Machine learning device, information processing device, inference device, machine learning method, information processing method, and inference method

Abstract

To provide a machine leaning device that generates a learning model for inferring information on maintenance on the basis of the state and characteristics of a press working machine.SOLUTION: A machine learning device 4 for generating a learning model 44a includes: a training data storage unit 42 which stores a plurality of sets of training data which respectively include state information indicating a state of a prescribed part of an electric press working machine, input information including characteristic information indicating characteristics of the prescribed part of the electric press working machine, and maintenance information on maintenance of the electric press working machine and in which the input information and the maintenance information are associated with each other; a machine learning unit 43 which makes the learning model 44a learn a correlation between the input information and the maintenance information by inputting a plurality of sets of training data to the learning model 44a; and a learned model storage unit 44 for storing the learning model 44a that been trained by the machine learning unit 43.SELECTED DRAWING: Figure 7

Description

[Technical Field]

[0001] The present invention relates to a machine learning device, an information processing device, an inference device, a machine learning method, an information processing method, and an inference method. [Background technology]

[0002] Conventionally, a die press device has been disclosed that displays the management status taking into account not only the number of uses but also the storage status, etc., by displaying the usage history and posting history of the die in detail (see Patent Document 1). According to the technology described in Patent Document 1, it is possible to display the management status taking into account not only the number of uses but also the storage status, etc., thereby realizing more reliable history management of processing tools. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 2018-063668 Summary of the Invention [Problem to be solved by the invention]

[0004] However, the mold press device in Patent Document 1 only displays the management status taking into account the number of times used and storage conditions, and does not take into account the condition and characteristics of the mold, etc., nor does it predict the need for maintenance.

[0005] The present invention aims to provide a machine learning device that can be used for various types of press machines and that generates a learning model for inferring maintenance information based on the condition and characteristics of the press machine. [Means for solving the problem]

[0006] The machine learning device according to the present invention comprises: A machine learning device that generates a learning model for inferring information regarding maintenance of an electric press machine, a learning data storage unit that stores a plurality of sets of learning data in which the input information and the maintenance information are associated with each other, the learning data storage unit including input information including status information indicating a status of a predetermined portion of the electric press working machine and characteristic information indicating characteristics of the predetermined portion of the electric press working machine, and maintenance information related to maintenance of the electric press working machine; a machine learning unit that inputs a plurality of sets of the learning data into the learning model, thereby causing the learning model to learn a correlation between the input information and the maintenance information; and a trained model storage unit that stores the trained model trained by the machine learning unit. [Effects of the Invention]

[0007] The machine learning device of the present invention can be used for various types of press machines and can generate a learning model for inferring maintenance information based on the state and characteristics of the press machine. [Brief explanation of the drawings]

[0008] [Figure 1] 1 shows an example of an information processing system 1 according to the present embodiment. [Figure 2] An example of information handled by the information processing system 1 of this embodiment is shown. [Figure 3] 1 shows an example of an electric press machine 100 according to this embodiment. [Figure 4] 1 shows an example of a slide mechanism of the electric press machine 100 of this embodiment. [Figure 5] 1 shows an example of a horizontal cross section including a slider 111 of the electric pressing machine 100 of this embodiment. [Figure 6] 1 shows an example of a system configuration of an electric press machine 100 according to this embodiment. [Figure 7] 1 shows an example of the system configuration of a machine learning device 4 of an information processing system 1 according to this embodiment. [Figure 8] 1 shows an example of a neural network model used in the machine learning device 4 of this embodiment. [Figure 9] 1 shows an example of a flowchart of a machine learning method performed by the machine learning device 4 of this embodiment. [Figure 10] 1 shows an example of the system configuration of an information processing device 5 of this embodiment. [Figure 11] 1 shows an example of a flowchart of a wear information inference method according to the present embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0009] Fig. 1 shows an example of an information processing system 1 of this embodiment. Fig. 2 shows an example of information handled by the information processing system 1 of this embodiment. Fig. 2(a) shows an example of input information 10, Fig. 2(b) shows an example of status information 11, Fig. 2(c) shows an example of characteristic information 12, and Fig. 2(d) shows an example of maintenance information 13.

[0010] The information processing system 1 functions as a system for managing a state in which maintenance of the electric pressing machine 100 is required. The information processing system 1 mainly includes a physical quantity measuring device 2 that measures the physical quantity of a measurement target, a database device 3 that stores various information, a machine learning device 4 that creates a learning model 44a (described later), an information processing device 5 that infers a state in which maintenance of the electric pressing machine 100 is required using the learning model 44a, and a user terminal device 6 operated by a user. Each of the devices 2 to 6 is, for example, configured as a general-purpose or dedicated computer, and is connected to a wired or wireless network 7 so that various data can be transmitted and received between them. Note that the number of the devices 2 to 6 and the connection configuration of the network 7 are not limited to the example shown in FIG. 1 and may be changed as appropriate.

[0011] The physical quantity measuring device 2 measures the physical quantity of a measurement target and sets the measured physical quantity as input information 10. The measurement target in this embodiment may be, for example, at least one of the electric press machine 100, the die 105, the workpiece, or the lubricant. The physical quantity measuring device 2 in this embodiment acquires, as the input information 10, state information 11 obtained by measuring the state of the electric press machine 100, the die 105, the workpiece, or the lubricant, and transmits the state information 11 to at least one of the database device 3, the machine learning device 4, the information processing device 5, and the user terminal device 6 via the network 7.

[0012] The status information 11 indicates the status of a predetermined part of the electric press machine 100. For example, the status information 11 includes at least one of main body status information 11a indicating the status of the main body of the electric press machine 100, die status information 11b indicating the status of the die 105, workpiece status information 11c indicating the status of the workpiece, and lubricant status information 11d indicating the status of the lubricant. The status information 11 is time-series data for each predetermined number of shots since the die 105 was replaced.

[0013] The main body status information 11a may be, for example, at least one of the slider load, slider position, or number of shots. The slider load may be the load applied to the slider 111 during processing, and a load (strain) sensor may be used as the physical quantity measuring device 2. The slider position may be the position of the slider 111 during processing, and a position sensor for the slider 111 or a measuring device that captures an image of the slider 111 with a camera and measures the position by image processing may be used as the physical quantity measuring device 2. The number of shots may be the total number of processing operations since the mold 105 was replaced, and a counter or cam switch that measures the number of shots of the slider 111 may be used as the physical quantity measuring device 2.

[0014] The mold state information 11b may be at least one of mold load, mold temperature, and mold displacement. The mold load may be the load applied to the mold 105 during processing, and a load (strain) sensor may be used as the physical quantity measuring device 2. The mold temperature may be the temperature of the mold 105 during processing, and a temperature sensor or a thermocouple may be used as the physical quantity measuring device 2. The mold displacement may be the change or difference in the dimensions of the mold 105 before and after processing, and a measuring instrument that captures an image of the mold 105 with a camera and measures the position by image processing, or a displacement sensor may be used as the physical quantity measuring device 2.

[0015] The workpiece condition information 11c may be at least one of post-processing dimensions, surface properties, workpiece temperature, and workpiece displacement. The post-processing dimensions may be the dimensions of the workpiece after processing, and the physical quantity measuring device 2 may be a load (strain) sensor, a measuring instrument that captures an image of the workpiece and measures the dimensions through image processing, or a displacement sensor. The surface properties may be the surface properties of the workpiece after processing, and the physical quantity measuring device 2 may be a measuring instrument that captures an image of the workpiece and measures the surface roughness through image processing. The workpiece temperature may be the temperature of the workpiece during processing, and a temperature sensor may be used as the physical quantity measuring device 2. The workpiece displacement may be the change or difference in the dimensions of the workpiece before and after processing, and the physical quantity measuring device 2 may be a measuring instrument that captures an image of the mold 105 with a camera and measures the position through image processing, or a displacement sensor.

[0016] The lubricant state information 11d may be, for example, at least one of lubricant viscosity, ambient temperature, or ambient humidity. The lubricant viscosity may be the viscosity of the lubricant during processing, and a viscometer may be used as the physical quantity measuring device 2. The lubricant temperature may be the temperature of the lubricant during processing, and a temperature sensor or a thermocouple may be used as the physical quantity measuring device 2. The ambient humidity may be the ambient humidity during processing, and a humidity sensor may be used as the physical quantity measuring device 2.

[0017] The database device 3 stores various types of information to be stored, and sets this as input information 10. In this embodiment, the storage object may be, for example, at least one of a mold 105, a workpiece, or a lubricant. The database device 3 in this embodiment stores characteristic information 12 indicating the characteristics of the mold 105, the workpiece, or the lubricant as input information 10 input from a user terminal device 6 or the like, and transmits the stored characteristic information 12 to at least one of the machine learning device 4, the information processing device 5, or the user terminal device 6 via the network 7.

[0018] The characteristic information 12 indicates the characteristics of a predetermined part of the electric press machine 100. For example, the characteristic information 12 includes at least one of die characteristic information 12a indicating the characteristics of the die 105, workpiece characteristic information 12b indicating the characteristics of the workpiece, and lubricant characteristic information 12c indicating the characteristics of the lubricant.

[0019] The mold characteristic information 12a may be at least one of the mold material name, the mold load characteristic, and the mold temperature characteristic. The mold material name may be the name of the material used in the mold 105, or may be the model number of the mold 105, etc. The mold load characteristic may be the strength against the load of the mold 105, the deformation rate, etc. The mold temperature characteristic may be the strength against the temperature of the mold 105, the deformation rate, etc. For example, it is preferable to associate the mold load characteristic and the mold temperature characteristic with the mold material name and store them.

[0020] The workpiece characteristic information 12b may be at least one of the workpiece material name, workpiece load characteristics, or workpiece temperature characteristics. The workpiece material name may be the name of the material used for the workpiece. The workpiece load characteristics may be the strength, deformation rate, etc. of the workpiece against the load. The workpiece temperature characteristics may be the strength, deformation rate, etc. of the workpiece against the temperature. For example, it is preferable to associate the workpiece load characteristics and workpiece temperature characteristics with the workpiece material name and store them.

[0021] The lubricant characteristic information 12c may be at least one of the lubricant material name, lubricant temperature characteristic, and lubricant humidity characteristic. The lubricant material name may be the name of the material used in the lubricant. The lubricant temperature characteristic may be the viscosity, expansion rate, or contraction rate of the lubricant relative to the temperature. The lubricant humidity characteristic may be the viscosity, expansion rate, or contraction rate of the lubricant relative to the humidity. For example, it is preferable to associate the lubricant temperature characteristic and the lubricant humidity characteristic with the lubricant material name and store them.

[0022] The machine learning device 4 operates as a subject of the learning phase of machine learning, and acquires a learning dataset used for machine learning created from input information 10 and maintenance information 13 acquired from at least one of the physical quantity measuring device 2, the database device 3, the information processing device 5, or the user terminal device 6, and generates by machine learning a learning model 44a (described later) to be used in the information processing device 5. The trained learning model 44a is provided to the information processing device 5 via the network 7, a recording medium, or the like.

[0023] As shown in FIG. 2(d), the maintenance information 13 may be at least one of a deteriorated location, a deterioration state, or a maintenance timing. The deteriorated location may be a location that has deteriorated due to wear or the like, or a location that is close to the time of wear or the like, such as the main body, the mold 105, or the lubricant. The deterioration state indicates the state of deterioration such as wear. For example, the input information 10 indicates the deterioration state of the main body, such as the molding speed, load, position, drive system, mold mounting state, sensors, and foreign matter contamination. The input information 10 also indicates the deterioration state of the mold 105, such as thermal expansion / contraction, shear fracture surface ratio, burr / sagging length, roughness and scratches, punch-die clearance wear, wear of the punch and die themselves, or foreign matter contamination. The input information 10 also indicates the deterioration state of the lubricant, such as foreign matter contamination or viscosity reduction. The maintenance timing indicates the time when maintenance is required. Here, deterioration includes at least one of wear of the mold 105 or lubricant deterioration.

[0024] The information processing device 5 operates as the main body of the inference phase of machine learning, and uses the learning model 44a generated by the machine learning device 4 to predict the locations or timing of maintenance required for the electric press processing machine 100 from the input information 10, and transmits the predicted maintenance information 13 for the electric press processing machine 100 to at least one of the user terminal devices 6, etc.

[0025] The user terminal device 6 is a terminal device used by a user, and may be a stationary device or a portable device. The user terminal device 6 accepts various input operations via a display screen of, for example, an application program, a web browser, or the like, and displays various information such as input information 10 or maintenance information 13 on the display screen.

[0026] The network 7 is configured by wired communication, wireless communication, or a combination of wired communication and wireless communication according to any communication standard. Specifically, for example, a standardized communication network such as the Internet, a communication network managed within a building such as a local network, or a combination of these communication networks can be used.

[0027] Each of the devices 3 to 6 is configured, for example, as a general-purpose or dedicated computer, and is configured to be able to transmit and receive various data to and from each other via the network 7. The number of each of the devices 3 to 6 is not limited to the example in FIG. 1, and may be one or more.

[0028] Fig. 3 shows an example of the electric press machine 100 of this embodiment. Fig. 4 shows an example of a slide mechanism of the electric press machine 100 of this embodiment. Fig. 5 shows an example of a horizontal cross section including a slider 111 of the electric press machine 100 of this embodiment. Note that the support pillars 102 and crown 103 are omitted from Fig. 4.

[0029] The electric press machine 100 includes a bed 101, a support 102, a crown 103, a scale post 104, a slider 111, a motor 120 as a drive unit, a ball screw 130 as a power transmission unit, and a position detection unit 140.

[0030] The bed 101 is a base member for placing the electric press machine 100 on the ground. The support pillars 102 are columns supported by the bed 101 and extend upward. In this embodiment, there are four support pillars 102, which are installed near the four corners of the bed 101. The crown 103 is placed on the support pillars 102 and mounts the motor 120 thereon. The bed 101, the support pillars 102, and the crown 103 form the frame of the electric press machine 100. Note that the number of support pillars 102 is not limited to four, and it is sufficient that there are at least two or more support pillars that can support the crown 103. Furthermore, the support pillars are not limited to being columnar, and may be plate-shaped.

[0031] The slider 111 is movably attached to the support 102. In this embodiment, the four corners of the slider 111 are movably installed on the support 102. The slider 111 has a slight gap or the like formed between it and the support 102, giving it a tiltable structure.

[0032] The motor 120 is mounted on the crown 103 and drives a ball screw 130, which serves as a power transmission unit. As shown in FIG. 4, the ball screw 130 has a screw shaft 130a and a connecting portion 130b. The screw shaft 130a passes through the crown 103 and is connected to the output shaft of the motor 120. The connecting portion 130b is a ball joint or the like. Therefore, the ball screw 130 mounts the slider 111 so that the slider 111 can rotate in any direction, and transmits the driving force generated by the motor 120 to the slider 111. As a result, the motor 120 drives the slider 111.

[0033] In the first example, there are four motors 120, each corresponding to the four corners of the crown 103 and the slider 111. The four motors 120 and the four ball screws 130 operate independently. The number of motors 120 is not limited to four, but at least two or more may be used.

[0034] The position detection unit 140 is preferably a linear scale or the like that reads the scale post 104 and measures the height at which the slider 111 is positioned relative to the bed 101. In this embodiment, there are four of them, corresponding to the four corners of the slider 111. Note that at least two or more position detection units 140 are required.

[0035] The scale post 104 is attached vertically to the bed 101 at one end and the crown 103 at the other end. In this embodiment, the scale post 104 is attached to the outside of the slider 111 near the four corners.

[0036] The mold 105 has an upper mold 105a and a lower mold 105b. The upper mold 105a is attached to a slider 111 and moves together with the slider 111. The lower mold 105b is attached to a bed 101. The upper mold 105a and the lower mold 105b are arranged opposite each other, and a material is processed by being sandwiched between the upper mold 105a and the lower mold 105b.

[0037] 5, the ball screw 130 of the electric press machine 100 of the first example is designated as a first ball screw 131 at the top right of the page, and counterclockwise therefrom as a second ball screw 132, a third ball screw 133, and a fourth ball screw 134. The motor 120 and the position detection unit 140 correspond to the ball screw 130 in the same manner, and the first to fourth numbers are assigned to them.

[0038] FIG. 6 shows an example of the system configuration of the electric press machine 100 of this embodiment.

[0039] The electric press machine 100 has an operation panel 190 operated by a user, a control unit 170 that drives and controls the first motor 121 to the fourth motor 124 for the first axis to the fourth axis in response to commands from the operation panel 190, and a memory unit 180 that pre-stores information regarding the drive energy for each stage to be supplied to each of the first motor 121 to the fourth motor 124.

[0040] Also, corresponding to each axis, there are provided a first servo amplifier 161 to a fourth servo amplifier 164 that receive signals from the control unit 170 and drive and control the first motor 121 to the fourth motor 124, a first encoder 151 to a fourth encoder 154 that detect the rotation speed of the first motor 121 to the fourth motor 124, and a first position detection unit 141 to a fourth position detection unit 144 that detect the position of each axis.

[0041] The control unit 170 has a command unit 170a that sends command values ​​to the first servo amplifier 161 to the fourth servo amplifier 164 corresponding to each axis, and a calculation unit 170b that calculates the command values ​​from the detection values ​​of the first position detection unit 141 to the fourth position detection unit 144. The operation panel 190, the control unit 170, or the memory unit 180 may be installed separately from the electric press machine 100 using a personal computer, a mobile terminal, or the like.

[0042] Next, a description will be given of the position control of the electric press machine 100 of the first example. Note that the description will be given here for a case where the electric press machine 100 has four axes.

[0043] The electric press machine 100 of the first example automatically and repeatedly performs the operation of pressing a workpiece during the actual press working period in which the molded product is actually produced. The slider 111 can be set to a horizontal position or a predetermined inclined position with high precision at each stage of the press working operation during the actual press working period. The electric press machine 100 of this embodiment preferably has a teaching period prior to the actual press working period.

[0044] FIG. 7 shows an example of the system configuration of the machine learning device 4 of the information processing system 1 of this embodiment.

[0045] The machine learning device 4 includes a learning data acquisition unit 41, a learning data storage unit 42, a machine learning unit 43, and a trained model storage unit 44. The machine learning device 4 is configured, for example, by a computer or the like. In this case, the learning data acquisition unit 41 is configured by a communication interface or an input / output device or the like, the machine learning unit 43 is configured by a processor or the like, and the learning data storage unit 42 and the trained model storage unit 44 are configured by storage or the like.

[0046] The learning data acquisition unit 41 is an interface unit connected to various external devices via a network 7 or the like, and acquires learning data including at least input information 10 and maintenance information 13. The external devices are, for example, an input unit provided in a simulated test device or the like, a physical quantity measuring device 2, a database device 3, an information processing device 5, or a user terminal device 6 used by a user.

[0047] The learning data storage unit 42 is a database that stores one or more sets of learning data acquired by the learning data acquisition unit 41. The specific configuration of the database that constitutes the learning data storage unit 42 may be designed as appropriate.

[0048] The machine learning unit 43 performs machine learning using the learning data stored in the learning data storage unit 42. That is, the machine learning unit 43 inputs multiple sets of learning data to the learning model 44a, and causes the learning model 44a to learn the correlation between the input information 10 contained in the learning data and the maintenance information 13, thereby generating a trained learning model 44a. As a specific method of learning by the machine learning unit 43 of this embodiment, a case where a neural network is adopted will be described.

[0049] The trained model storage unit 44 is a database that stores the trained learning model 44a generated by the machine learning unit 43. The trained learning model 44a stored in the trained model storage unit 44 is provided to the real system via any communication network, recording medium, etc. Note that although the training data storage unit 42 and the trained model storage unit 44 are shown as separate storage units in FIG. 7, they may also be configured as a single storage unit.

[0050] When "supervised learning" is adopted as machine learning, the learning data includes, as input data, input information 10 having state information 11 and characteristic information 12. The learning data also includes maintenance information 13 as output data associated with the input data. Therefore, the learning data is configured by associating input data including the input information 10 with output data including the maintenance information 13. In supervised learning, the output data is referred to as, for example, teacher data or correct answer labels.

[0051] Here, the correlation between the input information 10 and the maintenance information 13 contained in the learning data will be described.

[0052] Deterioration of the die 105 or lubricant is determined by at least one of the state of the electric press machine 100, the state of the die 105, the state of the workpiece, or the state of the lubricant, as well as the characteristics of the die 105, the characteristics of the workpiece, and the characteristics of the lubricant. However, it is difficult to directly measure the deterioration of the die 105 or the lubricant. Therefore, the physical quantity measuring device 2 measures at least one of the state of the electric press machine 100, the state of the die 105, the state of the workpiece, or the state of the lubricant as input data for learning, thereby making it possible to infer the deterioration state.

[0053] When acquiring the above-mentioned learning data, the learning data acquiring unit 41 also uses, as input data, characteristic information 12 including at least one of the characteristics of the die 105, the characteristics of the workpiece, and the characteristics of the lubricant stored in the database device 3. The characteristic information 12 may be input from the user terminal device 6.

[0054] When input information 10 is input as input data for learning data from at least one of the physical quantity measuring device 2, the database device 3, the information processing device 5, or the user terminal device 6, the learning data acquisition unit 41 acquires maintenance information 13 based on the input information 10. The maintenance information 13 is input from the database device 3, the information processing device 5, or the user terminal device 6 by the user determining the state of the mold 105 and the lubricant, and is used as output data (teaching data). Then, the learning data acquisition unit 41 configures one learning data by associating the input data with the output data, and stores the same in the learning data storage unit 42.

[0055] FIG. 8 shows an example of a neural network model used in the machine learning device 4 according to this embodiment.

[0056] The learning model 44a is configured as a neural network model as shown in Fig. 8. The neural network model is configured from l neurons (x1 to x1) in the input layer, m neurons (y11 to y1m) in the first hidden layer, n neurons (y21 to y2n) in the second hidden layer, and o neurons (z1 to zo) in the output layer.

[0057] Each neuron in the input layer is associated with a respective piece of input data included in the training data. Each neuron in the output layer is associated with a respective piece of output data included in the training data. Note that the input data may be subjected to predetermined pre-processing before being input to the input layer, and the output data may be subjected to predetermined post-processing after being output from the output layer.

[0058] The first and second hidden layers are also called hidden layers, and the neural network may have multiple hidden layers in addition to the first and second hidden layers, or may have only the first hidden layer as a hidden layer. Furthermore, synapses connecting the neurons of each layer are established between the input layer and the first hidden layer, between the first hidden layer and the second hidden layer, and between the second hidden layer and the output layer, and each synapse is assigned a weight wi (i is a natural number).

[0059] A neural network model uses training data to input input data contained in the training data into the input layer, and compares the output data output from the output layer as the inference result with the output data (teacher data) contained in the training data, thereby learning the correlation between the input data and the output data.

[0060] Specifically, each neuron in the input layer receives input data included in the training data, and the value of each neuron in the output layer is calculated by performing a process for all neurons other than the input layer, in which the value of the neuron on the input side connected to the neuron in question is calculated as the sum of a series of multiplication values ​​of the value of the neuron on the input side connected to the neuron in question and the weight wi associated with the synapse connecting the neuron on the output side and the neuron on the input side.

[0061] Then, the values ​​(z1 to zo) output to each neuron in the output layer as the inference results are compared with the values ​​(t1 to to) of the teacher data corresponding to each output data included in the learning data to determine the error, and a process (back propagation) is performed to adjust the weight wi associated with each synapse so that the error becomes small.

[0062] When a predetermined learning termination condition is met, such as repeating the above series of steps a predetermined number of times or the above error becoming smaller than an allowable value, the machine learning is terminated and a trained neural network model (all weights wi associated with each synapse) is generated.

[0063] FIG. 9 is a flowchart showing an example of a machine learning method performed by the machine learning device 4 according to this embodiment.

[0064] First, in step 21, the learning data acquisition unit 41 prepares a desired number of pieces of learning data as a preliminary preparation for starting machine learning, and stores the prepared learning data in the learning data storage unit 42 (ST21). The number of pieces of learning data to be prepared here may be set in consideration of the inference accuracy required for the ultimately obtained learning model 44a.

[0065] Several methods can be used to prepare training data. For example, when inferring deterioration of the die 105 or lubricant in a specific electric press machine 100 or test device, the deterioration of the die 105 or lubricant is acquired using the training data acquisition unit 41, and the user uses the user terminal device 6 to input the results in association with these measurement values, thereby preparing input data and output data that constitute the training data. Then, by repeating this process, it is possible to prepare multiple sets of training data.

[0066] Next, in step 22, the machine learning unit 43 prepares a pre-learning learning model 44a to start machine learning (ST22). The pre-learning learning model 44a prepared here is configured with the neural network model exemplified in FIG. 19, and the weight of each synapse is set to an initial value. Each neuron in the input layer is associated with a respective piece of input information 10 as input data included in the learning data. Each neuron in the output layer is associated with maintenance information 13.

[0067] Next, in step 23, the machine learning unit 43 acquires, for example, one piece of learning data at random from the plurality of sets of learning data stored in the learning data storage unit 42 (ST23).

[0068] Next, in step 24, the machine learning unit 43 inputs input data included in one piece of learning data to the input layer of the prepared learning model 44a before learning (or during learning) (ST24). As a result, output data is output as an inference result from the output layer of the learning model 44a, and this output data was generated by the learning model 44a before learning (or during learning). Therefore, in the state before learning (or during learning), the output data output as an inference result represents information different from the output data (teacher data) included in the learning data.

[0069] Next, in step 25, the machine learning unit 43 performs machine learning by comparing the output data (teacher data) included in the one learning data acquired in step 22 with the output data output from the output layer as an inference result in step 23 and adjusting the weight of each synapse (ST25). In this way, the machine learning unit 43 causes the learning model 44a to learn the correlation between the input data and the output data.

[0070] Next, in step 26, the machine learning unit 43 determines whether or not it is necessary to continue machine learning based on, for example, the error between the output data and the teacher data and the remaining number of unlearned learning data stored in the learning data storage unit 42 (ST26).

[0071] If the machine learning unit 43 determines in step 26 to continue machine learning (No in step 26), the process returns to step 23 and performs steps 23 to 25 on the learning model 44a being trained multiple times using untrained training data.On the other hand, if the machine learning unit 43 determines in step 26 to end machine learning (Yes in step 26), the process proceeds to step 27.

[0072] Then, in step 27, the machine learning unit 43 stores the trained learning model 44a generated by adjusting the weights associated with each synapse in the trained model storage unit 44 (ST27), thereby completing the series of machine learning methods shown in Fig. 20. In the machine learning method, step 21 corresponds to a learning data storage step, steps 22 to 26 correspond to a machine learning step, and step 27 corresponds to a trained model storage step.

[0073] As described above, the machine learning device 4 and machine learning method according to this embodiment can be used for various types of press machines, and can generate a learning model 44a for inferring maintenance information 13 based on input information 10 indicating the state and characteristics of the press machine as input data included in the learning data. Furthermore, the machine learning device 4 and machine learning method according to this embodiment can infer the deteriorated location, deterioration state, or appropriate maintenance timing of the die 105 or lubricant.

[0074] FIG. 10 shows an example of the system configuration of the information processing device 5 of this embodiment.

[0075] The information processing device 5 of this embodiment includes a processing unit 51 and a trained model storage unit 52. The processing unit 51 includes an input information acquisition unit 51a and a maintenance information output unit 51b. The information processing device 5 is configured, for example, by a computer. In this case, the input information acquisition unit 51a is configured by a communication interface or an input / output device, the maintenance information output unit 51b is configured by a processor, and the trained model storage unit 52 is configured by storage. The information processing device 5 may be incorporated into the die 105 or the electric press machine 100, or may be incorporated into a higher-level management device of the electric press machine 100 (for example, an equipment controller, an equipment management system that manages multiple pieces of equipment, etc.).

[0076] The input information acquiring unit 51a is an interface unit that is connected to the physical quantity measuring device 2, the database device 3, the machine learning device 4, and the user terminal device 6, and acquires the input information 10.

[0077] The maintenance information output unit 51b inputs the input information 10 acquired by the input information acquisition unit 51a into the learning model 44a and performs inference processing to infer the maintenance information 13. For the inference processing, the machine learning device 4 and the trained learning model 44a that has undergone supervised learning using the machine learning method are used.

[0078] The maintenance information output unit 51b not only has the function of performing inference processing using the learning model 44a, but also has a preprocessing function of adjusting the input information 10 acquired by the input information acquisition unit 51a into a desired format, etc., and inputting it to the learning model 44a as preprocessing of the inference processing, and a postprocessing function of applying a predetermined logical formula or calculation formula to the values ​​of the output data output from the learning model 44a, thereby finally inferring the maintenance information 13. Note that the inference results of the maintenance information output unit 51b are preferably stored in the trained model storage unit 52 or another storage device (not shown), and past inference results can be used, for example, as learning data to be used for online learning or re-learning in order to further improve the inference accuracy of the learning model 44a.

[0079] The trained model storage unit 52 is a database that stores trained learning models 44a used in the inference process of the maintenance information output unit 51b. The number of learning models 44a stored in the trained model storage unit 52 is not limited to one. For example, multiple learning models 44a with different numbers of input data or different machine learning methods may be stored and selectively available.

[0080] FIG. 11 shows an example of a flowchart of the maintenance information inference method of this embodiment.

[0081] First, in step 31, the input information acquisition unit 51a acquires the input information 10 (ST31).

[0082] Next, in step 32, the maintenance information output unit 51b performs preprocessing on the input data and inputs the preprocessed data to the input layer of the learning model 44a, and acquires the output data output from the output layer of the learning model 44a (ST32).

[0083] Next, in step 33, the maintenance information output unit 51b infers the maintenance information 13 as an inference step (ST33).

[0084] As described above, the information processing device 5 and information processing method according to this embodiment can be used for various types of press machines, and can infer maintenance information 13 based on input information 10 indicating the state and characteristics of the press machine as input data included in the learning data. Furthermore, the information processing device 5 and information processing method according to this embodiment can infer the deteriorated location, deterioration state, or appropriate maintenance timing of the die 105 or lubricant.

[0085] The present invention is not limited to the above-described embodiment, and various modifications can be made without departing from the spirit and scope of the present invention, all of which are included in the technical concept of the present invention.

[0086] In the above embodiment, a case has been described in which a neural network is employed as a specific method of machine learning by the machine learning unit 43, but any other machine learning method may be employed by the machine learning unit 43. Examples of other machine learning methods include tree-type methods such as decision trees and regression trees, ensemble learning such as bagging and boosting, neural network-type methods (including deep learning) such as recurrent neural networks and convolutional neural networks, clustering-type methods such as hierarchical clustering, non-hierarchical clustering, k-nearest neighbors, and k-means, multivariate analysis such as principal component analysis, factor analysis, and logistic regression, and support vector machines.

[0087] The present invention can also be provided in the form of a program (machine learning program) for causing a general-purpose computer to execute each step of the machine learning method according to the above embodiment. Also, the present invention can also be provided in the form of a program (position correction information inference program) for causing a general-purpose computer to execute each step of the position correction information inference method according to the above embodiment.

[0088] The present invention can be provided not only in the form of the information processing device 5 (maintenance information inference method or maintenance information inference program) according to the above embodiment, but also in the form of an inference device (inference method or inference program) used to infer maintenance information 13. In this case, the inference device (inference method or inference program) can include a memory and a processor, and the processor executes a series of processes. The series of processes includes an input information acquisition process (input information acquisition step) that acquires input information 10, and a maintenance information output process (maintenance information output step) that infers maintenance information 13 once the input information 10 is acquired in the input information acquisition process.

[0089] By providing it in the form of an inference device (inference method or inference program), it can be more easily applied to various devices than when it is implemented in the information processing device 5. It is naturally understandable to those skilled in the art that when the inference device (inference method or inference program) infers the maintenance information 13, it may apply an inference method implemented by the maintenance information output unit 51b of the information processing device 5 using the machine learning device 4 and trained learning model 44a generated by the machine learning method according to the above embodiment. [Explanation of symbols]

[0090] 1. Information processing system 10...Input information, 11...Status information, 11a...Main body status information, 11b...Mold status information, 11c...Workpiece status information, 11d...Lubricant status information, 12...Characteristic information, 12a...Mold property information, 12b...Workpiece material property information, 12c...Lubricant property information, 13...Maintenance information, 2...Physical quantity measurement device, 3...Database device, 4...machine learning device, 41...learning data acquisition unit, 42...learning data storage unit, 43...machine learning unit, 44...trained model storage unit, 44a...learning model 5...information processing device, 51...processing unit, 51a...input information acquisition unit, 51b...maintenance information output unit, 52...trained model storage unit

Claims

1. A machine learning device that generates a learning model for inferring information regarding maintenance of an electric press machine, a learning data storage unit that stores a plurality of sets of learning data in which the input information and the maintenance information are associated with each other, the learning data storage unit including input information including status information indicating a status of a predetermined portion of the electric press working machine and characteristic information indicating characteristics of the predetermined portion of the electric press working machine, and maintenance information related to maintenance of the electric press working machine; a machine learning unit that inputs a plurality of sets of the learning data into the learning model, thereby causing the learning model to learn a correlation between the input information and the maintenance information; a trained model storage unit that stores the trained model trained by the machine learning unit, Machine learning device.

2. The status information is Body status information indicating the status of the body of the electric press processing machine; die status information indicating the status of a die used in the electric press machine; workpiece state information indicating the state of the workpiece to be processed by the electric press; Or, Lubricant status information indicating the status of the lubricant used in the electric press machine; At least one of Contains The machine learning device according to claim 1 .

3. The main body status information is a slider load indicating a load applied to at least a portion of a slider of the main body of the electric press machine during processing; a slider position indicating a position of at least a portion of the slider during processing; Or, the number of shots of the slider after replacing the mold; At least one of Contains The machine learning device according to claim 2 .

4. The mold state information is a mold load indicating a load applied to at least a portion of the mold during processing; a mold temperature indicating the temperature of at least a portion of the mold during processing; or Mold displacement, which indicates the change in the dimensions of the mold before and after processing; At least one of Contains The machine learning device according to claim 2 .

5. The workpiece state information is post-processing dimensions indicating dimensions of at least a portion of the workpiece after processing; Surface roughness of at least a portion of the workpiece after processing; a workpiece temperature indicating the temperature of at least a portion of the workpiece during processing; Or, Workpiece displacement, which indicates the change in dimensions of the workpiece before and after processing; At least one of Contains The machine learning device according to claim 2 .

6. The lubricant condition information is a lubricant viscosity, which indicates the viscosity of at least a portion of the lubricant during processing; a lubricant temperature indicating the temperature of at least a portion of the lubricant during processing; Or, Ambient humidity, which indicates the humidity of at least a portion of the surroundings of the lubricant during processing; At least one of Contains The machine learning device according to claim 2 .

7. The characteristic information is Die characteristic information indicating characteristics of a die used in the electric press machine; Workpiece characteristic information indicating the characteristics of the workpiece to be processed by the electric press machine; Or, Lubricant characteristic information indicating the characteristics of a lubricant used in the electric press machine; At least one of Contains The machine learning device according to claim 1 .

8. The mold characteristic information is a mold material name indicating the name of the material of the mold; a mold load characteristic indicating deformation of the mold relative to a load; or a mold temperature characteristic showing deformation of the mold relative to temperature; At least one of Contains The machine learning device according to claim 7 .

9. The workpiece characteristic information is a workpiece material name indicating the material name of the workpiece; a workpiece load characteristic indicating deformation of the workpiece relative to a load; or a workpiece temperature characteristic showing deformation of the workpiece with respect to temperature; At least one of Contains The machine learning device according to claim 7 .

10. The lubricant characteristic information is a lubricant material name indicating the material name of the lubricant; lubricant temperature characteristics indicating the viscosity, expansion rate or contraction rate of the lubricant with respect to temperature; Or, lubricant humidity characteristics indicating the viscosity, expansion rate, or contraction rate of the lubricant with respect to humidity; At least one of Contains The machine learning device according to claim 7 .

11. The maintenance information includes: a deteriorated portion indicating a deteriorated portion or a portion approaching deterioration of the electric press machine; a deterioration state indicating a deterioration state including wear of the electric press processing machine; Or, When maintenance of the electric press processing machine is required, At least one of Contains The machine learning device according to claim 1 .

12. 12. An information processing device that infers the maintenance information using a learning model generated by the machine learning device according to claim 1, an input information acquisition unit that acquires input information including the state information and the characteristic information; an inference unit that inputs the input information acquired by the input information acquisition unit into the learning model and infers the maintenance information, Information processing device.

13. An inference device used to infer maintenance information related to maintenance of an electric press machine, the inference device comprises a memory and a processor; The processor performs an input information acquisition process to acquire input information including status information indicating a status of a predetermined part of the electric press machine and characteristic information indicating a characteristic of the predetermined part of the electric press machine; When the input information is acquired in the input information acquisition process, an inference process is executed to infer the maintenance information. Reasoning device.

14. A machine learning method for generating a learning model for inferring maintenance information related to maintenance of an electric press machine, comprising: a learning data storage step of storing a plurality of sets of learning data in a learning data storage unit, the learning data including input information including status information indicating a status of a predetermined portion of the electric press machine and characteristic information indicating characteristics of the predetermined portion of the electric press machine, and maintenance information relating to maintenance of the electric press machine, the input information and the maintenance information being associated with each other; a machine learning process in which a correlation between the input information and the maintenance information is learned by the learning model using a machine learning unit by inputting a plurality of sets of the learning data into the learning model; and a trained model storage step of storing the trained model trained by the machine learning unit in a trained model storage unit. Machine learning methods.

15. An information processing method for inferring maintenance information related to maintenance of an electric press machine using the learning model generated by the machine learning method according to claim 14, comprising: an input information acquisition step of acquiring input information including status information indicating a status of a predetermined portion of the electric press working machine and characteristic information indicating a characteristic of the predetermined portion of the electric press working machine; an inference step of inputting the input information acquired by the input information acquisition step into the learning model and inferring the maintenance information, Information processing methods.

16. An inference method used to infer maintenance information related to maintenance of an electric press machine, comprising: an input information acquisition step of acquiring input information including status information indicating a status of a predetermined portion of the electric press working machine and characteristic information indicating a characteristic of the predetermined portion of the electric press working machine; When the input information is acquired in the input information acquisition step, an inference step of inferring the maintenance information is executed. Reasoning method.

Images