the discharge is weak, the amount by which the electrode E is moved toward the workpiece 17, that is, the amount by which the distance between the electrode E and the workpiece 17 is made close, is called the approach movement amount. When the approach movement amount is insufficient, the discharge marks formed on the machining surface 17a under machining conditions where the heat energy generated by the discharge is strong will remain. In this case, discharge marks with a discharge mark diameter larger than the discharge mark diameter formed on the machining surface 17a under machining conditions where the heat energy generated by the discharge is weak remain, and the surface roughness of the machining surface 17a becomes rough.

[0066] Therefore, in profile grinding electrical discharge machining, when the approach movement amount at the time of switching machining conditions during machining in which the voltage value and current value of the high-frequency pulse voltage in the machining conditions are adjusted to gradually weaken the heat energy generated by the discharge is inappropriate, the discharge marks formed under the machining conditions before switching the machining conditions remain, resulting in Figure 5 the pits in region (i) of the distribution g(r) therein.

[0067] In addition, when at least one of the conditions of the high-frequency pulse applied between the electrode E and the workpiece 17 and the machining chip discharge conditions is inappropriate, concentrated discharge occurs, thereby forming a discharge mark having a discharge mark diameter extremely larger than the discharge mark diameter setting value c, resulting in Figure 5The pit in region (ii) of the distribution g(r). The diameter of the discharge trace of the pit as described above sometimes reaches 10 to 20 times the discharge trace diameter setting value c. As described above, in the pit formed on the machining surface 17a, there are pits with different main causes and sizes.

[0068] Therefore, like the distribution f(r) and the distribution g(r), the machining surface quality data calculation unit 22 generates a distribution map of the discharge trace diameters of the discharge traces formed on the machining surface 17a based on the machining surface image, and calculates the first machining surface quality data according to the distribution map of the discharge trace diameters of the discharge traces. The machining surface quality data calculation unit 22 uses the following formula (1) to calculate the first machining surface quality data according to the Figure 5 distribution map of. Figure 5 distribution map.

[0069]

Formula 1

[0070] First machining surface quality data (S) = ∫{r × (g(r) - f(r))}dr…(1)

[0071] In formula (1), r is the difference between the discharge trace diameter of g0 and later in the distribution g(r) and the discharge trace diameter setting value c. In addition, the operation of "g(r) - f(r)" obtains the difference between the distribution of the discharge trace diameter in the case where the distribution of the discharge trace diameter is not ideal and the distribution of the discharge trace diameter in the case where the distribution of the discharge trace diameter is ideal. For example, consider the part where the distribution curve of f(r) touches the horizontal axis, that is, the position where the discharge trace diameter is g1. For example, if the number of discharge trace diameters of g(r) at the position of g1 is 50 and the number of discharge trace diameters of f(r) at the position of g1 is 0, then g(r) - f(r) = 50.

[0072] The value of the first machining surface quality data calculated by formula (1) becomes a value that is larger when the machining surface quality of the machining surface 17a is worse and smaller when the machining surface quality of the machining surface 17a is better.

[0073] The pits with large discharge trace diameters are generated in the state of abnormal electrical discharge machining. The pits with small discharge trace diameters are generated when the approach movement amount is inappropriate. As described above, there are pits with different main causes and sizes on the machining surface 17a. The value of the first machining surface quality data calculated by formula (1) becomes a value that is larger when the machining surface quality of the machining surface 17a is worse. Therefore, when the value of the first machining surface quality data is extremely large, it is predicted that the state of the electrical discharge machining machine 1 itself is abnormal, or the machining conditions are significantly inappropriate. Therefore, by optimizing the state of the electrical discharge machining machine 1 itself or the machining conditions, the pits in region (ii) with a significantly large discharge trace diameter can be eliminated.

[0074] In addition, when the value of the first processed surface quality data is not extremely large, the learning device 23 performs optimization for reducing the relatively small pits in region (i) caused by inappropriate approach movement amount, thereby suppressing the generation of pits and improving the uniformity of the processed surface 17a.

[0075] Furthermore, the processed surface quality data is the value of an evaluation index representing the evaluation of the processing result of the processed surface 17a of the workpiece 17. Therefore, the processed surface quality data calculation unit 22 can be said to be an evaluation calculation unit that calculates the value of the evaluation index representing the evaluation of the processing result of the processed surface 17a of the workpiece 17. The evaluation of the processing result of the processed surface 17a of the workpiece 17 is the evaluation of the uniformity of the discharge marks formed on the processed surface 17a of the workpiece 17 after form engraving electric discharge machining. Therefore, the processed surface quality data is the value of an evaluation index representing the evaluation of the uniformity of the discharge marks formed on the processed surface 17a of the workpiece 17 after form engraving electric discharge machining.

[0076] The processed surface quality data calculation unit 22 sends the calculated first processed surface quality data to the learning device 23. In addition, the processed surface quality data calculation unit 22 can be arranged inside the learning device 23 or outside the electric discharge machine 1.

[0077] The learning device 23 learns the processing conditions that can achieve the target processed surface quality based on the processing conditions used in actual processing and the processing results obtained by electric discharge machining using these processing conditions. In this embodiment, the case where the processing result is the processed surface quality of the processed surface 17a is described.

[0078] <Learning stage>

[0079] The learning device 23 is a computer that generates a trained model 30 based on the processing conditions used in actual processing and the processed surface quality data obtained through actual processing, and learns the processing conditions that can improve the processed surface quality of the processed surface 17a. The trained model storage unit 24 stores the trained model 30 generated by the learning device 23. In addition, the trained model storage unit 24 can be arranged inside the learning device 23. Additionally, the learning device 23 and the trained model storage unit 24 can also be arranged outside the electric discharge machine 1.

[0080] The learning device 23 has a data acquisition unit 231 and a model generation unit 232.

[0081] The data acquisition unit 231 acquires workpiece information, processing conditions, and processed surface quality data as learning data. In Embodiment 1, the case where the data acquisition unit 231 acquires workpiece information, processing conditions, and the first processed surface quality data as learning data is described.

[0082] The model generation unit 232 learns the processing conditions based on the learning data including the workpiece information, the processing conditions, and the first processed surface quality data. That is, the model generation unit 232 generates a trained model 30 for inferring the processing conditions that improve the processed surface quality of the workpiece 17 according to the workpiece information and the processed surface quality data.

[0083] The learning algorithm used by the model generation unit 232 can use known algorithms such as supervised learning, unsupervised learning, and reinforcement learning. As an example, the case of applying reinforcement learning is described. In reinforcement learning, an agent (acting entity) in a certain environment observes the current state (parameters of the environment) and decides the action to be taken. The environment changes dynamically through the action of the agent, and a reward is given to the agent corresponding to the change of the environment. The agent repeats and learns the action policy that maximizes the reward through a series of actions. As representative methods of reinforcement learning, Q-learning and TD-learning are known. For example, in the case of Q-learning, the general update formula of the action value function Q(s, a) is represented by the following formula (2).

[0084]

Formula 2

[0085]

[0086] In formula (2), s t represents the environment at time t, and a t represents the action at time t. Through the action a t , the state (environment) becomes s t+1 . r t+1 represents the reward brought by the change of its state, γ represents the discount rate, and α represents the learning coefficient. In addition, γ is in the range of 0 < γ ≤ 1, and α is in the range of 0 < α ≤ 1. The processing conditions become the action a t , the workpiece information and the processed surface quality data become the state s t , and the learning device 23 learns the best action a t under the state s t at time t.

[0087] The update formula represented by formula (2) is that if the action value Q of the action a with the highest Q value at time t+1 is greater than the action value Q of the action a executed at time t, the action value Q is increased, and in the opposite case, the action value Q is decreased. In other words, the learning device 23 updates the action value function Q(s, a) in such a way that the action value Q of the action a at time t approaches the best action value Q at time t+1. Thus, the best action value Q in a certain environment is continuously propagated in sequence as the action value Q in its previous environment.

[0088] As described above, when the model generation unit 232 generates the trained model 30 through reinforcement learning, the model generation unit 232 includes a reward calculation unit 233 and a function update unit 234.

[0089] The reward calculation unit 233 calculates the reward r for the processing conditions based on the processing conditions and the value of the evaluation index, that is, the processed surface quality data. For example, the reward calculation unit 233 increases the reward r when the processed surface quality data is small (for example, gives a reward of "1"), and on the contrary, decreases the reward r when the processed surface quality data is large (for example, gives a reward of "-1"). The reward calculation unit 233 stores a first threshold 31, which is a reference value for determining whether to increase or decrease the reward for the processing conditions for the application by comparing with the first processed surface quality data. The first threshold value 31 is set by the user in the control device 2, and thus can be changed to any value.

[0090] The function update unit 234 updates the function for determining the next processing conditions, that is, the action, based on the reward. That is, the function update unit 234 updates the trained model 30. The function update unit 234 updates the function for determining the processing conditions according to the reward calculated by the reward calculation unit 233 and outputs it to the trained model storage unit 24. For example, in the case of Q-learning, the function update unit 234 uses the action value function Q(s t , a t ) represented by formula (2) as the function for calculating the processing conditions. The action value function Q(s t , a t ) can be said to be a processing condition generation function for calculating the processing conditions.

[0091] The learning device 23 repeatedly executes the above learning. The trained model storage unit 24 stores the action value function Q(s t , a t ) updated by the function update unit 234, that is, the trained model 30.

[0092] Next, use Figure 6, the processing order of the learning process performed by the learning device 23 will be described. Figure 6 It is a flowchart showing the processing order of the learning process performed by the learning device according to Embodiment 1.

[0093] The data acquisition unit 231 acquires the workpiece information, processing conditions, and the first processed surface quality data as the first learning data (step S110).

[0094] The model generation unit 232 calculates the return for the applied processing conditions based on the workpiece information, processing conditions, and the first processed surface quality data (step S120). Specifically, the return calculation unit 233 acquires the first processed surface quality data and determines whether to increase or decrease the return for the applied processing conditions based on a predetermined return criterion, i.e., the first threshold 31. The return calculation unit 233 determines that the return for the applied processing conditions should be increased when the first processed surface quality data is less than the first threshold 31. The return calculation unit 233 determines that the return for the applied processing conditions should be decreased when the first processed surface quality data is greater than or equal to the first threshold 31.

[0095] The return calculation unit 233 increases the return when the first processed surface quality data is less than the first threshold 31 (step S130). On the other hand, the return calculation unit 233 decreases the return when the first processed surface quality data is greater than or equal to the first threshold 31 (step S140).

[0096] The function update unit 234 updates the action value function Q(s t , a t ) represented by Equation (2) stored in the trained model storage unit 24 based on the return calculated by the return calculation unit 233 (step S150).

[0097] The learning device 23 repeatedly executes the steps from step S110 to step S150 above, and stores the generated action value function Q(s t , a t ) as the trained model 30 in the trained model storage unit 24.

[0098] In addition, the learning device 23 according to Embodiment 1 stores the trained model 30 in the trained model storage unit 24 provided outside the learning device 23, but the trained model storage unit 24 may also be configured inside the learning device 23.

[0099] In the evaluation point of the processing result in the first embodiment, that is, the first processed surface quality data, the larger the discharge mark diameter compared to the distribution f(r) in the distribution g(r), the larger the value. In addition, the first processed surface quality data is such that the larger the number of discharge marks with a relatively large discharge mark diameter in the distribution f(r) increases in the distribution g(r), the larger the value. Therefore, the learning device 23 generates a trained model 30 that reduces the action value function Q(s t , a t ), thereby enabling learning of the relationship between the workpiece information for inferring the processing conditions and the processing conditions, where the processing conditions are those that can improve the processed surface quality of the workpiece 17, that is, the processing conditions that can improve the uniformity of the processed surface 17a.

[0100] <Effective use stage>

[0101] The inference device 25 is a computer that uses the trained model 30 to infer the processing conditions for improving the processed surface quality of the processed surface 17a based on the workpiece information.

[0102] The inference device 25 includes a data acquisition unit 251 and an inference unit 252. The data acquisition unit 251 acquires workpiece information. The inference unit 252 uses the trained model 30 to infer the processing conditions and outputs the inferred processing conditions as the processing conditions 32 to the processing condition storage unit 26. That is, the inference unit 252 can infer the processing conditions suitable for the workpiece information by inputting the workpiece information acquired by the data acquisition unit 251 into the trained model 30. In addition, the inference device 25 can also be arranged outside the electric discharge machine 1.

[0103] In addition, in the first embodiment, the case where the inference device 25 uses the trained model 30 learned by the model generation unit 232 to infer the processing conditions has been described, but the inference device 25 can also obtain the trained model 30 from other learning devices other than the learning device 23 and infer the processing conditions based on the trained model 30.

[0104] Next, use Figure 7 , to describe the processing sequence of the process of inferring the processing conditions by the inference device 25. Figure 7 is a flowchart showing the processing sequence of the inference process performed by the inference device according to the first embodiment and the control process performed by the control device according to the first embodiment.

[0105] The data acquisition unit 251 acquires the workpiece information as the inference data (step S210). The inference unit 252 inputs the inference data, i.e., the workpiece information, into the trained model 30 stored in the trained model storage unit 24 (step S220) to obtain the processing conditions. The inference unit 252 outputs the obtained data, i.e., the processing conditions, to the processing condition storage unit 26 (step S230).

[0106] The processing condition output unit 27 reads out the processing conditions corresponding to the workpiece information from the processing condition storage unit 26. The processing condition output unit 27 outputs the position conditions among the read processing conditions to the machine control unit 14, and outputs the electrical conditions among the read processing conditions to the power supply control unit 15. The machine control unit 14 and the power supply control unit 15 control the electrical discharge machining using the output processing conditions (step S240). Thus, the electrical discharge machining machine 1 can improve the machining surface quality of the machining surface 17a.

[0107] In addition, in the first embodiment, the case where reinforcement learning is applied to the learning algorithm used in the inference unit 252 is described, but it is not limited thereto. Regarding the learning algorithm, in addition to reinforcement learning, supervised learning, unsupervised learning, or semi-supervised learning can also be applied.

[0108] In addition, as the learning algorithm used in the model generation unit 232, deep learning for learning the extraction of the feature quantity itself can also be used. The model generation unit 232 can perform machine learning according to other known methods, such as neural networks, genetic programming, functional logic programming, support vector machines, etc.

[0109] In addition, the learning device 23 and the inference device 25 can be, for example, devices separate from the control device 2 connected to the control device 2 via a network. In addition, the learning device 23 and the inference device 25 can also be built into the control device 2. And the learning device 23 and the inference device 25 can also exist on a cloud server.

[0110] In addition, the model generation unit 232 can use the learning data obtained from multiple control devices 2 to learn the processing conditions. In addition, the model generation unit 232 can obtain the learning data from multiple control devices 2 used in the same area, or can use the learning data collected from multiple control devices 2 operating independently in different areas to learn the processing conditions. In addition, the learning device 23 can also add the control device 2 that collects the learning data to the object midway, or remove it from the object. And the learning device 23 for learning the processing conditions of a certain control device 2 can also be applied to other control devices 2 different from the control device 2, and the trained model 30 can be updated by re-learning the processing conditions for the other control devices 2.

[0111] As described above, in the electric discharge machine 1 according to the first embodiment, the learning device 23 learns the machining conditions, and the inference device 25 infers the machining conditions. Therefore, the workload of parameter adjustment of the machining conditions by the user can be eliminated, and the machining surface quality of the machining surface 17a of the workpiece 17 can be improved.

[0112] In the case of profile electric discharge machining, if the workpiece information is different, the machining conditions for obtaining good machining surface quality are different. Therefore, a large amount of machining results and the experience of skilled workers are required to generate machining conditions for obtaining a machining surface 17a with good machining surface quality. However, according to the first embodiment, machining conditions for easily obtaining a machining surface 17a with good machining surface quality can be generated.

[0113] In addition, by effectively using the first machining surface quality data, which is the machining surface quality data obtained by evaluating the machining surface quality of the machining surface 17a based on the distribution of the pits in the machining surface 17a, it is possible to learn the appropriate conditions for the approaching movement amount of the machining conditions highly correlated with the distribution of the pit sizes and the appropriate conditions for the high-frequency pulse conditions in the electric discharge machining.

[0114] Therefore, according to the first embodiment, a learning device capable of learning the appropriate machining conditions for electric discharge machining that can improve the machining surface quality of the machining surface 17a of the workpiece 17 is obtained.

[0115] Second Embodiment.

[0116] Figure 8 FIG. is a diagram showing the structure of the learning device of the electric discharge machine in the second embodiment. In the second embodiment, other learning methods in the learning device 23 will be described.

[0117] In the second embodiment, the model generation unit 232 learns the machining conditions that can improve the machining surface quality of the machining surface 17a based on the learning data including the workpiece information, the machining conditions, and the second machining surface quality data described later. That is, the model generation unit 232, in the same manner as in the first embodiment, generates a trained model 30 for inferring the machining conditions that can improve the machining surface quality of the machining surface 17a according to the workpiece information and the machining surface quality data.

[0118] In the second embodiment, the model generation unit 232 stores a second threshold value 33, which is a reference value for comparing with the second machining surface quality data to determine whether to increase or decrease the reward for the applied machining conditions. The second threshold value 33 is set by the user in the control device 2, and thus can be changed to any value.

[0119] If profile-discharge machining is performed in a machining fluid using the electric discharge machine 1, machining chips are generated between the machining surface 17a of the workpiece 17 and the electrode E. Moreover, the distribution of the machining chips in the machining fluid existing between the machining surface 17a and the electrode E in the in-plane direction of the machining surface 17a becomes non-uniform, and local agglomeration of the machining chips occurs in the in-plane direction of the machining surface 17a, thereby causing non-uniformity in the machining surface quality of the machining surface 17a.

[0120] That is, if the concentration of the machining chips in the machining fluid existing between the machining surface 17a and the electrode E increases to be greater than or equal to a certain specific concentration, a state where the potential between the machining surface 17a and the electrode E excessively decreases is formed. Moreover, at a portion where the concentration of the machining chips in the machining fluid between the machining surface 17a and the electrode E is greater than or equal to a certain specific concentration, the discharge between the machining surface 17a and the electrode E easily changes to an arc state, and the discharge easily concentrates on one portion. Therefore, at a portion where the concentration of the machining chips in the machining fluid between the machining surface 17a and the electrode E is greater than or equal to a certain specific concentration, the discharge mark diameter of the discharge mark becomes larger due to the concentration of the discharge, and a crater is formed.

[0121] Therefore, in order to stably perform profile-discharge machining in a machining fluid using the electric discharge machine 1, it is necessary to keep the potential between the machining surface 17a and the electrode E constant. Therefore, when performing profile-discharge machining in a machining fluid using the electric discharge machine 1, machining is performed while periodically performing a machining chip discharging operation, i.e., a lifting operation, to discharge the machining chips from the machining fluid between the machining surface 17a and the electrode E.

[0122] Figure 9 FIG. is a diagram for explaining the inter-electrode distance control and the lifting operation control in the profile-discharge machining in a machining fluid performed by the electric discharge machine in Embodiment 2. When performing the inter-electrode distance control, the electric discharge machine 1 brings the electrode E close to the machining surface 17a of the workpiece 17 in the machining fluid 43, and generates a discharge pulse 41 between the electrode E and the machining surface 17a (s1). The machining surface 17a is subjected to electric discharge machining by this discharge pulse 41, and machining chips 42 are generated.

[0123] Then, the electric discharge machine 1 finishes the inter-electrode distance control and starts the lift motion control. In this case, the electric discharge machine 1 raises the electrode E by the lift motion distance L1 from the machining surface 17a (s2). Then, the electric discharge machine 1 lowers the electrode E to return the electrode E to a certain position of the workpiece 17 (s3). Here, when the electrode E is lowered, the machining fluid 43 containing the machining chips 42 flows out from the machining completed area 45 toward the outside of the machining completed area 45. Moreover, by the outflow action of the machining fluid 43 at this time, the machining chips are discharged from the space between the electrode E and the machining surface 17a. Through the lift motion as described above, the machining chips 42 are discharged from the space between the electrode E and the machining surface 17a. The electric discharge machine 1 performs profile electric discharge machining in the machining fluid 43 while periodically repeating the above processes (s1) to (s3). The action of raising the electrode E to discharge the machining chips 42 from the machining fluid 43 existing between the machining surface 17a and the electrode E as described above and then returning the electrode E to the original position is called the lift motion. The lift motion is not limited to the motion of the electrode E in the machining direction. The lift motion includes, for example, the action of moving the electrode E in the XY plane during machining in the Z direction as the machining direction and then returning the electrode E to the original position after moving the electrode E in the XY direction.

[0124] There is a machining chip discharge effect suitable for each machining shape in the lift motion. The machining chip discharge effect is the effect of reducing the machining chip concentration in the machining fluid 43 existing between the machining surface 17a and the electrode E after the lift motion ends. After the lift motion ends, it can be said that the lower the machining chip concentration in the machining fluid 43 between the machining surface 17a and the electrode E, the higher the discharge effect of the machining chips 42 in the lift motion, that is, the higher the machining chip discharge effect.

[0125] Figure 10 It is a characteristic diagram showing the relationship between the non-uniformity of the distribution of machining chips between the machining surface and the electrode in the in-plane direction of the machining surface in the profile electric discharge machining in the machining fluid performed by the electric discharge machine of Embodiment 2 and the machining chip discharge effect of the lift motion. In addition, in Figure 10 , the relationship between the machining chip discharge effect of the lift motion and the amount of machining chips in the machining fluid existing between the machining surface 17a of the workpiece 17 and the electrode E immediately after the lift motion ends is also shown. Hereinafter, the distribution of machining chips between the machining surface 17a and the electrode E in the in-plane direction of the machining surface 17a may sometimes be simply referred to as the distribution of machining chips.

[0126] Figure 10The horizontal axis in [the figure] shows the machining chip discharge effect of the lifting operation. In the horizontal axis, on the left side, the machining chip discharge effect of the lifting operation is low, and the amount of machining chips in the machining fluid existing between the machining surface 17a and the electrode E is relatively large. In addition, in the horizontal axis, on the right side, the machining chip discharge effect of the lifting operation is high, and the amount of machining chips in the machining fluid existing between the machining surface 17a and the electrode E is relatively small. Figure 10 The left vertical axis in [the figure] shows the non-uniformity of the distribution of machining chips between the machining surface and the electrode E in the in-plane direction of the machining surface. In the left vertical axis, the lower the position, the more uniform the distribution of machining chips, and the higher the position, the more non-uniform the distribution of machining chips.

[0127] Figure 10 The right vertical axis in [the figure] shows the amount of machining chips in the machining fluid between the machining surface 17a and the electrode E immediately after the lifting operation ends. In the right vertical axis, the lower the position, the less the amount of machining chips, and the higher the position, the more the amount of machining chips. In Figure 10 [the figure], the characteristic curve representing the relationship between the non-uniformity of the distribution of machining chips and the machining chip discharge effect of the lifting operation is represented by a solid line 46. In Figure 10 [the figure], the characteristic curve representing the relationship between the amount of machining chips in the machining fluid between the machining surface 17a and the electrode E immediately after the lifting operation ends and the machining chip discharge effect of the lifting operation is represented by a dashed line 47.

[0128] As Figure 10 shown, there are regions where the distribution of machining chips becomes uniform both in the region with a high machining chip discharge effect of the lifting operation and in the region with a low machining chip discharge effect of the lifting operation. There are regions where there is a tendency for the distribution of machining chips to become non-uniform both in the region with an excessively high machining chip discharge effect of the lifting operation and in the region with an excessively low machining chip discharge effect of the lifting operation.

[0129] In order to suppress the non-uniformity of the distribution of machining chips, there is an appropriate value for the machining chip discharge effect of the lifting operation. In addition, the state between the machining surface 17a and the electrode E in a state where the non-uniformity of the distribution of machining chips is suppressed is, for example, as Figure 10 shown, there are state A, state B, state C, and state D. Here, an example is shown in the case where the shape of the machining portion of the electrode E, that is, the shape of the portion of the electrode E facing the machining surface 17a, is a quadrilateral shape. In addition, the shape of the machining portion of the electrode E is not limited to a quadrilateral shape, and if it can machine the desired machining surface 17a, it can be set to any shape. Therefore, the machining shape of the machining surface 17a is not limited to a quadrilateral shape.

[0130] In states A, B, C, and D, the shaded area 51 has a relatively large amount of machining chips in the machining fluid existing between the machining surface 17a of the workpiece 17 and the electrode E immediately after the lifting operation, compared to the non-shaded area 52.

[0131] State A is a state in which a relatively large amount of machining chips compared to state D are uniformly distributed between the machining surface 17a and the electrode E immediately after the lifting operation. State A corresponds to the state of the portion 46A between the first point 46A1 and the second point 46A2 across the left lowest point 46AL at the solid line 46 in Figure 10 The first point 46A1 is a point to the left of the left lowest point 46AL at the solid line 46. The second point 46A2 is a point to the right of the left lowest point 46AL at the solid line 46 and is a point between the uppermost point 46T and the left lowest point 46AL at the solid line 46. In addition, the first point 46A1 and the second point 46A2 vary according to the respective conditions in the profile electric discharge machining. State D is a state in which a relatively very small amount of machining chips compared to state A are uniformly distributed between the machining surface 17a and the electrode E immediately after the lifting operation, or a state in which the machining chips between the machining surface 17a and the electrode E are completely removed. State D corresponds to the state of the portion 46D between the third point 46D1 and the fourth point 46D2 across the right lowest point 46DL at the solid line 46 in Figure 10 The third point 46D1 is a point to the left of the right lowest point 46DL at the solid line 46 and is a point between the uppermost point 46T and the right lowest point 46DL. The fourth point 46D2 is a point to the right of the right lowest point 46DL at the solid line 46. In addition, the third point 46D1 and the fourth point 46D2 vary according to the respective conditions in the profile electric discharge machining. Moreover, there is an appropriate level for the machining chip discharge effect of the lifting operation to achieve the respective states of state A and state D.

[0132] If state A is the target, the amount of machining chips discharged in the lifting operation with a low machining chip discharge effect of the lifting operation is small. Therefore, the machining chip concentration between the machining surface 17a and the electrode E during the electric discharge machining is too high, and the discharge between the machining surface 17a and the electrode E changes to an arc state and the electric discharge machining does not proceed.

[0133] In addition, when the chip discharge effect of the lifting operation is higher than that in state A, it becomes a state where there is a region with less chips locally between the machining surface 17a and the electrode E in the in-plane direction of the machining surface 17a, that is, state B or state C. In addition, if the chip discharge effect of the lifting operation is too high compared to the appropriate value of the chip discharge effect of the lifting operation in state D, it becomes a state where the discharge potential between the electrode E and the machining surface 17a is high, and it becomes a situation where pulsed discharge is difficult to occur. Therefore, during machining by weak pulsed discharge, the machining easily becomes unstable.

[0134] When the shape of the machining part of the electrode E is quadrilateral, it is a grade with a relatively low chip discharge effect of the lifting operation. When the chip discharge effect of the lifting operation is higher than that in state A, as shown in state B, due to the property of fluid fluidity, chips remain in an X-shaped line relative to the machining surface 17a. State B corresponds to the state of the part 46B between the second point 46A2 and the uppermost point 46T at the solid line 46 in Figure 10 . In addition, when the shape of the machining part of the electrode E opposite to the machining surface 17a is quadrilateral, it is a grade with a higher chip discharge effect of the lifting operation than that in state B. When the chip discharge effect of the lifting operation is lower than that in state D, as shown in state C, due to the property of fluid fluidity, chips remain at the corners relative to the machining surface 17a. State C corresponds to the state of the part 46C between the uppermost point 46T and the third point 46D1 at the solid line 46 in Figure 10 .

[0135] Even if the chip concentration in the machining fluid between the machining surface 17a and the electrode E is at the same level, due to the shape of the machining part of the electrode E and the chip discharge effect of the lifting operation, local chip segregation occurs relative to the machining surface 17a. Therefore, Figure 10 the horizontal axis of [] is not marked as chip concentration, but as the chip discharge effect of the lifting operation.

[0136] Moreover, in the state of state B or state C, the uniformity of the discharge trace diameter and the surface roughness of the part of the machining surface 17a corresponding to the pattern of the shaded area 51 after electrical discharge machining deteriorate, and the machining surface 17a becomes non-uniform.

[0137] The machining surface non-uniformity pattern 50 described below shows the segregation of the characteristic chip concentration distribution in the in-plane direction of the machining surface 17a that occurs during electrical discharge machining in the states of state B and state C. The machining surface non-uniformity pattern is a pattern representing the segregation of the characteristic chip concentration distribution indicating poor chip concentration distribution in the in-plane direction of the machining surface 17a that occurs during electrical discharge machining.

[0138] Figure 11 and Figure 12 is a diagram showing an example of the uneven machining surface pattern in Embodiment 2. In Figure 11 and Figure 12 , the pattern of the shaded area 53 is the pattern corresponding to the shaded area 51 of Figure 10 . In Figure 11 and Figure 12 , the non-shaded area 54 corresponds to the non-shaded area 52 of Figure 10 . The pattern of the shaded area 53 is the pattern in which the machining surface 17a becomes uneven.

[0139] Figure 11 's uneven machining surface pattern 50 is an example of patternizing the distribution of machining chips 42 that occurs when, as shown in Figure 9 , during the upward movement when the electrode E rises, the machining chips 42 concentrate in a diagonal shape of the four sides of the machining surface 17a due to the inflow of the machining fluid 43 from the four corners of the quadrilateral shape of the machining surface 17a to the machining completion area 45. Figure 11 The distribution of the machining chips 42 shown occurs when the machining chips 42 do not spread properly due to the outflow of the machining fluid 43 from the machining completion area 45 during the downward movement of the electrode E during the upward movement.

[0140] ​ 's uneven machining surface pattern 50 corresponds to state B. In the profile electric discharge machining where it is necessary to maintain the distribution of the machining chips 42 uniformly by setting the concentration of the machining chips in the machining fluid 43 between the machining surface 17a and the electrode E to an appropriate concentration, a state like state A is required as the distribution of the machining chips 42. State B occurs when the machining chip discharge effect of the upward movement is slightly higher than that of state A.

[0141] In addition, ​ 's uneven machining surface pattern 50 is an example of patternizing the distribution of machining chips that occurs when, with the aim of completely discharging or discharging as much as possible the machining chips 42 between the machining surface 17a and the electrode E as in state D, the machining chip discharge effect of the upward movement is insufficient and the machining chips concentrate on the outer peripheral side of the machining shape in the machining completion area 45.

[0142] In the present Embodiment 2, the machining surface quality data calculation unit 22 calculates the machining surface quality data based on an image obtained by photographing the machining surface 17a of the workpiece 17 after the electric discharge machining. ​ is a diagram showing the machining surface quality data calculation unit that stores the uneven machining surface pattern in Embodiment 2. In addition, the machining surface quality data calculation unit 22 stores in advance ​The uneven processed surface pattern shown ​ is a diagram showing the first processed surface quality data in the region of the processed surface image divided in Embodiment 2 ​ is a diagram showing a state where the first processed surface quality data and the image of the uneven processed surface pattern in the divided region of the processed surface image are overlapped in Embodiment 2

[0143] The processed surface quality data calculation unit 22 divides the processed surface image 60 into a plurality of divided regions 61 as shown ​ The processed surface quality data calculation unit 22 calculates the first processed surface quality data calculated by the method described in Embodiment 1 above for each divided region 61 divided in the processed surface image. In ​ the values shown in each divided region 61 divided in the processed surface image 60 are the first processed surface quality data of each divided region 61

[0144] Furthermore, the processed surface quality data calculation unit 22 overlaps the processed surface image 60 with the added value of the first processed surface quality data and the image of the uneven processed surface pattern 50 with each divided region 61 as shown ​ The processed surface quality data calculation unit 22 sets the ratio of the portion where the first processed surface quality data is calculated and which overlaps the region of the uneven processed surface pattern 50 as the degree of consistency with respect to the uneven processed surface pattern 50

[0145] That is, the degree of consistency is the degree of consistency between the uneven processed surface pattern 50 and the first processed surface quality data. Here, the uneven processed surface pattern 50 is a pattern predetermined as a pattern of a region where the processing roughness of the processed surface 17a after shape engraving electric discharge machining is worse than the desired reference. The first processed surface quality data is the processed surface quality data obtained by evaluating the processed surface quality of the processed surface 17a based on the distribution of pits, which are discharge traces larger than a predetermined size of discharge traces in the processed surface 17a obtained from the image of the processed surface 17a after shape engraving electric discharge machining, in the divided region 61 obtained by dividing the processed surface 17a obtained from the image of the processed surface 17a after shape engraving electric discharge machining. The desired reference can be appropriately changed according to the uniformity required for the processed surface 17a. In Embodiment 2, the degree of consistency is set as the processed surface quality data, that is, the second processed surface quality data. That is, the processed surface quality data calculation unit 22 also calculates and quantifies the processed surface quality of the processed surface 17a of the workpiece 17 after electric discharge machining based on the image of the processed surface 17a in Embodiment 2

[0146] In ​ an example of the degree of consistency between the uneven processed surface pattern 50 and the first processed surface quality data is shown ​ In ​The total amount of first surface quality data that matches the designated area of the surface unevenness pattern 50, namely the hatched area 53, is 12, and the total amount of first surface quality data for the entire surface 17a is 14. In this case, the degree of agreement is 85.76% = 12 ÷ 14. In other words, the degree of agreement can be expressed as the ratio of the total amount of first surface quality data calculated for the segmented area 61 that overlaps with the area of the surface unevenness pattern 50 to the total amount of first surface quality data calculated for the segmented area 61.

[0147] The machined surface quality data calculation unit 22 calculates the degree of consistency for all of the plurality of machined surface unevenness patterns 50 stored in the machined surface quality data calculation unit 22. Alternatively, if the operator can predict in advance the pattern in which the machined surface 17a will become uneven based on the processing object information and processing conditions, and the operator deems that the unevenness pattern of the machined surface 17a is close to the predicted unevenness pattern, the operator may input information specifying one or more machined surface unevenness patterns 50 into the machined surface quality data calculation unit 22. In this case, the machined surface quality data calculation unit 22 calculates the degree of consistency for the specified machined surface unevenness patterns 50.

[0148] Next, use ​ Next, a description will be given of a processing procedure for the learning process performed by the learning device 23 of the electrical discharge machine 1 in the second embodiment. ​ This is a flowchart showing the processing procedure of the learning process performed by the learning device in the second embodiment.

[0149] The data acquisition unit 231 acquires the processing object information, processing conditions, and second processed surface quality data as second learning data (step S310 ).

[0150] The model generation unit 232 calculates the reward for the applied processing conditions based on the processing object information, the processing conditions, and the second processing surface quality data (step S320). Specifically, the reward calculation unit 233 obtains the second processing surface quality data and, based on a predetermined reward benchmark, i.e., a second threshold value 33, determines whether to increase or decrease the reward for the applied processing conditions. If the second processing surface quality data is less than the second threshold value 33, the reward calculation unit 233 determines to increase the reward for the applied processing conditions. If the second processing surface quality data is greater than or equal to the second threshold value 33, the reward calculation unit 233 determines to decrease the reward for the applied processing conditions.

[0151] When the second processed surface quality data is less than the second threshold value 33, the reward calculation unit 233 increases the reward (step S330). On the other hand, when the second processed surface quality data is greater than or equal to the second threshold value 33, the reward calculation unit 233 decreases the reward (step S340).

[0152] Based on the reward calculated by the reward calculation unit 233, the function update unit 234 updates the action value function Q(s t , a t ) represented by formula (2) stored in the trained model storage unit 24 (step S350).

[0153] The learning device 23 repeatedly executes the above steps from step S310 to step S350, and stores the generated action value function Q(s t , a t ) in the trained model storage unit 24 as the trained model 30.

[0154] In addition, similar to the case of the first embodiment, the inference device 25 uses the trained model 30 generated as described above and the workpiece information, and infers the machining conditions for improving the machining surface quality of the machining surface 17a based on the workpiece information.

[0155] As described above, in the second embodiment, the non-uniformity of the distribution of machining chips between the machining surface and the electrode E in the in-plane direction of the machining surface during electrical discharge machining, that is, the agglomeration of machining chips, is defined as the machining surface non-uniformity pattern 50. Moreover, the learning device 23 generates an action value function Q(s t , a t ) that reduces the above-mentioned degree of consistency, that is, the trained model 30, thereby enabling learning of the relationship between the workpiece information for inferring machining conditions and the machining conditions, and the machining conditions can machine the machining surface 17a in such a way that the first processed surface quality data of the divided region 61 obtained by dividing the machining surface 17a does not fit the machining surface non-uniformity pattern 50. That is, the learning device 23 can learn the relationship between the workpiece information for inferring machining conditions and the machining conditions, and the machining conditions can form a machining surface 17a in which the regions with different surface roughness do not fit the machining surface non-uniformity pattern 50.

[0156] By calculating the second processed surface quality data, that is, the degree of consistency, it is possible to learn the appropriate conditions for the discharge and discharge actions of the precipitate and machining chips that are highly correlated with the non-uniformity result of the machining surface 17a in electrical discharge machining.

[0157] Next, use ​ to describe the processing sequence of other processes for the learning device 23 of the electrical discharge machine 1 in the second embodiment.​ This is a flowchart showing the processing sequence of other learning processes performed by the learning device in Embodiment 2.

[0158] The data acquisition unit 231 acquires the first learning data and the second learning data. That is, the data acquisition unit 231 acquires the workpiece information, processing conditions, first processed surface quality data, and second processed surface quality data as learning data (step S410).

[0159] Then, in step S420, the same processing as in step S120 is performed, in step S430, the same processing as in step S130 is performed, and in step S440, the same processing as in step S140 is performed.

[0160] In addition, in step S450, the same processing as in step S320 is performed, in step S460, the same processing as in step S330 is performed, and in step S470, the same processing as in step S340 is performed.

[0161] The function update unit 234 updates the action value function Q(s t , a t ) represented by Equation (2) stored in the trained model storage unit 24 based on the reward calculated by the reward calculation unit 233 (step S480).

[0162] The learning device 23 repeatedly executes the steps from step S410 to step S480 above, and stores the generated action value function Q(s t , a t ) as the trained model 30 in the trained model storage unit 24.

[0163] In addition, the inference device 25, in the same manner as in Embodiment 1, uses the trained model 30 generated as described above and the workpiece information, and infers the processing conditions for improving the processed surface quality of the processed surface 17a based on the workpiece information.

[0164] By performing the above processing, the learning device 23 generates the action value function Q(s t , a t ) that reduces the first processed surface quality data, that is, the trained model 30. Thus, it is possible to learn the relationship between the workpiece information for inferring the processing conditions and the processing conditions, and the processing conditions can improve the processed surface quality of the processed surface 17a of the workpiece 17, that is, can improve the uniformity of the processed surface 17a. In addition, the learning device 23 can learn the relationship between the workpiece information for inferring the processing conditions and the processing conditions, and the processing conditions can form a processed surface 17a in which the region with a surface roughness difference is not suitable for the processing of the processed surface non-uniform pattern 50.

[0165] In the case where, through the above learning, state A is targeted as the state of the distribution of machining chips, machining conditions for suppressing non-uniformity of the machining surface 17a can also be appropriately generated in the case where state D is targeted.

[0166] Therefore, according to the second embodiment, similarly to the case of the first embodiment described above, the machining surface quality of the machining surface 17a of the workpiece 17 can be improved, and learning of appropriate machining conditions for electrical discharge machining can be performed.

[0167] Here, the hardware configuration of the machining condition setting unit 16 will be described. ​ It is a diagram showing an example of the hardware configuration for implementing the machining condition setting unit according to the first and second embodiments.

[0168] The machining condition setting unit 16 can be implemented by a processor 100, a memory 200, an input device 300, and an output device 400. Examples of the processor 100 are a CPU (also referred to as a Central Processing Unit, a central processing device, a processing device, an arithmetic device, a microprocessor, a microcomputer, a DSP (Digital Signal Processor)), or a system LSI (Large Scale Integration). Examples of the memory 200 are a RAM (Random Access Memory) and a ROM (ReadOnly Memory).

[0169] The machining condition setting unit 16 is implemented by the processor 100 reading out and executing a computer-executable control program stored in the memory 200 for executing the operations of the control device 2. The control program, which is the program for executing the operations of the machining condition setting unit 16, can be said to cause the computer to execute the sequence or method executed by the machining condition setting unit 16.

[0170] The memory 200 is used as a temporary memory when the processor 100 performs various processes. The memory 200 stores, for example, a machining condition setting program, machining conditions, etc. executed by the machining condition setting unit 16. The input device 300 receives machining object information, machining conditions, and image information of the machining surface when the workpiece 17 is subjected to electrical discharge machining and sends them to the processor 100. The output device 400 outputs machining conditions, etc. to an external device such as the display unit 13.

[0171] The machining condition setting program can be provided as a computer program product by being stored in a computer-readable storage medium in the form of an installable or executable file. In addition, the machining result evaluation program can also be provided to the machining condition setting unit 16 via a network such as the Internet.

[0172] In addition, regarding the functions of the machining condition setting unit 16, a part of them can be implemented by dedicated hardware such as a dedicated circuit, and a part of them can be implemented by software or firmware. In addition, the control device 2 can be set to have the same hardware structure as the machining condition setting unit 16.

[0173] The program for executing the operation of the learning device 23, that is, the learning program, has a module structure including a data acquisition unit 231 and a model generation unit 232. In addition, the program for executing the operation of the inference device 25, that is, the inference program, has a module structure including a data acquisition unit 251 and an inference unit 252.

[0174] The structures shown in the above embodiments represent an example. It is also possible to combine with other known technologies, combine the embodiments with each other, and omit or change a part of the structure without departing from the gist.

[0175] Explanation of reference numerals

[0176] 1 Electric discharge machine, 2 Control device, 12 Driving unit, 13 Display unit, 14 Machine control unit, 15 Power supply control unit, 16 Machining condition setting unit, 17 Workpiece, 17a Machining surface, 18 Platform, 19 Base, 21 Input unit, 22 Machining surface quality data calculation unit, 23 Learning device, 24 Trained model storage unit, 25 Inference device, 26 Machining condition storage unit, 27 Machining condition output unit, 28 Machining condition correction unit, 30 Trained model, 31 First threshold, 32 Machining condition, 33 Second threshold, 41 Discharge pulse, 42 Machining chip, 43 Machining fluid, 45 Machining completed area, 46 Solid line, 47 Dotted line, 50 Machining surface non-uniform pattern, 51, 53 Hatched areas, 52, 54 Non-hatched areas, 60 Machining surface image, 61 Segmented area, 100 Processor, 200 Memory, 231, 251 Data acquisition unit, 232 Model generation unit, 233 Reward calculation unit, 234 Function update unit, 252 Inference unit, 300 Input device, 400 Output device.

Claims

1. A learning device that learns machining conditions used in profile electric discharge machining of a workpiece in an electric discharge machine, The learning device is characterized by comprising: a data acquisition unit that acquires learning data, the learning data including workpiece information, machining conditions, and values of evaluation indexes, the workpiece information including information related to a target workpiece and information related to an electrode used in the electric discharge machine, the machining conditions being the machining conditions used when profile electric discharge machining the workpiece into the target workpiece, and the values of the evaluation indexes indicating the machining surface quality of the machining surface of the workpiece after profile electric discharge machining using the machining conditions; and a model generation unit that uses the learning data to generate a trained model for inferring machining conditions according to the workpiece information, the machining conditions improving the machining surface quality of the machining surface; The evaluation index is machining surface quality data obtained by evaluating the machining surface quality of the machining surface based on the distribution of discharge marks on the machining surface larger than a pre-determined size in the image of the machining surface after profile electric discharge machining.

2. A learning device that learns machining conditions used in profile electric discharge machining of a workpiece in an electric discharge machine, The learning device is characterized by comprising: a data acquisition unit that acquires learning data, the learning data including workpiece information, machining conditions, and values of evaluation indexes, the workpiece information including information related to a target workpiece and information related to an electrode used in the electric discharge machine, the machining conditions being the machining conditions used when profile electric discharge machining the workpiece into the target workpiece, and the values of the evaluation indexes indicating the machining surface quality of the machining surface of the workpiece after profile electric discharge machining using the machining conditions; and a model generation unit that uses the learning data to generate a trained model for inferring machining conditions according to the workpiece information, the machining conditions improving the machining surface quality of the machining surface; The evaluation index is the degree of consistency between a pattern determined in advance, i.e., a machining surface non-uniformity pattern, which is a pattern of areas where the machining roughness of the machining surface after profile electric discharge machining is worse than a desired reference, and machining surface quality data obtained by evaluating the machining surface quality of the machining surface based on the distribution of discharge marks on the machining surface larger than a pre-determined size in the image of the machining surface after profile electric discharge machining in the divided regions obtained by dividing the machining surface in the image of the machining surface after profile electric discharge machining.

3. The learning device according to claim 1 or 2, characterized in that the evaluation index is the uniformity of discharge marks formed on the machining surface of the workpiece after profile electric discharge machining.

4. The learning device according to claim 1 or 2, characterized in that It has a machined surface quality data calculation unit that calculates the value of the evaluation index using an image of the machined surface of the workpiece after the shaped electrical discharge machining.

5. The learning device according to claim 3, characterized in that: It has a machined surface quality data calculation unit that calculates the value of the evaluation index using an image of the machined surface of the workpiece after the shaped electrical discharge machining.

6. The learning device according to claim 1 or 2, characterized in that: The model generation unit has: A reward calculation unit that calculates the reward for the machining conditions based on the machining conditions and the value of the evaluation index; and A function update unit that updates the trained model based on the reward.

7. The learning device according to claim 3, characterized in that: The model generation unit has: A reward calculation unit that calculates the reward for the machining conditions based on the machining conditions and the value of the evaluation index; and A function update unit that updates the trained model based on the reward.

8. The learning device according to claim 4, characterized in that: The model generation unit has: A reward calculation unit that calculates the reward for the machining conditions based on the machining conditions and the value of the evaluation index; and A function update unit that updates the trained model based on the reward.

9. The learning device according to claim 5, characterized in that: The model generation unit has: A reward calculation unit that calculates the reward for the machining conditions based on the machining conditions and the value of the evaluation index; and A function update unit that updates the trained model based on the reward.

10. An electrical discharge machining machine that obtains the learning result of the learning device according to any one of claims 1 to 9, i.e., the trained model, and performs shaped electrical discharge machining on a workpiece, characterized in that it has: An inference unit that infers machining conditions based on the trained model according to machining object information; and A control unit that controls the shaped electrical discharge machining operation on the workpiece based on the machining conditions inferred by the inference unit.

11. A learning method that learns the machining conditions used during the shaped electrical discharge machining of a workpiece in an electrical discharge machining machine, characterized by including the following steps: Obtaining learning data, which includes machining object information, machining conditions, and the value of an evaluation index. The machining object information includes information related to the target workpiece and information related to the electrode used in the electrical discharge machining machine. The machining conditions are the machining conditions used when shaping the workpiece into the target workpiece, and the value of the evaluation index represents the machining surface quality of the machined surface of the workpiece after the shaped electrical discharge machining using the machining conditions; and Using the learning data, a trained model for inferring machining conditions is generated based on the workpiece information, the machining conditions being such that the surface quality of the machined surface is improved. The evaluation index is surface quality data obtained by evaluating the surface quality of the machined surface based on the distribution of discharge marks in the machined surface that are larger than a predetermined size, obtained from an image of the machined surface after the shaped electrode discharge machining.

12. A learning method for learning machining conditions used in shaped electrode discharge machining of a workpiece in an electric discharge machine. The learning method is characterized by including the following steps: Obtaining learning data including workpiece information, machining conditions, and values of an evaluation index, the workpiece information including information related to a target workpiece and information related to an electrode used in the electric discharge machine, the machining conditions being the machining conditions used when shaped electrode discharge machining the workpiece into the target workpiece, and the value of the evaluation index indicating the surface quality of the machined surface of the workpiece after shaped electrode discharge machining using the machining conditions; and Using the learning data, a trained model for inferring machining conditions is generated based on the workpiece information, the machining conditions being such that the surface quality of the machined surface is improved. The evaluation index is the degree of consistency between a pre-determined pattern, i.e., a surface non-uniformity pattern, which is a pattern of areas where the surface roughness of the machined surface after the shaped electrode discharge machining is worse than the desired reference, and surface quality data obtained by evaluating the surface quality of the machined surface based on the distribution of discharge marks in the machined surface that are larger than a predetermined size, in the segmentation regions obtained by segmenting the machined surface from an image of the machined surface after the shaped electrode discharge machining.

13. A computer-readable storage medium storing a program for causing a learning device to learn machining conditions used in shaped electrode discharge machining of a workpiece in an electric discharge machine. The storage medium is characterized in that the program includes the following steps: Obtaining learning data including workpiece information, machining conditions, and values of an evaluation index, the workpiece information including information related to a target workpiece and information related to an electrode used in the electric discharge machine, the machining conditions being the machining conditions used when shaped electrode discharge machining the workpiece into the target workpiece, and the value of the evaluation index indicating the surface quality of the machined surface of the workpiece after shaped electrode discharge machining using the machining conditions; and Using the learning data, a trained model for inferring machining conditions is generated based on the workpiece information, the machining conditions being such that the surface quality of the machined surface is improved. The evaluation index is surface quality data obtained by evaluating the surface quality of the machined surface based on the distribution of discharge marks larger than a predetermined size in the machined surface obtained from an image of the machined surface after the shaped engraving electrical discharge machining.

14. A computer-readable storage medium storing a program for causing a learning device to learn machining conditions used in shaped engraving electrical discharge machining of a workpiece in an electrical discharge machine, The storage medium is characterized in that the program includes the following steps; acquiring learning data, the learning data including workpiece information, machining conditions, and values of an evaluation index, the workpiece information including information related to a target workpiece and information related to an electrode used in the electrical discharge machine, the machining conditions being machining conditions used when shaping and engraving the workpiece into the target workpiece, and the value of the evaluation index indicating the surface quality of the machined surface of the workpiece after the shaped engraving electrical discharge machining using the machining conditions; and using the learning data to generate a trained model for inferring machining conditions according to the workpiece information, the machining conditions improving the surface quality of the machined surface, the evaluation index is the consistency between a pattern predetermined as a pattern of a region where the machining roughness of the machined surface after the shaped engraving electrical discharge machining is worse than a desired reference, i.e., a surface non-uniformity pattern, and surface quality data obtained by evaluating the surface quality of the machined surface based on the distribution of discharge marks larger than a predetermined size in the machined surface obtained from an image of the machined surface after the shaped engraving electrical discharge machining in a segmentation region obtained by segmenting the machined surface obtained from the image of the machined surface after the shaped engraving electrical discharge machining.

Images