Processing device, processing machine control device, processing method, and control program

Abstract

To provide a processing device and a processing method capable of automatically forming a machine learning model to monitor whether a device is operating normally without bothering a worker, and automatically monitoring the device using the machine learning model.SOLUTION: A processing device 1 comprises a processing machine 2 capable of processing a workpiece W, a detection unit 10 for detecting information about the processing machine 2, and a control unit 3 for controlling an operation of the processing machine 2, wherein: the control unit 3 acquires the above information from the detection unit 10 while processing the workpiece W by the processing machine 2; the control unit uses the acquired information to form a machine learning model used for monitoring the processing machine 2 during processing of the workpiece W; the control unit uses information acquired separately from the information used to form the machine learning model so as to determine whether the machine learning model can be used; and when the control unit determines that the machine learning model can be used, the control unit operates a device while monitoring the processing machine 2 using the machine learning model.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a processing device, a control device for a processing machine, a processing method, and a control program. [Background technology]

[0002] In processing equipment such as presses that press and process workpieces using dies, if processing continues despite an abnormality in the equipment, the situation will worsen, resulting in a large number of defective products, and various other problems such as equipment breakdowns may occur. For this reason, development is underway for devices and systems that can automatically diagnose whether a device is operating normally or if an abnormality has occurred.

[0003] For example, Patent Document 1 discloses an invention of a predictive diagnostic system that diagnoses the condition of mechanical equipment, which acquires time-series data from sensors attached to the equipment while the equipment is in operation, and uses the obtained data to diagnose the condition of the mechanical equipment through statistical methods, thereby diagnosing how long the equipment will be able to operate stably in the future.

[0004] Furthermore, for example, Patent Document 2 discloses an invention of a combustion abnormality prediction device that predicts abnormalities by using a machine learning model that uses an operator's abnormality judgment about a combustion furnace as training data, and that periodically re-learns and updates the machine learning model using plant data up to the point when the operator judged the abnormality and the operator's operation data thereafter. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Japanese Patent Application Laid-Open No. 2016-128973 [Patent Document 2] Patent Publication No. 2021-76371 Summary of the Invention [Problem to be solved by the invention]

[0006] However, the system described in Patent Document 1 has a problem in that a worker such as an operator must manually specify parameters such as thresholds. Furthermore, the operator must set parameters for each type of workpiece or product to be machined by the machine equipment and for each manufacturing condition, which can make the parameter setting process complicated.

[0007] Furthermore, while the device described in Patent Document 2 is a combustion furnace, if it is considered more broadly as a processing device, the actions taken by the operator when an abnormality occurs in the processing device are used as training data, which raises concerns that the quality of the training data may vary depending on the operator's level of proficiency. Furthermore, training data must be manually prepared and provided to the device every time an abnormality occurs, which requires laborious and complicated work for the operator.

[0008] The present invention has been made in consideration of the above points, and aims to provide a processing device, a control device for a processing machine, a processing method, and a control program that can automatically create a machine learning model for monitoring whether a device is operating normally without bothering an operator, and automatically monitor the device using the machine learning model. [Means for solving the problem]

[0009] According to one aspect of the processing device of the present invention, a processing machine capable of processing the workpiece; a detection unit that detects information about the processing machine; a control unit for controlling the operation of the processing machine; Equipped with The control unit While the processing machine is processing the workpiece, the detection unit acquires information about the processing machine; Information about the processing machine acquired Part of to create a machine learning model used to monitor the processing machine during processing of the workpiece, To create the machine learning model was not used Using information about the processing machine, determine whether the machine learning model is usable; When it is determined that the machine learning model is usable, the machine learning model is used to of Make it work.

[0010] According to one aspect of the control device for a processing machine of the present invention, In a control device for a processing machine that controls the operation of a processing machine capable of processing a workpiece, Information about the processing machine acquired from a detection unit that detects information about the processing machine while the processing machine is processing the workpiece. Part of to create a machine learning model used to monitor the processing machine during processing of the workpiece, To create the machine learning model was not used Using information about the processing machine, determine whether the machine learning model is usable; When it is determined that the machine learning model is usable, the machine learning model is used to of Make it work.

[0011] According to one aspect of the processing method of the present invention, an information acquisition step of acquiring information about the processing machine by a detection unit while processing a workpiece with the processing machine; Information about the processing machine acquired Part of a machine learning model creation process for creating a machine learning model to be used for monitoring the processing machine during processing of the workpiece using the above-mentioned method; To create the machine learning model was not used a determination step of determining whether the machine learning model is usable using information about the processing machine; When it is determined that the machine learning model is usable, the machine learning model is used to of a monitoring step of operating the It has.

[0012] According to one aspect of the control program of the present invention, a processing machine capable of processing the workpiece; a detection unit that detects information about the processing machine; A computer of a processing device comprising: an information acquisition means for acquiring information about the processing machine by the detection unit while the processing machine is processing a workpiece; Information about the processing machine acquired Part of a machine learning model creation means for creating a machine learning model to be used for monitoring the processing machine during processing of the workpiece, using the above; To create the machine learning model was not used a determination means for determining whether the machine learning model is usable using information about the processing machine; When it is determined that the machine learning model is usable, the machine learning model is used to of Make it work operation means, Function as. [Effects of the Invention]

[0013] According to the present invention, it is possible to automatically create a machine learning model for monitoring whether a device is operating normally, without bothering workers, and to automatically monitor the device using the machine learning model. [Brief explanation of the drawings]

[0014] [Figure 1] FIG. 2 is a schematic diagram illustrating a configuration example of a press device. [Figure 2] 1 is a flowchart showing a processing flow in a processing method according to the present embodiment. [Figure 3] Graph showing (a) an example of the load applied to the upper mold transmitted from a sensor attached to the upper mold, and (b) an example of the load applied to the lower mold transmitted from a sensor attached to the lower mold. [Figure 4] FIG. 1 shows a set of means and points corresponding to noise plotted in two-dimensional space. [Figure 5]10 is a graph showing, over time, (a) an example of an average calculated from the load applied to the upper die, and (b) an example of an average calculated from the load applied to the lower die. [Figure 6] FIG. 10 is a diagram illustrating an example of a state in which feature vectors are input to a one-class support vector machine and plotted in a multidimensional space, and an example of a discrimination plane. [Figure 7] FIG. 10 is a diagram illustrating a state where the values ​​are plotted closer to the origin than the discriminant plane. [Figure 8] 10A and 10B are diagrams showing examples of display on a display screen. [Figure 9] FIG. 10 is a diagram illustrating an example in which a machine learning model is associated with processing conditions for multiple types of workpieces and stored in a storage unit. [Figure 10] FIG. 10 is a diagram showing an example of processing conditions for a prototype. [Figure 11] Graph showing (a) an example of the average of 20 pieces calculated from the load applied to the upper mold, and (b) an example of the average of 20 pieces calculated from the load applied to the lower mold. DETAILED DESCRIPTION OF THE INVENTION

[0015] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, embodiments of a processing device, a control device for a processing machine, a processing method, and a control program according to the present invention will be described with reference to the drawings. In the following, a case where the processing device is a press device that presses a workpiece to process it will be described, but the present invention is not limited to this case. Also, in the following, a case where the press device is a type that converts the rotation of an eccentric shaft into translational motion of a slide to perform pressing will be described, but the type may also be a press device that uses a cylinder to perform translational motion of the slide, and the form is not limited thereto. Furthermore, in the following description, terms such as up and down, rising, falling, etc. will be used in accordance with the up and down direction in Figure 1 (described later), but the present invention is not limited to this case.

[0016] 1 is a schematic diagram showing an example of the configuration of a press apparatus. In this embodiment, the press apparatus 1 includes a processing machine 2 and a control unit 3. The processing machine 2 includes an eccentric shaft 4 such as an eccentric shaft or a crankshaft, a clutch brake 5, a connecting rod 6, a slide 7, an upper die 8, a lower die 9, and the like.

[0017] The eccentric shaft 4 of the processing machine 2 is composed of an eccentric shaft, a crank shaft, or the like. The clutch / brake 5 may be a wet type that is hydraulically actuated or a dry type that is air-pressure actuated. Also, instead of forming the clutch and brake integrally as the clutch / brake 5 as in this embodiment, the clutch and brake may be configured to be separate.

[0018] When the clutch of the clutch brake 5 is engaged, the eccentric shaft 4 is connected to an external drive source (not shown) such as a motor or a flywheel, causing the eccentric shaft 4 to rotate around its axis. When the clutch is disengaged, the connection between the external drive source and the eccentric shaft 4 is cut off. When the clutch brake 5 is actuated, the rotation of the eccentric shaft 4 is slowed down or stopped.

[0019] The rotational movement of the eccentric shaft 4 around its axis is converted into vertical translational movement of the slide 7 via the connecting rod 6. Therefore, the rotation of the eccentric shaft 4 causes the slide 7 to rise and fall. An upper die 8 is fixed to the underside of the slide 7, and a lower die 9 is fixed below the upper die 8. The upper die 8 is moved by the slide 7 in a direction toward the lower die 9, i.e., lowered, so that the workpiece W is sandwiched between the upper die 8 and the lower die 9 and pressed to process the workpiece W. After the processing, the slide 7 rises to a predetermined stop position and stops.

[0020] In addition, in this embodiment, the upper die 8 and the lower die 9 are equipped with sensors 10 that detect the load F applied to the upper die 8 or the lower die 9 when the workpiece W is pressed and processed by the upper die 8 and the lower die 9 as described above. In this embodiment, the sensor 10 corresponds to the detection unit of the present invention. The sensor 10 transmits information about the detected load F applied to the upper mold 8 or the lower mold 9 to the control unit 3.

[0021] In this embodiment, the following description will be given of a case where the sensor 10, which is the detection unit, detects the load F applied to the upper die 8 and the lower die 9 when pressing and processing the workpiece W as information regarding the processing machine 2. However, the detection unit can also be configured to detect information regarding the processing machine 2, such as the load or pressure applied to the components involved in pressing when pressing and processing the workpiece W, the position or speed of a movable component, the rotation angle or rotation speed of a rotatable component, the temperature of the component, or the temperature of a liquid supplied to or lubricating the component.

[0022] That is, for example, the detection unit can be configured to detect the load applied to members involved in pressing, such as the slide 7, and the pressure applied to the clutch brake 5, etc., when pressing and processing the workpiece W. It is also possible to configure the device to detect the position and speed of movable members such as the slide 7 and upper mold 8. Furthermore, it is also possible to configure it so as to detect the rotation angle and rotation speed around the axis of a rotatable member such as the eccentric shaft 4.

[0023] The control unit 3 may be configured as a general-purpose computer equipped with a CPU (Central Processing Unit) or as a dedicated device. The control unit 3 corresponds to the control device of the processing machine according to the present invention, and operates according to a control program. The control unit 3 includes a display screen 31 and a storage unit 32. Although the display screen 31 and the storage unit 32 are shown in FIG. 1 as being provided in the control unit 3 itself, either or both of the display screen 31 and the storage unit 32 may be provided separately from the control unit 3.

[0024] The control unit 3 controls the operation of the processing machine 2, such as by controlling the clutch brake 5 according to the input motion pattern, etc., to lower the slide 7 to process the workpiece W, or by raising the slide 7 and stopping it at a predetermined stopping position.

[0025] In addition, while the workpiece W is being processed by the processing machine 2, the control unit 3 acquires information about the processing machine 2 using the detection unit, and uses the acquired information to create a machine learning model through machine learning to be used for monitoring the processing machine 2 while the workpiece W is being processed. Then, using information obtained separately from the information used to create the machine learning model, it is determined whether the machine learning model is usable, and if it is determined that the machine learning model is usable, the machine learning model is used to operate the processing machine 2 while monitoring whether it is operating normally.

[0026] In this embodiment, the control unit 3 is configured to continue operation of the processing machine 2 if it determines that the information about the processing machine 2 acquired by the detection unit is normal based on the machine learning model that is determined to be usable. In this embodiment, when the control unit 3 determines based on the machine learning model that the information regarding the processing machine 2 acquired by the detection unit indicates that an abnormality has occurred, it outputs information regarding the occurrence of the abnormality and notifies the user that an abnormality has occurred.

[0027] Hereinafter, the creation of the above-mentioned machine learning model and the monitoring of the processing machine 2 by the control unit 3 of the press machine 1, which is the processing device according to this embodiment, will be specifically described. In addition, a processing method according to this embodiment will also be described with reference to FIG.

[0028] First, the control unit 3 starts the operation of the press machine 1 (step S1). That is, after an external drive source such as a motor or a flywheel is operated, the slide 7 is moved up and down by controlling the clutch brake 5 of the processing machine 2, and processing of the workpiece W is started.

[0029] Next, the control unit 3 acquires information about the processing machine 2 using the detection unit while the processing machine 2 is processing the workpiece W (information acquisition step, step S2). That is, in this embodiment, when the slide 7 and the upper die 8 are moved up and down to process the workpiece W, the control unit 3 receives and acquires information about the load F applied to the upper die 8 or the lower die 9 transmitted from the sensor 10, which is the detection unit.

[0030] Figure 3(a) is an example of the load F8 applied to the upper mold 8 transmitted from the sensor 10 attached to the upper mold 8, and Figure 3(b) is an example of the load F9 applied to the lower mold 9 transmitted from the sensor 10 attached to the lower mold 9. 3(a) and 3(b) show examples of the time transition of the loads F8 and F9 during one press cycle. For simplicity's sake, the following description will be made on the case where processing is performed on information on the two loads F8 and F9 obtained from the two sensors 10. However, in reality, information on the load F, i.e., information on the processing machine 2, is obtained for each sensor 10, i.e., detection unit, and the following processing is performed on all of them.

[0031] When the control unit 3 acquires the information on the loads F8 and F9, it calculates the average x8 of the distribution shown in Fig. 3(a) as a feature quantity of the acquired load F8 applied to the upper mold 8. Similarly, for the acquired load F9 applied to the lower mold 9, it calculates the average x9 of the distribution of the load F9 shown in Fig. 3(b) as a feature quantity of the load F9. The feature values ​​do not necessarily have to be the averages x8 and x9 of the loads F8 and F9, but may be, for example, standard deviations, and it is possible to configure the system to calculate appropriate feature values ​​based on the data shown in Figures 3(a) and 3(b).

[0032] Furthermore, the data shown in Figures 3(a) and (b) are obtained each time pressing is performed, and the averages x8 and x9 calculated from the loads F8 and F9 applied to the upper die 8 and the lower die 9 obtained at the same time are hereinafter referred to as the set of averages x8 and x9. That is, each time the workpiece W is pressed, the control unit 3 calculates a set of averages x8 and x9.

[0033] In addition, in the early stages of operation of the press machine 1, the data detected by the sensor 10 may contain noise. In such a case, for example, points consisting of each calculated feature amount, i.e., in the above case, points (x8, x9) corresponding to the pair of averages x8 and x9, can be plotted in a two-dimensional space with each average on the horizontal and vertical axes as shown in Figure 4, and abnormal data can be removed using a technique such as clustering.

[0034] In this case, since the amount of noise is usually less than the amount of correctly detected data, it is possible to configure the system to remove, as noise, the points belonging to the cluster with the fewer points out of the two clusters obtained as a result of clustering, i.e., the pair of averages x8 and x9. This noise removal process is carried out as needed.

[0035] Next, the control unit 3 uses the acquired information about the processing machine 2 to create a machine learning model used to monitor the processing machine 2 while processing the workpiece W through machine learning (machine learning model creation step, step S3). In this embodiment, the control unit 3 creates a machine learning model using unsupervised learning as the machine learning. In the following, a case where a model is created using a one-class support vector machine as the unsupervised learning will be described, but it is also possible to configure the control unit 3 to use, for example, a method using Mahalanobis distance.

[0036] To explain more specifically below, let us assume that 30 sets of averages x8 and x9 calculated from the loads F8 and F9 applied to the upper mold 8 and the lower mold 9 obtained as described above are obtained, as shown in Figures 5(a) and 5(b). At that point, the control unit 3 arbitrarily selects, for example, 20 pairs of averages x8 and x9 indicated by black circles in Figures 5(a) and 5(b) from the 30 pairs of averages x8 and x9, and sets each as a feature vector (x8, x9). Note that instead of selecting 20 pairs of averages x8 and x9 when 30 pairs of averages x8 and x9 are obtained in this way, it may be configured such that, for example, 20 pairs of averages x8 and x9 are obtained and set them as feature vectors (x8, x9).

[0037] The control unit 3 then inputs each feature vector (x8, x9) into a one-class support vector machine and plots it in a multidimensional space as shown in FIG. 6. In this case, it plots it in a two-dimensional space. Then, of the planes that separate the area including the origin from the area including all 20 plots, the plane that is the farthest from the origin is determined as the discrimination plane P. In this case, since the multidimensional space is a two-dimensional space, the discrimination plane P is a straight line. In this way, the control unit 3 creates the discrimination plane P as a machine learning model used to monitor the processing machine 2 while the workpiece W is being processed.

[0038] Next, the control unit 3 tests the created machine learning model. That is, it determines whether the machine learning model is usable or not by using information about the processing machine 2 that is acquired separately from the information about the processing machine 2 that was used to create the machine learning model (determination step, step S4). That is, in this embodiment, the control unit 3 inputs the remaining 10 pairs of averages x8 and x9, for example, indicated by white circles in Figures 5(a) and (b) out of the 30 pairs of averages x8 and x9 as described above, into a one-class support vector machine as feature vectors (x8, x9), and plots them in the same multidimensional space as the multidimensional space shown in Figure 6.

[0039] As a result, if the result is plotted closer to the origin than the discrimination plane P as the machine learning model created as described above, as shown in FIG. 7, the control unit 3 determines that the created machine learning model, i.e., the discrimination plane P, cannot be used (step S4; NO). In this case, the control unit 3 discards the created machine learning model, i.e., the discrimination plane P (step S5), and restarts the process from the information acquisition step (step S2) in which the detection unit acquires information about the processing machine 2 while processing the workpiece W using the processing machine 2.

[0040] On the other hand, when the remaining 10 pairs of averages x8 and x9 are input as feature vectors (x8, x9) to the one-class support vector machine as described above, if they are not plotted closer to the origin than the discrimination plane P as the machine learning model, the control unit 3 determines that the created machine learning model, i.e., the discrimination plane P, is usable (step S4; YES). Then, the control unit 3 adopts the machine learning model that has been determined to be usable, that is, the discrimination plane P (step S6).

[0041] Then, the control unit 3 operates the processing machine 2 based on the adopted machine learning model, i.e., the machine learning model determined to be usable, while monitoring whether the processing machine 2 is operating normally (monitoring process, step S7). That is, each time a press is performed, the average pair x8 and x9 of the loads F8 and F9 transmitted from the sensor 10 is calculated, and the feature vector (x8, x9) is input into a one-class support vector machine and plotted in a multidimensional space.

[0042] Then, if the position of the plot in the multidimensional space is on the discrimination plane P or farther from the origin than the discrimination plane P, the control unit 3 determines that the processing machine 2 is operating normally (step S8; NO), and continues monitoring while continuing to operate the processing machine 2 (step S7). Furthermore, if the position of the plot in the multidimensional space is closer to the origin than the discrimination plane P, the control unit 3 determines that an abnormality has occurred in the processing machine 2 (step S8; YES), and outputs information regarding the occurrence of the abnormality (step S9) to notify the user that an abnormality has occurred. If necessary, the operator stops the operation of the processing machine 2. Alternatively, the control unit 3 may be configured to automatically stop the operation of the processing machine 2.

[0043] As described above, in the machining device, machining control device, machining method, and control program according to the present embodiment, the control unit 3 performs training of a one-class support vector machine, for example, by using a set of averages x8 and x9 calculated from the loads F8 and F9 transmitted as information related to the machining device 2 from the sensor 10, which is the detection unit, and creates, through machine learning, a discrimination plane P used to monitor the machining device 2 during machining of the workpiece W. The created machine learning model is then tested, and if the machine learning model is usable, the machine learning model is used to operate the machining device 2 while monitoring whether it is operating normally.

[0044] In this way, in the processing device, processing machine control device, processing method, and control program of this embodiment, the control unit 3 automatically creates a machine learning model using machine learning to monitor whether the device is operating normally, making it possible to automatically create a machine learning model without bothering workers, and to automatically monitor the device using that machine learning model. This eliminates the need for workers to perform the tedious and complicated task of creating machine learning models, and also makes it possible to prevent machine learning models from being produced with poor quality depending on the worker's level of expertise.

[0045] In this embodiment, as described above, the case where unsupervised learning is performed as machine learning has been described, but machine learning is not limited to unsupervised learning, and may be supervised learning, reinforcement learning, or the like.

[0046] Furthermore, even when processing the same type of workpiece W, if the state of the processing device 1 such as a press device or external environmental factors change, the above-mentioned discrimination plane P, i.e., the machine learning model used to monitor the processing machine 2, may also change. Therefore, it is desirable to configure the processing device 1 so that a machine learning model is created each time the device 1 starts operating, as shown in Fig. 2. In addition, if the processing device 1 is operated continuously for a long period of time, it is also possible to configure the processing device 1 so that a machine learning model is created periodically, even when the device 1 does not start operating.

[0047] Furthermore, in this embodiment, as described above, when the control unit 3 determines that an abnormality has occurred, it outputs information regarding the occurrence of the abnormality (step S9) to notify the user that an abnormality has occurred. In this case, it is possible to configure the controller 3 to notify the user by displaying a message on the display screen 31 of the controller 3, by sound or light emission, or by simultaneously performing all of these.

[0048] When the abnormality is to be displayed on the display screen 31 and reported, for example, it is possible to calculate the abnormality level of the set of the average x8 and x9 of the loads F8 and F9 and display it in chronological order. The degree of abnormality can be calculated, for example, as the reciprocal of the distance from the origin of a plot in a multidimensional space corresponding to the set of averages x8 and x9 of the loads F8 and F9. The degree of abnormality may also be calculated by other methods.

[0049] 8(a), for example, the abnormality degrees of the sets of averages x8 and x9 of the loads F8 and F9 used when it was determined that the processing machine 2 was operating normally are displayed in chronological order on the upper side of the display screen 31. In this case, since the processing machine 2 is operating normally, each abnormality degree has a small value. The degree of abnormality up to the point where it is determined that an abnormality has occurred can be displayed in chronological order on the lower side of the display screen 31, and normal cases and cases where an abnormality has occurred can be displayed side by side.

[0050] Furthermore, for example, as shown in Figure 8(b), it is also possible to configure the display to superimpose in chronological order the abnormality level of the set of averages x8 and x9 of the loads F8 and F9 used when it was determined that the processing machine 2 was operating normally, and the abnormality level up until it was determined that an abnormality had occurred. In this way, by configuring the display to display the abnormality levels in parallel or in superimposed fashion, it is possible to visually notify workers and others that an abnormality has occurred in an easily understandable manner.

[0051] However, when configured as described above, there may be a state in which there is no machine learning model until the control unit 3 creates a machine learning model, determines whether the created machine learning model is usable, and begins monitoring whether the processing machine 2 is operating normally based on the machine learning model that is determined to be usable. Therefore, in order to avoid this situation, it is possible to create a standard machine learning model in advance, and configure the system so that this standard machine learning model is used to monitor whether the processing machine 2 is operating normally during the above period.

[0052] It is also possible to configure it as follows. For example, if the processing machine 2 is capable of processing a plurality of types of workpieces W, the memory unit 32 stores the past machine learning models created for each of the plurality of types of workpieces W. In this case, as shown in Fig. 9, the machine learning model can be configured to be stored in association with processing conditions such as product name, mold number, product shape, workpiece shape, temperature [°C] during pressing, and weight [kg]. In Fig. 9, B represents width [mm], L represents length [mm], and T represents height [mm].

[0053] The control unit 3 then searches for past machine learning models stored in the memory unit 32, and if a past machine learning model created for the same type of workpiece W as the workpiece W to be processed this time is stored in the memory unit 32, the control unit 3 can be configured to read out the past machine learning model from the memory unit 32 and operate the processing machine 2 while monitoring it using the past machine learning model until a new machine learning model is created and monitoring of the processing machine 2 begins based on the machine learning model.

[0054] As mentioned above, even when processing the same type of workpiece W, the machine learning model used to monitor the processing machine 2 may change if the state of the processing device 1, such as a press device, or external environmental factors change. However, when machining the same type of workpiece W, the machine learning model rarely changes drastically. Therefore, by configuring as described above, it is possible to effectively monitor the processing machine 2 using the previous machine learning model until a new machine learning model is created and monitoring of the processing machine 2 based on the new machine learning model begins.

[0055] In this case, the newly created machine learning model can be configured to be re-linked to the type of the corresponding work W and stored in the storage unit 32. In this case, the new model may be stored by overwriting a previous machine learning model, or the previous machine learning model may be configured to be kept as a history.

[0056] On the other hand, if no past machine learning model is stored in the storage unit 32 for the type of workpiece W to be machined this time, the following configuration can be used, for example. For example, when a prototype is made using the processing device 1, processing is performed on a new type of workpiece W that has never been processed before. Therefore, the storage unit 32 does not store any past machine learning models for the type of workpiece W of the prototype to be processed.

[0057] Therefore, if the control unit 3 does not have a past machine learning model stored in the memory unit 32 for the type of workpiece W to be processed, such as a prototype, the control unit 3 can be configured to read from the memory unit 32 a past machine learning model for another type of workpiece W whose processing conditions are closest to the processing conditions of the workpiece W to be processed, and to operate the processing machine 2 while monitoring it using the read past machine learning model until a new machine learning model is created for the workpiece W to be processed and monitoring of the processing machine 2 begins based on the machine learning model.

[0058] That is, for example, if the processing conditions for prototype ZZ are as shown in Figure 10, the processing conditions that are closest to these processing conditions among the products a to c shown in Figure 9 are the processing conditions for product B, so in this case, the control unit 3 reads out from the memory unit 32 a past machine learning model b for the workpiece W of product B, which is a different type from prototype ZZ. In practice, the similarity of the processing conditions can be determined by using, for example, the least squares method.

[0059] By configuring it in this manner, even when processing a type of workpiece W that has never been processed before with the processing device 1, such as a prototype, it is possible to use past machine learning models for types of workpiece W with similar processing conditions to create a new machine learning model and monitor the processing machine 2 effectively until monitoring of the processing machine 2 begins based on the machine learning model. In this case, when the control unit 3 creates a new machine learning model and starts monitoring the processing machine 2 based on the machine learning model, it then operates the processing machine 2 while monitoring whether the processing machine 2 is operating normally using the newly created machine learning model.

[0060] In the above configuration, the control unit 3 may determine that an abnormality has occurred based on a past machine learning model while the detection unit 10 is detecting information about the processing machine as shown in Figures 3(a) and 3(b) in order to create a new machine learning model. In that case, as described above, an operator may stop the operation of the processing machine 2, or the control unit 3 may automatically stop the operation of the processing machine 2. That is, for example, in the above example, it is possible that the operation of the processing machine 2 will stop when the detection unit 10 detects 20 sets of data on the loads F8 and F9 applied to the upper mold 8 and the lower mold 9, before detecting 30 sets of data, as shown in Figures 11(a) and (b).

[0061] Therefore, if the detection unit 10 is detecting information about the processing machine 2 to create a new machine learning model and determines that an abnormality has occurred based on a past machine learning model, the control unit 3 can be configured to obtain information about the processing machine 2 that the detection unit 10 has detected up to that point, and use the obtained information about the processing machine 2 to create a machine learning model through machine learning, as described above.

[0062] That is, in the example shown in Figures 11(a) and (b), the control unit 3 acquires data on 20 pairs of loads F8 and F9 detected by the detection unit 10, and uses this data to create a machine learning model through machine learning. By configuring it in this manner, it becomes possible to create a machine learning model using information about the processing machine 2 detected between the time when the control unit 3 determines that an abnormality has occurred based on past machine learning models and the time when the control unit 3 stops the operation of the processing machine 2, without discarding this information, and it becomes possible to effectively use the detected information in creating the machine learning model.

[0063] It goes without saying that the present invention is not limited to the above-described embodiments, and can be modified as appropriate without departing from the spirit of the present invention. For example, in the above embodiment, the control unit 3 creates a machine learning model using 20 sets of feature quantities out of 30 sets of averages x8 and x9 calculated from the loads F8 and F9 applied to the acquired upper mold 8 and lower mold 9, that is, 30 sets of feature quantities, and tests the machine learning model using the remaining 10 sets of feature quantities. However, the number of feature quantities used for creating and testing the machine learning model is not limited to this case. [Explanation of symbols]

[0064] 1 Processing equipment, press equipment 2 Processing machine 3. Control Unit 7 slides 8. Mold 10. Detection unit 32 Storage section Information on F8 and F9 processing machines, load double work

Claims

1. a processing machine capable of processing the workpiece; a detection unit that detects information about the processing machine; a control unit for controlling the operation of the processing machine; Equipped with The control unit While the processing machine is processing the workpiece, the detection unit acquires information about the processing machine; Using a portion of the acquired information about the processing machine, a machine learning model is created to be used for monitoring the processing machine while the workpiece is being processed; determining whether the machine learning model is usable using information about the processing machine that was not used to create the machine learning model; When it is determined that the machine learning model is usable, the machine learning model is used to operate the processing machine. Processing equipment.

2. the control unit continues the operation of the processing machine when it determines that the information about the processing machine acquired by the detection unit is normal based on the machine learning model that is determined to be usable. The processing device according to claim 1 .

3. When the control unit determines that the information about the processing machine acquired by the detection unit is abnormal based on the machine learning model determined to be usable, the control unit outputs information about the occurrence of the abnormality. The processing device according to claim 1 or 2.

4. the control unit creates the machine learning model using unsupervised learning as the machine learning. The processing device according to any one of claims 1 to 3.

5. The processing machine is capable of processing a plurality of types of the workpieces, a storage unit that stores the past machine learning models created for each of the plurality of types of work, until the machine learning model is newly created and monitoring of the processing machine is started based on the newly created machine learning model, the past machine learning model is read out, and the processing machine is operated while being monitored using the past machine learning model. The processing device according to any one of claims 1 to 4.

6. When the previous machine learning model created for the workpiece of the same type as the workpiece to be machined is stored in the storage unit, the control unit reads out the previous machine learning model from the storage unit. The processing device according to claim 5.

7. When the past machine learning model for the type of workpiece to be machined is not stored in the storage unit, the control unit reads out from the storage unit the past machine learning model for another type of workpiece having machining conditions closest to the machining conditions of the workpiece to be machined. The processing device according to claim 5 or 6.

8. When the control unit determines that an abnormality has occurred based on the past machine learning model while the detection unit is detecting information about the processing machine in order to newly create the machine learning model, the control unit acquires information about the processing machine that has been detected by the detection unit up to that point in time, and creates the machine learning model by machine learning using the acquired information about the processing machine. The processing device according to claim 6 or 7.

9. The processing device is a press device that moves a die fixed to a slide to press and process the workpiece. The processing device according to any one of claims 1 to 8.

10. The detection unit detects, as information related to the processing machine, a load or pressure applied to a member involved in pressing when pressing the workpiece to process it, a position or speed of a movable member, a rotation angle or rotation speed of a rotatable member, a temperature of a member, or a temperature of a liquid supplied to or lubricating a member. The processing device according to claim 9.

11. In a control device for a processing machine that controls the operation of a processing machine capable of processing a workpiece, creating a machine learning model to be used for monitoring the processing machine during processing of the workpiece, using a portion of information about the processing machine acquired from a detection unit that detects information about the processing machine while the workpiece is being processed by the processing machine; determining whether the machine learning model is usable using information about the processing machine that was not used to create the machine learning model; When it is determined that the machine learning model is usable, the machine learning model is used to operate the processing machine. Control device for processing machines.

12. an information acquisition step of acquiring information about the processing machine by a detection unit while processing a workpiece with the processing machine; a machine learning model creation step of creating a machine learning model to be used for monitoring the processing machine during processing of the workpiece, using a portion of the acquired information about the processing machine; a determination step of determining whether the machine learning model is usable using information about the processing machine that was not used to create the machine learning model; a monitoring step of operating the processing machine using the machine learning model when it is determined that the machine learning model is usable; having Processing method.

13. a processing machine capable of processing the workpiece; a detection unit that detects information about the processing machine; A computer of a processing device comprising: an information acquisition means for acquiring information about the processing machine by the detection unit while the processing machine is processing a workpiece; a machine learning model creation means for creating a machine learning model to be used for monitoring the processing machine during processing of the workpiece, using a portion of the acquired information about the processing machine; a determination means for determining whether the machine learning model is usable by using information about the processing machine that was not used in creating the machine learning model; an operating means for operating the processing machine using the machine learning model when it is determined that the machine learning model is usable; A control program that functions as a

Images