Machining diagnosis device, learning device, inference device, machining diagnosis method, machining diagnosis program, and machine tool system

Abstract

This machining diagnosis device (1) comprises: a data acquisition unit (11) that acquires a physical quantity relating to an operation state of a machine tool (90) for machining a workpiece and a machining condition at the time the physical quantity is obtained; a feature value calculation unit (12) that on the basis of the physical quantity and the machining condition, calculates, for each machining condition, a statistical feature value of a physical quantity obtained during machining under the same machining condition; and an inference unit (16) that performs machining diagnosis on the basis of the statistical feature value for each machining condition.

Description

Machining diagnosis device, learning device, inference device, machining diagnosis method, machining diagnosis program, and machine tool system

[0001] The present disclosure relates to a machining diagnosis device, a learning device, an inference device, a machining diagnosis method, a machining diagnosis program, and a machine tool system that estimate the results of machining performed by a numerically controlled (NC) machine tool.

[0002] Numerically controlled machine tools cannot produce good machining results if the tools used wear out during machining. Therefore, technologies have been proposed to determine whether the specified machining results can be achieved and whether tool replacement is necessary.

[0003] For example, Patent Document 1 describes a technology that collects the drive current, vibration, temperature, etc. of a spindle motor as status data that indicates the operating status of a numerically controlled machine tool, and predicts the machining quality and tool replacement time based on the collected status data. In the technology described in Patent Document 1, when machining is performed to repeatedly create machined products of the same shape, a regression equation is generated as a prediction model used to predict the machining quality, assuming that status data can be obtained repeatedly under the same machining conditions, for example.

[0004] Japanese Patent Application Laid-Open No. 2020-123409

[0005] However, the technology described in Patent Document 1 does not take into consideration machining conditions such as the rotation speed of the spindle, the feed rate of the tool during machining, the cutting depth of the workpiece, etc. Therefore, in machining for high-mix low-volume production where the same machining is not repeated, such as machining for making molds or machining of custom-made products, it is difficult to repeatedly collect status data under the same machining conditions, which poses a problem in that it is not possible to predict machining quality or determine when to replace the tool.

[0006] In machining to obtain a single machined product, the workpiece is typically machined while changing machining conditions, such as the cutting depth and feed rate of the tool. When machining is repeated to obtain a machined product of the same shape, the machining conditions change at the same time for each repeated machining, and the status data during machining changes at the same time each time. Therefore, even if the machining conditions are unknown, it is possible to predict the machining quality and determine when to replace the tool from the status data obtained from each repeated machining. On the other hand, in machining for high-mix, low-volume production, where the timing of changes in machining conditions varies for each machining, it is difficult to predict the machining quality or determine when to replace the tool based solely on the status data.

[0007] The present disclosure has been made in consideration of the above, and aims to provide a machining diagnosis device that is capable of predicting machining quality and determining when to replace tools, even when machining for high-mix, low-volume production.

[0008] In order to solve the above-mentioned problems and achieve the object, the machining diagnosis device according to the present disclosure is characterized by including a data acquisition unit that acquires physical quantities related to the operating state of a machining machine that processes a workpiece and the machining conditions under which the physical quantities were obtained, a feature quantity calculation unit that calculates, for each machining condition, statistical feature quantities of the physical quantities obtained during machining under the same machining conditions based on the physical quantities and the machining conditions, and an estimation unit that performs machining diagnosis based on the statistical feature quantities for each machining condition.

[0009] The machining diagnosis device according to the present disclosure has the effect of being able to predict machining quality and determine the time for tool replacement even when machining is performed for high-mix, low-volume production.

[0010] FIG. 1 is a diagram showing an example of a machine tool system realized by applying a machining diagnosis device according to an embodiment; FIG. 2 is a diagram showing an example of machining performed by a numerically controlled machine tool of a machine tool system according to an embodiment; FIG. 3 is a diagram showing an example of machining parameters necessary for creating a machining program according to an embodiment; FIG. 4 is a diagram showing an example of a configuration of a database referenced by a machining program generation unit of a numerical control device when generating a machining program;

[0011] Hereinafter, a machining diagnosis device, a learning device, an inference device, a machining diagnosis method, a machining diagnosis program, and a machine tool system according to embodiments of the present disclosure will be described in detail with reference to the drawings.

[0012] 1 is a diagram showing an example of a machine tool system 100 realized by applying a machining diagnosis device 1 according to an embodiment. In FIG. 1, the configuration of a numerically controlled machine tool 7 that is the target of diagnosis by the machining diagnosis device 1 is also shown.

[0013] First, we will explain the configuration of the numerically controlled machine tool 7. The numerically controlled machine tool 7 includes a numerical control device 70 that processes numerical control data (NC data) and outputs control signals, and a processing machine 90 that processes a workpiece WA in accordance with the control signals from the numerical control device 70.

[0014] The processing machine 90 includes a spindle 91, a tool 92 attached to the spindle 91, a spindle motor 93 that rotates and drives the spindle 91, a table 94 to which a workpiece WA is fixed, a movement mechanism 95 that moves the table 94, two servo motors 96x and 96y that position the movement mechanism 95 in the x-axis and y-axis directions, and a servo motor 96z that positions the spindle motor 93 in the z-axis direction. A drive current is supplied to the spindle motor 93 from the spindle amplifier 85, and a drive current is supplied to the servo motors 96x, 96y, and 96z from the servo amplifier 86. The servo motors 96x, 96y, and 96z can independently control the position of the tool 92 relative to the workpiece WA in the x-axis, y-axis, and z-axis directions.

[0015] A current sensor S1, a vibration sensor S2, and a temperature sensor S3 are arranged on the spindle motor 93. The processing machine 90 is also arranged with devices 98 such as contactors, solenoids, lamps, etc., as well as limit switches, sensors, and other switches 97 that detect the state of each part in the processing machine 90 and output detection signals.

[0016] The numerical control device 70 includes a processor 71 that processes information, a memory unit 72 that stores data, an external connection interface (hereinafter referred to as I / F) 73 connected to the machining diagnosis device 1, an operation panel I / F 74 connected to an NC operation panel 82, a spindle control I / F 75 connected to a spindle amplifier 85, and a servo control I / F 76 connected to a servo amplifier 86.

[0017] The processor 71 supplies a drive signal to the spindle amplifier 85 via the spindle control I / F 75 to drive the spindle motor 93, and supplies a drive signal to the servo amplifier 86 via the servo control I / F 76 to operate the servo motors 96x, 96y, and 96z, in accordance with the machining program and numerical information stored in the memory unit 72. This rotates the spindle 91, positions the tool 92, and cuts the workpiece WA. The machining program stored in the memory unit 72 is also called a numerical control program.

[0018] Furthermore, the digital input unit 77 inputs data from the current sensor S1, the vibration sensor S2, the temperature sensor S3, the limit switch, the sensors, and other switches 97, and supplies the data to the processor 71. The processor 71 stores the data in the storage unit 72. Furthermore, the processor 71 transmits the data from the current sensor S1, the vibration sensor S2, and the temperature sensor S3 to the machining diagnosis device 1 via the external connection I / F 73.

[0019] The external connection I / F 73 performs data communication with an external device. In particular, in this embodiment, under the control of the processor 71, the external connection I / F 73 adds time data and type data to the data output by the current sensor S1, the vibration sensor S2, and the temperature sensor S3, and transmits the data to the machining diagnosis device 1.

[0020] The operation panel I / F 74 is connected to an NC operation panel 82 which is equipped with a display unit, a key operation unit, etc. which are not shown in the figure, and transmits display data to the NC operation panel 82, receives key operation signals from the NC operation panel 82 and notifies the processor 71.

[0021] The numerical control device 70 also includes a digital input unit 77 that receives digital signals from the machining machine 90, a digital output unit 78 that transmits digital signals to the machining machine 90, a machining program generation unit 79 that generates a machining program to be stored in the memory unit 72, and an internal timer IT built into the processor 71.

[0022] Here, we will explain the machining program generation unit 79. The machining program generation unit 79 automatically generates a machining program based on various information indicating the machining details, such as a target shape which is the final shape of the workpiece WA obtained when machining of the workpiece WA is performed normally, a pre-machining shape which is the shape of the workpiece WA before machining, a workpiece material which is the material of the workpiece WA, a tool used which is a tool used to machine the workpiece WA, a machining pattern, a peripheral speed, a feed rate, a radial depth of cut, and an axial depth of cut.

[0023] Of the various types of information required to generate a machining program, the machining program generation unit 79 acquires the target shape, before-machining shape, workpiece material, tools to be used, and machining pattern from outside the numerically controlled machine tool 7 (for example, from an operator, etc.). The machining program generation unit 79 acquires the target shape, before-machining shape, workpiece material, tools to be used, and machining pattern via, for example, the NC operation panel 82 and the operation panel I / F 74.

[0024] Of the information acquired by the machining program generating unit 79 from outside the numerically controlled machine tool 7, the tools to be used and machining patterns are set individually for each of a plurality of machining sections that form the path along which the tool moves from the start of machining to obtain the target shape until the end of machining (hereinafter referred to as the tool path). The designation of the machining sections and the setting of the tools to be used and machining patterns in each machining section are performed, for example, by an operator.

[0025] Among the various pieces of information required to generate a machining program, the peripheral speed, feed rate, radial depth of cut, and axial depth of cut are settable according to the tool used and the machining pattern, and if inappropriate values ​​are set, normal machining will not be possible. For this reason, the peripheral speed, feed rate, radial depth of cut, and axial depth of cut are determined in advance for each combination of multiple types of tools usable for machining and multiple specifiable machining patterns, and are stored in a database. The database of peripheral speed, feed rate, radial depth of cut, and axial depth of cut is constructed, for example, in the storage unit 72. The peripheral speed, feed rate, radial depth of cut, and axial depth of cut are determined by the manufacturer of each tool selectable as the tool used, the manufacturer of the numerically controlled machine tool 7, or the like.

[0026] A specific example of the machining program generation unit 79 generating a machining program will be described. For example, when generating a machining program for turning a workpiece 203 using a tool 201 and a cutting edge 202 as shown in FIG. 2, the machining program generation unit 79 generates the machining program based on the machining parameters illustrated in FIG. 3. FIG. 2 is a diagram showing an example of machining performed by the numerically controlled machine tool 7 of the machine tool system 100 according to the embodiment, and FIG. 3 is a diagram showing an example of machining parameters required to generate the machining program. Of the machining parameters shown in FIG. 3, the machining program generation unit 79 acquires the "unit," "machining part," "finish allowance-Z," "tool," "start point-X," "start point-Z," "end point-X," and "end point-Z" from the outside via the NC operation panel 82 and the operation panel I / F 74. Furthermore, for the "radial depth of cut," "circumferential speed," and "feed rate" among the machining parameters shown in Fig. 3, the machining program generation unit 79 searches a database in which the "radial depth of cut," "circumferential speed," and "feed rate" are registered, using the "tool," "unit," and "machining unit" acquired from outside as keys. An example of the configuration of the database in which the "radial depth of cut," "circumferential speed," and "feed rate" are registered is shown in Fig. 4. Fig. 4 is a diagram showing an example of the configuration of a database referenced by the machining program generation unit 79 of the numerical control device 70 when generating a machining program. The database shown in Fig. 4 registers "workpiece material," "tool number," "machining pattern," "circumferential speed," "feed rate," "radial depth of cut," "axial depth of cut," etc.

[0027] In the turning process shown in FIG. 2 , the blade 202 of the tool 201 is brought into contact with the end face of the rotating workpiece 203, and the tool 201 is then moved in the X-axis direction from this state to cut (cut) the workpiece 203 in the Z-axis direction with a depth of cut according to the "radial depth of cut." If the depth of cut required to achieve the target shape is greater than the depth of cut per cut, the Z-axis positions of the tool 201 and the blade 202 are changed and similar cutting is repeated. For example, if the machining parameters for performing the turning process shown in FIG. 2 are as shown in FIG. 3 , the "start point - Z" is "4," the "end point - Z" is "0," and the "finishing allowance - Z" is "0.15," so the depth of cut required in the Z-axis direction is 3.85, calculated by subtracting the finishing allowance (0.15) from the Z-axis movement of the tool 201 and the blade 202 (4). The cutting depth in the Z-axis direction per cutting pass is indicated by the machining parameter "radial cutting depth," which is 1.5 in FIG. 3 . Because the cutting depth in the Z-axis direction per cutting pass (= 1.5) is smaller than the required cutting depth in the Z-axis direction (= 3.85), multiple cutting passes are required. Specifically, cutting passes must be repeated three times. The difference (= 0.85) between the total cutting depth in the Z-axis direction after two cutting passes with a cutting depth of 1.5 (= 3) and the required cutting depth in the Z-axis direction (= 3.85) is the cutting depth for the third cutting pass. The machining program generator 79 creates a machining program for such machining based on the machining parameters shown in FIG. 3 . In this embodiment, when the same machining is repeatedly performed on one section, the section cut in one machining pass may be referred to as a "pass." For example, performing three cutting passes on one section to machine it to the target shape is considered to be three passes.

[0028] The machining program generation unit 79 also obtains information (machining parameters) indicating the machining content and creates machining programs for other machining sections necessary to machine the workpiece to the target shape in a similar manner.

[0029] The machining program stored in the storage unit 72 is not limited to the machining program automatically generated by the machining program generation unit 79. The storage unit 72 may store a machining program manually created by an operator or the like, and the numerically controlled machine tool 7 may machine the workpiece WA in accordance with this machining program.

[0030] When manually creating a machining program stored in the memory unit 72, the program creator refers to the cutting depth, peripheral speed, and feed conditions listed in tool manufacturers' catalogs as recommended cutting values, and sets the peripheral speed, feed rate, radial depth of cut, and axial depth of cut, etc., to values ​​deemed appropriate, taking into consideration the material of the workpiece (workpiece material), the tool used, the machining pattern, etc. However, setting the peripheral speed, feed rate, radial depth of cut, and axial depth of cut, etc., to appropriate values ​​requires significant knowledge and experience, making it difficult for inexperienced creators to create programs. Inappropriate settings for the peripheral speed, feed rate, radial depth of cut, and axial depth of cut, etc., increase the likelihood of problems such as failure to achieve the desired machining quality, machine failure, and tool damage. Preventing these problems requires repeated trial machining and program adjustments, which takes time. Meanwhile, the machining program generation unit 79 automatically sets the peripheral speed, feed rate, radial depth of cut, axial depth of cut, etc., using a database in which the peripheral speed, feed rate, radial depth of cut, axial depth of cut, etc. are registered in advance for each combination of workpiece material, tool, and machining pattern. Therefore, the creator can automatically generate a program that performs repeated machining in accordance with the peripheral speed, feed rate, radial depth of cut, axial depth of cut, etc. retrieved from the settings, as described above, until the target shape is reached, without needing any experience or programming skills, and can create a machining program efficiently in a short time.

[0031] Next, the machining diagnosis device 1 will be described. The machining diagnosis device 1 is, for example, a personal computer (PC) or an industrial personal computer (IPC), a programmable logic controller (PLC), or other factory automation (FA) device. A numerical control device 70 may function as part or all of the machining diagnosis device 1. FIG. 5 is a diagram showing a hardware configuration of the machining diagnosis device 1 according to an embodiment. As shown in FIG. 5 , the machining diagnosis device 1 includes, as its hardware configuration, a processor 51, a main memory unit 52, an auxiliary memory unit 53, an input unit 54, an output unit 55, and a communication unit 56. The main memory unit 52, the auxiliary memory unit 53, the input unit 54, the output unit 55, and the communication unit 56 are all connected to the processor 51 via an internal bus 57.

[0032] The processor 51 includes an integrated circuit such as a CPU (Central Processing Unit), an MPU (Micro Processing Unit), a GPU (Graphics Processing Unit) for calculations, or an FPU (Floating Point Unit). The processor 51 executes a program P1 stored in the auxiliary storage unit 53 to realize various functions of the machining diagnosis device 1 and execute the processes described below.

[0033] The main memory unit 52 includes a RAM (Random Access Memory). The program P1 is loaded into the main memory unit 52 from the auxiliary memory unit 53. The main memory unit 52 is used as a working area for the processor 51.

[0034] The auxiliary storage unit 53 includes a non-volatile memory such as an EEPROM (registered trademark) (Electrically Erasable Programmable Read-Only Memory). In addition to the program P1, the auxiliary storage unit 53 stores various data used in the processing of the processor 51. In accordance with instructions from the processor 51, the auxiliary storage unit 53 supplies the processor 51 with data used by the processor 51 and stores data supplied from the processor 51.

[0035] The input unit 54 includes input devices typified by input keys, buttons, switches, a keyboard, a pointing device, and a DI (Digital Input) contact (photocoupler input). The input unit 54 acquires information input by a user of the machining diagnosis device 1 and other information provided from the outside, and notifies the processor 51 of the acquired information.

[0036] The output unit 55 includes output devices such as a light emitting diode (LED), a liquid crystal display (LCD), a digital output (DO) contact (photocoupler output), and a speaker. The output unit 55 presents various information to the user or outputs it to the outside in accordance with instructions from the processor 51.

[0037] The communication unit 56 has a network interface circuit and an analog signal circuit for communicating with external devices. Examples of external devices include the above-mentioned numerical control device 70 and external sensor 80. The external sensor 80 is a sensor provided outside the numerically controlled machine tool 7, such as a temperature sensor or a vibration sensor. The communication unit 56 receives signals from the outside and outputs information indicated by these signals to the processor 51. The communication unit 56 also transmits signals indicating the information output from the processor 51 to the external device, or outputs analog signals.

[0038] The above-described hardware configurations work together, and the machining diagnosis device 1 uses data collected from the numerically controlled machine tool 7 to learn a model for diagnosing the numerically controlled machine tool 7. Furthermore, once the model learning is complete, the machining diagnosis device 1 performs machining diagnosis on the processing machine 90 provided in the numerically controlled machine tool 7, i.e., diagnoses the results of machining performed by the processing machine 90, based on the learned model and data collected while the numerically controlled machine tool 7 is machining the workpiece WA. Details of the machining diagnosis will be described later.

[0039] 6 is a diagram illustrating an example of a functional block configuration of a machining and diagnosis device 1 according to an embodiment. The machining and diagnosis device 1 includes a data acquisition unit 11, a feature calculation unit 12, a storage unit 13, a learning target identification unit 14, a learning unit 15, an inference unit 16, and a notification unit 17.

[0040] The data acquisition unit 11 and the notification unit 17 are realized by the communication unit 56 shown in FIG. 5. The storage unit 13 is realized by the main storage unit 52 and auxiliary storage unit 53 shown in FIG. 5. The feature calculation unit 12, the learning object identification unit 14, the learning unit 15, and the inference unit 16 are realized by the processor 51, the main storage unit 52, and the auxiliary storage unit 53 shown in FIG. 5. That is, the feature calculation unit 12, the learning object identification unit 14, the learning unit 15, and the inference unit 16 are realized by the processor 51 executing a program P1 for operating as each of these units. It is assumed, but not limited to, that the program P1 for operating as the feature calculation unit 12, the learning object identification unit 14, the learning unit 15, and the inference unit 16 is stored in advance in the auxiliary storage unit 53. The program P1 may be written on a recording medium such as a CD (Compact Disc)-ROM or a DVD (Digital Versatile Disc)-ROM and supplied to a user of the machining diagnosis apparatus 1, and the user may install the program P1 into the auxiliary storage unit 53. Alternatively, the program P1 may be downloaded by the communication unit 56 from a server via a network.

[0041] Each part of the machining diagnosis device 1 shown in FIG. 6 will be described.

[0042] The data acquisition unit 11 acquires from the numerical control device 70 physical quantities related to the machining operation of the machining machine 90, which are collected while the machining machine 90 is machining the workpiece WA, and also acquires analog data from the external sensor 80. As the physical quantities collected during machining, the data acquisition unit 11 acquires, for example, detected values ​​from a current sensor S1, a vibration sensor S2, and a temperature sensor S3 arranged on the spindle motor 93 of the machining machine 90. The data acquisition unit 11 may also acquire the rotation speed, movement amount, torque, force, acceleration, temperature, etc. of each motor equipped in the machining machine 90. The numerical control device 70 periodically collects the above physical quantities at predetermined regular time intervals and controls the machining machine 90 based on the collected physical quantities.

[0043] The data acquisition unit 11 also acquires information referenced by the machining program generation unit 79 of the numerical control device 70 when generating the machining program, specifically, the workpiece material, the tool used, the peripheral speed, the feed rate, the radial depth of cut, the axial depth of cut, etc. The data acquisition unit 11 also acquires the name of the machining program generated by the machining program generation unit 79 of the numerical control device 70, the number of times each tool is used counted by the numerical control device 70, the cumulative tool usage time, information on the steps currently being executed in the machining program, etc. In the following description, the information referenced by the machining program generation unit 79 when generating the machining program (such as the workpiece material, the tool used, the peripheral speed, the feed rate, the radial depth of cut, and the axial depth of cut) acquired by the data acquisition unit 11 may be collectively referred to as machining conditions. The machining conditions include at least the workpiece material, the tool used, the peripheral speed, the feed rate, the radial depth of cut, and the axial depth of cut.

[0044] The feature amount calculation unit 12 calculates feature amounts of the physical quantities acquired by the data acquisition unit 11. Based on a plurality of physical quantities collected over a predetermined period, the feature amount calculation unit 12 calculates statistical feature amounts, such as the average value, maximum value, minimum value, median value, standard deviation, convexity, and skewness over a predetermined period.

[0045] The storage unit 13 stores various data necessary for the machining diagnosis device 1 to operate.

[0046] The learning target identification unit 14 extracts the features to be learned when the learning unit 15 learns a model for diagnosing the numerically controlled machine tool 7 from the features calculated by the feature calculation unit 12, using the machining condition information collected from the memory unit 72 and the machining program generation unit 79 of the numerical control device 70 as search conditions.

[0047] The learning unit 15 performs machine learning using learning data based on the feature amounts extracted by the learning target identification unit 14, and generates a trained model for diagnosing the numerically controlled machine tool 7. The learning unit 15 generates a trained model for each machining condition. In other words, the learning unit 15 learns the relationship between the feature amounts of physical quantities collected when machining under the same machining conditions and the quality information of the workpiece machined by the numerically controlled machine tool 7.

[0048] The inference unit 16 performs machining diagnosis of the numerically controlled machine tool 7 based on the learned model for each machining condition generated by the learning unit 15 and the feature quantities calculated by the feature quantity calculation unit 12. Specifically, the inference unit 16 determines whether the machining result by the machining machine 90 of the numerically controlled machine tool 7 is normal or not, and determines the state of the tool, such as the amount of wear and wear tendency. When determining whether the machining result is normal or not, the inference unit 16 performs, for example, one or both of the following: determining whether the difference between the dimensions of the machining section after machining is completed and the target dimensions; and determining whether the difference between the shape of the workpiece after machining is completed and the target shape is also within a predetermined range. The inference unit 16 also estimates the dimensions, curvature, straightness, and surface roughness of the workpiece after machining by the machining machine 90 is completed. Note that the inference unit 16 may perform all or some of the machining result determination, state determination, and estimation of the workpiece after machining is completed (estimation of the dimensions, curvature, straightness, and surface roughness) in the machining diagnosis.

[0049] The notification unit 17 notifies the numerical control device 70 and the like of the inference result by the inference unit 16, that is, the diagnosis result of the numerically controlled machine tool 7.

[0050] Next, a learning phase in which the machining diagnosis device 1 generates a trained model for diagnosing the numerically controlled machine tool 7, and an utilization phase in which the machining diagnosis device 1 diagnoses the numerically controlled machine tool 7 using the trained model will be described.

[0051] <Learning Phase> FIG. 7 is a flowchart showing an example of an operation of the machining diagnosis device 1 to generate a trained model for diagnosing the numerically controlled machine tool 7.

[0052] The machining diagnosis device 1 first acquires data necessary for generating a model (step S11). Specifically, the data acquisition unit 11 acquires from the numerical control device 70 physical quantities and machining conditions related to machining operations collected while the machining machine 90 is machining the workpiece WA. The data acquisition unit 11 may acquire the above data while the numerically controlled machine tool 7 is machining the workpiece WA with the machining machine 90, or may acquire data collected when the numerically controlled machine tool 7 is machining the workpiece WA with the machining machine 90 and stored in the numerical control device 70 after machining is completed. The machining conditions acquired by the data acquisition unit 11 when the physical quantities were collected are the same as the machining conditions used when the machining program generation unit 79 generated the machining program used when the numerical control device 70 controlled the machining machine 90 to machine the workpiece WA. Note that the workpiece material only needs to be acquired once during machining of the same workpiece WA.

[0053] The machining diagnosis device 1 then calculates feature values ​​of the physical quantities acquired in step S11 (step S12). Specifically, the feature value calculation unit 12 calculates the feature values ​​of the physical quantities. Even when machining is performed under the same machining conditions, the time required to complete the machining varies depending on the machining range (length of the machining section). Therefore, the feature value calculation unit 12 calculates statistical feature values ​​that are not affected by the time required to complete the machining, rather than feature values ​​whose values ​​change depending on the time required to complete the machining, such as the integral value of the physical quantity. Specifically, the feature value calculation unit 12 calculates statistical feature values ​​such as the average value, maximum value, minimum value, median value, standard deviation, convexity, and skewness over a predetermined period. For example, the feature value calculation unit 12 calculates the average value, maximum value, minimum value, median value, and other values ​​of the physical quantities collected between the start of machining under certain machining conditions and the change of the machining conditions. The feature value calculation unit 12 may divide the section in which machining is performed under the same machining conditions into subsections of a fixed width and calculate feature values ​​for each subsection. The feature calculation unit 12 may calculate the feature using the latest acquired data and previously acquired data each time the data acquisition unit 11 acquires the data in step S11, or may calculate the feature using previously acquired data at a predetermined timing. The predetermined timing may be when the number of acquired data reaches a certain number, or when the machining diagnosis device 1 receives a predetermined operation from the user, for example, when it receives an operation instructing the generation of a trained model. The previously acquired data may include a mixture of data obtained in machining using different machining programs, i.e., data obtained when machining with different target shapes is performed on different workpieces. Alternatively, the feature calculation unit 12 may calculate the feature using the data acquired by the data acquisition unit 11 during each predetermined period of time.

[0054] The machining diagnosis device 1 then creates learning data for each machining condition based on the feature values ​​calculated in step S12 (step S13). Specifically, the learning object identification unit 14 groups the feature values ​​calculated by the feature value calculation unit 12 into feature values ​​with the same machining conditions to create learning data for each machining condition. When repeatedly performing the same machining on the same section to machine the workpiece to a target shape (target dimensions), the learning object identification unit 14 identifies an approach section, which is a section from when the machining conditions to be used are set and the tool begins to approach the workpiece until the tool contacts the workpiece and cutting begins at the depth of cut specified by the machining conditions, and excludes the physical feature values ​​collected in the approach section from the learning data. The approach section is provided to prevent the tool from violently colliding with the workpiece. The learning object identification unit 14 also excludes the physical feature values ​​collected in the target dimension adjustment section (described later) from the learning data. 8, the learning target identification unit 14 acquires feature quantities in a repeated pass (step S21) and identifies feature quantities to be used as learning targets (step S22). The feature quantities in a repeated pass are feature quantities of physical quantities collected when the same processing is repeatedly performed on the same section.

[0055] In step S21, the learning object identification unit 14 extracts feature values ​​for the repeat pass from the feature values ​​calculated by the feature value calculation unit 12. In step S22, the learning object identification unit 14 selects the remaining feature values ​​from the repeat pass, excluding the feature values ​​in the approach section and the target dimension adjustment section, as learning targets. FIG. 9 is a diagram illustrating the approach section and the target dimension adjustment section. In FIG. 9, the horizontal axis represents the cutting time (seconds), which is the elapsed time from the start of the cutting operation, and the vertical axis represents the feature value. In the example shown in FIG. 9, after the start of cutting, the tool contacts the workpiece and cutting begins around 40 seconds after the cutting time has exceeded, and the feature value increases with each pass. The magnitude of the feature value is correlated with the cutting depth. When the cutting depth becomes constant (the specified cutting depth), the magnitude of the feature value for each pass becomes approximately constant. However, if the workpiece surface is uneven and has irregularities, the depth of cut may not be the specified depth of cut indicated by the machining conditions immediately after the tool contacts the workpiece and actually starts cutting, and physical quantities and feature quantities may gradually increase. Taking such characteristics into consideration, the learning object identification unit 14 identifies an approach section based on the change in feature quantity per pass. For example, the learning object identification unit 14 determines the approach section as the section from when cutting starts and the feature quantity begins to increase per pass until the increase in feature quantity per pass becomes 10% or less. The change of 10% used to identify the approach section is an example. The learning object identification unit 14 may also determine the approach section as the section until the change becomes 5% or less. Furthermore, if the depth of cut required to achieve the target shape is not an integer multiple of the depth of cut per cutting pass, the depth of cut in the last pass of the repeated passes (hereinafter referred to as the “final pass”) will be smaller than the depth of cut in the previous pass, and the feature quantity will also decrease. Therefore, if the feature amount in the final pass is significantly reduced compared to the feature amount in the previous pass, for example, if the amount of reduction (change) exceeds 10%, the learning object identification unit 14 determines that the final pass is a target dimension adjustment section and excludes it from the learning object. Note that if the cutting depth required to achieve the target shape is an integer multiple of the cutting depth per cutting process, the final pass will not be a target dimension adjustment section.In this way, the learning object identification unit 14 identifies the feature of the physical quantity collected in a section that does not correspond to either the approach section or the target dimension adjustment section among the feature in the repeated pass, and sets it as the feature of the learning object.

[0056] When identifying the feature of the learning target from the feature in the repeated path, the learning target identification unit 14 may determine, as the learning target, the feature in the path in which the feature such as the average value or median of the physical quantity is the largest, and the change in the feature between adjacent paths is within a predetermined range, for example, the feature in the path in which the change in the feature is within ±10%.

[0057] The learning object specifying unit 14 may limit the feature quantities of the learning object to feature quantities of physical quantities collected when the same processing is repeatedly performed on the same section to process it to the target dimension. In this case, the learning unit 15 can learn the trend of feature quantities when the same processing is repeated in machine learning for generating a trained model.

[0058] Returning to the description of FIG. 7 , the machining diagnosis device 1 then generates a trained model for each machining condition (step S14). Specifically, the learning unit 15 performs machine learning using the training data for each machining condition created by the learning target identification unit 14 to generate multiple trained models corresponding to each machining condition. The learning involves generating a model from the feature values ​​used as training data for each machining condition, calculating principal component distance values ​​and neural network anomaly levels from the training data used for model generation or other training target data, and extracting peak values ​​in the training target data as normal / abnormal judgment thresholds. Alternatively, a quality prediction model may be generated by machine learning using the training data for each machining condition, with the quality value as the objective variable. That is, the learning unit 15 generates a trained model for inferring machining results based on feature values ​​of physical quantities collected while the machining machine 90 of the numerically controlled machine tool 7 is machining the workpiece WA. In machining after the completion of learning, if the trained model's principal component distance value or the abnormality degree of the neural network based on the feature values ​​calculated after machining does not exceed a specified normal / abnormal judgment threshold, machining is deemed normal, or if the trained model predicts quality values, the error between the predicted values ​​of dimensions (hereinafter sometimes referred to as actual dimensions), curvature, straightness, surface roughness, etc. and the dimensions of the target shape (hereinafter sometimes referred to as target dimensions) is within a specified range. In machine learning, the learning unit 15 uses machine learning methods such as principal component analysis, multiple regression analysis, support vector machines, decision trees, gradient boosting decision trees, and neural networks.

[0059] The learning unit 15 may use the error between the actual dimension and the target dimension as the machining result and learn the relationship between the error and the feature values ​​used as the learning data. In this case, the learning unit 15 generates a trained model for inferring machining quality based on the feature values ​​of physical quantities collected while the machining machine 90 of the numerically controlled machine tool 7 is machining the workpiece WA, specifically, a trained model for inferring the error between the target dimension and the actual dimension.

[0060] The learning unit 15 may learn the relationship between the wear state of the tool used in machining and the feature values ​​used as training data. In this case, the learning unit 15 generates a trained model for inferring the wear state of the tool based on the feature values ​​of physical quantities collected while the machining device 90 of the numerically controlled machine tool 7 is machining the workpiece WA. The wear state may be information on whether the tool blade is worn to the extent that replacement is required, the actual amount of wear from the initial state (new condition) of the tool blade, or a numerical value indicating the degree of wear, for example, where new condition is 0% and condition requiring replacement is 100%. When the machining device 90 repeatedly performs the same machining on one section to cut to the target dimension, the learning unit 15 may learn the relationship between the feature values ​​and the tool wear state in each repeated pass, or may learn the relationship between the trend of the feature values ​​for each repeated pass and the tool wear state. The difference between the feature values ​​before and after repeated passes may be calculated, and the relationship between the trend of the difference and the tool wear state may be learned. When the tool wears, the feature values ​​of physical quantities collected during machining change. In other words, there is a correlation between the tool wear state and feature values ​​or the difference between feature values. Therefore, by learning the trend of feature values ​​for each repeated pass, it is possible to generate a trained model that can infer the tool wear state with high accuracy.

[0061] As described above, in learning a model for diagnosing the numerically controlled machine tool 7, the machining diagnosis device 1 acquires the physical quantities collected during machining and the machining conditions when the physical quantities were collected from the numerical control device 70, creates learning data for each machining condition based on the feature quantities of the physical quantities and the machining conditions, performs learning using the learning data for each machining condition, and generates a learned model for each machining condition. Furthermore, when generating learning data for each machining condition, the machining diagnosis device 1 creates learning data by identifying feature quantities calculated based on the physical quantities collected while actually machining a workpiece under the acquired machining conditions, from the feature quantities of the physical quantities collected during machining.

[0062] The machining diagnosis device 1 may generate a plurality of types of trained models for each machining condition. For example, the machining diagnosis device 1 may create, for each machining condition, a trained model for inferring whether the machining performed by the machining machine 90 is normal or abnormal and a trained model for inferring the wear state of a tool. For example, the machining diagnosis device 1 may generate, for each machining condition, a trained model for inferring the error between a target dimension and an actual dimension and a trained model for inferring the wear state of a tool.

[0063] <Utilization Phase> FIG. 10 is a flowchart showing an example of an operation in which the machining diagnosis device 1 according to the embodiment uses a trained model to diagnose the machining result of the numerically controlled machine tool 7.

[0064] The machining diagnosis device 1 first acquires data required for the diagnosis process of the numerically controlled machine tool 7 (step S31). Specifically, the data acquisition unit 11 periodically acquires physical quantities and machining conditions related to the machining operation collected from the processing machine 90 currently processing the workpiece WA from the numerical control device 70. The data acquired by the data acquisition unit 11 in step S31 is the same type of data as the data collected in step S11 of the learning phase described above.

[0065] The machining diagnosis device 1 then calculates feature quantities of the physical quantities included in the data acquired in step S31 (step S32). Specifically, the feature quantity calculation unit 12 calculates feature quantities similar to the feature quantities calculated in step S12 of the above-described learning phase at a predetermined timing. The machining diagnosis device 1 calculates feature quantities, for example, at a cycle that is an integer multiple of the cycle at which the data acquisition unit 11 acquires data in step S31. When the machining conditions included in the data acquired by the data acquisition unit 11 in step S31 have changed, i.e., when the machining conditions included in the latest data acquired by the data acquisition unit 11 are different from the machining conditions included in the previously acquired data, the machining diagnosis device 1 may calculate feature quantities of the physical quantities collected during machining under the machining conditions before the change. The feature quantity calculation unit 12 outputs the calculated feature quantities and the machining conditions when the physical quantities used to calculate the feature quantities were collected to the inference unit 16.

[0066] The machining diagnosis device 1 then estimates a machining result by the machining machine 90 of the numerically controlled machine tool 7 based on the feature amounts calculated in step S32 and the trained model generated in the above-mentioned learning phase (step S33). Specifically, the inference unit 16, which is an estimation unit, executes machining diagnosis to estimate the machining result based on the feature amounts input from the feature amount calculation unit 12 and the trained model corresponding to the machining conditions input from the feature amount calculation unit 12. The inference unit 16 inputs the feature amounts to the trained model, and acquires the estimated machining result output from the trained model accordingly.

[0067] The machining diagnosis device 1 then outputs the estimation result in step S33 (step S34). Specifically, the notification unit 17 outputs the machining result estimated by the inference unit 16 in step S33 to the numerical control device 70.

[0068] In addition, the machining diagnosis device 1 may estimate the wear state of the tool in step S33, and when it estimates that the amount of wear has reached a level that requires tool replacement, in step S34, notify the numerical control device 70 of information about the tool that needs replacement and instruct the numerical control device 70 to replace the tool.

[0069] As explained in the learning phase, even when machining is repeatedly performed under the same machining conditions, machining is not performed at the cutting depth indicated by the machining conditions in the approach section. That is, the physical quantities collected in the section from the start of machining to the end of the approach section are significantly different from the physical quantities collected in other sections where machining is performed under the same machining conditions. Therefore, it is not possible to perform a correct diagnosis using the trained model in the approach section, and if a diagnosis is performed, the diagnosis result will be abnormal machining. Therefore, after starting repetitive machining, the machining diagnosis device 1 waits until the approach section ends and the diagnosis result becomes normal for the first time, and then starts actual diagnosis. Specifically, the machining diagnosis device 1 treats the diagnosis result from the start of repetitive machining until the diagnosis result becomes normal for the first time as invalid. FIG. 11 is a flowchart showing an example of a diagnostic operation by the machining diagnosis device 1 according to an embodiment.

[0070] 10 to estimate the machining result (step S41), and checks whether the estimation result indicates normal machining (step S42). If the estimation result indicates abnormal machining (step S42: No), the machining diagnosis device 1 determines that the estimation result is invalid, returns to step S41, and estimates the machining result again.

[0071] If the estimation result is normal machining (step S42: Yes), the machining diagnosis device 1 starts diagnosis (step S43) and estimates the machining result (step S44). In step S44, the same processes as steps S31 to S34 shown in Fig. 10 are executed to estimate the machining result and output the estimated machining result.

[0072] After estimating the machining result in step S44, the machining diagnosis device 1 checks whether the estimated result is abnormal machining (step S45). If the estimated result is normal machining (step S45: No), the machining diagnosis device 1 returns to step S44 and again estimates the machining result and outputs the estimated machining result. On the other hand, if the estimated result is abnormal machining (step S45: Yes), the machining diagnosis device 1 checks whether the machining is in the final pass, i.e., whether the estimated result is abnormal machining in the final pass (step S46). If the estimated result is abnormal machining in the final pass (step S46: Yes), this is because machining was performed with a cutting depth different from that in other passes, and the machining diagnosis device 1 simply ends its operation. On the other hand, if the estimated result is abnormal machining in a pass other than the final pass (step S46: No), it is possible that the change in the feature value tends to differ from that in normal machining due to breakage or loss of the tool used, resulting in the estimated result being abnormal machining. Therefore, the machining diagnosis device 1 notifies the numerical control device 70 of the numerically controlled machine tool 7 that the tool is abnormal (step S47), and ends the diagnosis operation.

[0073] In this way, the machining diagnosis device 1 acquires from the numerical control device 70 the physical quantities related to the machining operation collected from the machining machine 90 that is machining the workpiece WA, and the machining conditions under which the physical quantities were collected, and performs machining diagnosis of the numerically controlled machine tool 7, specifically, estimates the machining results by the machining machine 90 equipped in the numerically controlled machine tool 7, using the feature quantities of the acquired physical quantities and the learned model corresponding to the acquired machining conditions.

[0074] In this embodiment, the numerical control device 70 is provided with the machining program generation unit 79, and the machining program generation unit 79 generates the machining program, but the present invention is not limited to this configuration. A configuration in which a CAM (Computer Aided Manufacturing) system creates the machining program based on drawing data of the target shape of the workpiece WA created by a CAD (Computer Aided Design) system may also be used.

[0075] When a CAM system creates a machining program, an operator sets the cutoff amount in the CAM system based on a drawing of the target shape created by the CAD system, referring to past experience with similar machining (cutting conditions used with the same tool in the past), and also registers coordinate information indicating the tool's passing points in the CAM system. After the operator completes the necessary settings and registration, and then issues an instruction to create a machining program, the CAM system automatically generates the machining program based on the registered coordinate information and the set machining conditions (workpiece material, tool used, peripheral speed, feed rate, radial cutting depth, and axial cutting depth). The CAM system stores information on the machining conditions for each line of the machining program (hereinafter referred to as "line-by-line machining condition information"). For this reason, the machining diagnosis device 1 acquires the line-by-line machining condition information from the CAM system in advance, along with the name of the machining program, and stores it in the memory unit 13. When machining is performed by the numerically controlled machine tool 7, the machining diagnosis device 1 acquires from the numerical control device 70 the name of the machining program being executed, the physical quantities collected by the numerical control device 70, and information on the step in execution of the machining program (hereinafter referred to as "current step information"), and identifies the machining conditions under which the acquired physical quantities were obtained by comparing the acquired current step information with the line-by-line machining condition information stored in the storage unit 13. In this way, the machining diagnosis device 1 can create learning data for each machining condition, perform machine learning, and generate a trained model for each machining condition. Furthermore, the machining diagnosis device 1 can diagnose the machining results of the numerically controlled machine tool 7 using the trained model for each machining condition.

[0076] As described above, the machining diagnosis device 1 according to the present embodiment acquires, from a numerically controlled machine tool 7 including a numerical control device 70 and a processing machine 90, physical quantities collected while the processing machine 90 processes a workpiece WA and the processing conditions when the physical quantities were collected, performs machine learning using the acquired physical quantities and processing conditions, and generates a trained model for diagnosing the numerically controlled machine tool 7 for each processing condition. After generating the trained model, the machining diagnosis device 1 acquires the physical quantities collected while the processing machine 90 processes the workpiece WA and the processing conditions when the physical quantities were collected, and performs machining diagnosis of the numerically controlled machine tool 7 based on the trained model corresponding to the acquired processing conditions and the feature quantities of the acquired physical quantities. The machining diagnosis device 1 according to the present embodiment performs machining diagnosis taking into account the processing conditions when the physical quantities were collected (workpiece material, tool used, peripheral speed, feed rate, radial cut-in amount, axial cut-in amount). Therefore, machining diagnosis of the numerically controlled machine tool 7 can be performed even when the processing machine 90 of the numerically controlled machine tool 7 to be diagnosed performs machining for high-mix low-volume production.

[0077] The machining diagnosis device 1 may be configured to perform different diagnoses for each machining section during the machining diagnosis operation of the numerically controlled machine tool 7. For example, the machining diagnosis device 1 may diagnose the wear state of the tool and the machining result when the machining machine 90 repeatedly performs the same machining for one machining section to cut to the target dimension, or may diagnose only the machining result when the machining machine 90 does not repeatedly perform the same machining for one machining section (performs machining only once). Furthermore, the machining diagnosis device 1 may diagnose either the wear state of the tool or the machining result, or both, only when the machining machine 90 repeatedly performs the same machining for one machining section to cut to the target dimension.

[0078] The configurations shown in the above embodiments are merely examples, and may be combined with other known technologies, and parts of the configurations may be omitted or modified without departing from the spirit of the invention.

[0079] For example, in the embodiment, the configuration in which the learning unit 15 and the inference unit 16 are provided inside the machining diagnosis device 1 has been described, but one or both of the learning unit 15 and the inference unit 16 may be provided outside the machining diagnosis device 1. For example, the learning unit 15 may be provided in an external learning device, and the inference unit 16 may be provided in an external inference device, and these learning devices and inference devices may be connected to the machining diagnosis device 1 via a communication network or the like. Furthermore, the machining diagnosis device 1 may be included in the numerically controlled machine tool 7.

[0080] 1 Machining diagnosis device, 7 Numerical control machine tool, 11 Data acquisition unit, 12 Feature calculation unit, 13, 72 Memory unit, 14 Learning target identification unit, 15 Learning unit, 16 Inference unit, 17 Notification unit, 51, 71 Processor, 52 Main memory unit, 53 Auxiliary memory unit, 54 Input unit, 55 Output unit, 56 Communication unit, 70 Numerical control device, 73 External connection I / F, 74 Operation panel I / F, 75 Spindle control I / F, 76 Servo control I / F, 77 Digital input unit, 78 Digital output unit, 79 Machining program generation unit, 80 External sensor, 82 NC operation panel, 85 Spindle amplifier, 86 Servo amplifier, 90 Machining machine, 91 Spindle, 92, 201 Tool, 93 Spindle motor, 94 Table, 95 Moving mechanism, 96x, 96y, 96z Servo motor, 100 machine tool system, 202 blade, 203 workpiece.

Claims

1. A machining diagnosis device comprising: a data acquisition unit that acquires physical quantities related to the operating state of a processing machine that processes a workpiece and the processing conditions under which said physical quantities were obtained; a feature quantity calculation unit that calculates, for each processing condition, statistical feature quantities of physical quantities obtained during processing under the same processing conditions based on said physical quantities and said processing conditions; and an estimation unit that performs machining diagnosis based on the statistical feature quantities for each processing condition.

2. The machining diagnosis device according to claim 1, characterized in that the machining conditions are the material of the workpiece, the tool used in the machining, the peripheral speed, feed rate, radial depth of cut, and axial depth of cut when performing the machining.

3. The machining diagnosis device according to claim 1 or 2, characterized in that the machining machine is provided in a numerically controlled machine tool having a machining program generation unit that automatically generates a machining program based on the input material of the workpiece, the tools used in machining, and a database of machining parameters, the machining is performed using the machining program generated by the machining program generation unit, and the data acquisition unit acquires the machining conditions from the machining program generation unit.

4. The machining diagnosis device according to claim 1 or 2, characterized in that the machining machine performs the machining using a machining program generated by a CAM system based on the target shape of the workpiece, and the data acquisition unit acquires the machining conditions from the CAM system.

5. A machining diagnosis device according to any one of claims 1 to 4, characterized in that the estimation unit estimates whether the machining performed by the machining machine is normal or abnormal based on the statistical feature amount.

6. The machining diagnosis device according to any one of claims 1 to 5, wherein the estimation unit estimates the quality of machining performed by the machining machine based on the statistical feature amount.

7. A machining diagnosis device according to any one of claims 1 to 6, characterized in that the machining diagnosis result by said estimation unit includes the wear state of a tool used by said processing machine in machining when said physical quantity was obtained, and further comprising: a notification unit that instructs tool replacement when said wear state indicates that the tool has reached an amount of wear that requires replacement.

8. A machining diagnosis device according to any one of claims 1 to 7, characterized in that the estimation unit performs the machining diagnosis based on statistical feature quantities of physical quantities obtained when the machining machine repeatedly performs machining under the same machining conditions on the same section.

9. The machining diagnosis device according to claim 8, wherein the estimation unit performs the machining diagnosis using the remaining statistical feature values ​​after excluding the statistical feature values ​​of physical quantities obtained when the cutting depth of the tool into the workpiece does not match the specified cutting depth indicated by the machining conditions.

10. The machining diagnosis device according to claim 9, wherein the estimation unit repeatedly performs the machining diagnosis based on feature amounts of physical quantities obtained during repeated machining of the same section by the machining machine under the same machining conditions, invalidates the result of the machining diagnosis performed before the amount of cut of the tool into the workpiece reaches a specified amount of cut indicated by the machining conditions, and considers the result of the machining diagnosis performed after the amount of cut of the tool into the workpiece reaches the specified amount of cut indicated by the machining conditions to be the valid machining diagnosis result.

11. The machining diagnosis device according to claim 9, characterized in that, after the cutting depth of the tool into the workpiece reaches a specified cutting depth indicated by the machining conditions, if the result of the machining diagnosis based on the feature values ​​of physical quantities obtained in machining of passes other than the final pass among repeated machining under the same machining conditions indicates abnormal machining, the estimation unit estimates that the tool used in machining has broken or been chipped.

12. A machining diagnosis device according to any one of claims 1 to 11, characterized in that the feature amount calculation unit calculates the statistical feature amount based on the physical amount obtained in machining using different machining programs.

13. A machining diagnosis device according to any one of claims 1 to 12, further comprising: a learning unit that performs machine learning using learning data created based on the statistical feature amounts for each of the machining conditions, and generates, for each of the machining conditions, a trained model for performing the machining diagnosis based on the statistical feature amounts; and wherein, when the data acquisition unit acquires the physical quantities and the machining conditions, the estimation unit performs the machining diagnosis using the feature amounts calculated by the feature amount calculation unit based on the acquired physical quantities and the trained model corresponding to the acquired machining conditions.

14. The machining diagnosis device according to claim 13, further comprising a learning target specifying unit that determines the remaining statistical feature values ​​to be used in the machine learning after excluding, from the statistical feature values ​​of physical quantities obtained when the cutting depth of the tool into the workpiece does not match the specified cutting depth indicated by the machining conditions, statistical feature values ​​of physical quantities obtained when the cutting depth of the tool into the workpiece does not match the specified cutting depth indicated by the machining conditions, from the statistical feature values ​​of physical quantities obtained when the machining machine repeatedly performs machining on the same section under the same machining conditions.

15. A learning device comprising: a data acquisition unit that acquires physical quantities related to the operating state of a processing machine that processes a workpiece and the processing conditions under which the physical quantities were obtained; a feature calculation unit that calculates, for each processing condition, statistical feature quantities of the physical quantities obtained during processing under the same processing conditions based on the physical quantities and the processing conditions; and a learning unit that performs machine learning using learning data created based on the statistical feature quantities for each processing condition, and generates, for each processing condition, a trained model for performing processing diagnosis based on the statistical feature quantities.

16. An inference device comprising: a data acquisition unit that acquires physical quantities related to the operating state of a processing machine that processes a workpiece and the processing conditions under which the physical quantities were obtained; a feature calculation unit that calculates, for each processing condition, statistical feature quantities of the physical quantities obtained during processing under the same processing conditions based on the physical quantities and the processing conditions; and an estimation unit that performs processing diagnosis based on a trained model for performing processing diagnosis based on the statistical feature quantities, the trained model being generated by a learning device that performs machine learning using learning data created based on the statistical feature quantities for each processing condition, and the statistical feature quantities for each processing condition calculated by the feature calculation unit.

17. A machining diagnosis method in which a machining diagnosis device estimates the machining results of a workpiece by a machining machine, comprising: a data acquisition step of acquiring physical quantities related to the operating state of the machining machine and the machining conditions under which the physical quantities were obtained; a feature calculation step of calculating, for each machining condition, statistical feature quantities of physical quantities obtained during machining under the same machining conditions based on the physical quantities and the machining conditions; and a diagnosis step of diagnosing the machining based on the statistical feature quantities for each machining condition.

18. A machining diagnosis program that causes a computer to execute the following steps: a data acquisition step of acquiring physical quantities relating to the operating state of a processing machine that processes a workpiece and the processing conditions under which said physical quantities were obtained; a feature calculation step of calculating, for each processing condition, statistical feature quantities of physical quantities obtained during processing under the same processing conditions based on said physical quantities and said processing conditions; and a diagnosis step of diagnosing machining based on said statistical feature quantities for each processing condition.

19. A machine tool system comprising: a processing machine that processes a workpiece; a numerical control device that controls the processing machine in accordance with a processing program; and a processing diagnosis device that diagnoses processing by the processing machine, wherein the processing diagnosis device acquires physical quantities related to the operating state of the processing machine and the processing conditions under which the physical quantities were obtained, calculates statistical feature quantities of the physical quantities obtained during processing under the same processing conditions based on the physical quantities and the processing conditions, for each of the processing conditions, and diagnoses the processing based on the statistical feature quantities for each of the processing conditions.

20. The machine tool system according to claim 19, characterized in that the numerical control device determines the machining conditions and generates the machining program based on the input material of the workpiece, the tools to be used in machining, and a database of machining parameters.

Images