STATE DETECTION DEVICE AND STATE DETECTION PROCEDURE

Abstract

State determination device (1) for determining an operating state of an industrial machine, wherein the state determination device (1) comprises: a data acquisition unit (30) configured to acquire data relating to the industrial machine; a training data extraction unit (32) configured to generate, on the basis of the data relating to the industrial machine obtained by the data acquisition unit (30), several partial time series data elements obtained by shifting time series data relating to physical quantities from the data relating to the industrial machine in the direction of a time axis, and to extract several training data containing the several partial time series data elements, and a learning unit (110) which is trained to perform machine learning using the learning data extracted by the learning data extraction unit (32), thereby generating a learning model.

Description

Field of invention

[0001] The present invention relates to a condition determination device and a condition determination method, and more precisely relates to a condition determination device and a condition determination method for supporting the maintenance of industrial machines. Description of the state of the art

[0002] Maintenance of an industrial machine, such as an injection molding machine, is performed regularly or when an anomaly occurs. During maintenance, the maintenance personnel determine the anomaly in the machine's operating condition by using physical parameters that indicate the machine's condition and are recorded during operation. They then perform maintenance tasks such as replacing abnormal components. An industrial machine can be any type of machine, including injection molding machines, machine tools, mining machines, woodworking machines, agricultural machines, construction machines, and the like.

[0003] For example, there is a known procedure for the maintenance of a check valve in the injection cylinder of an injection molding machine, a type of industrial machine, in which a screw is regularly removed from the injection cylinder so that the dimensions of the check valve can be measured directly. However, this procedure requires production to be stopped for this measurement, inevitably reducing productivity.

[0004] To solve this problem, a well-known anomaly diagnosis method exists. This method diagnoses an anomaly by indirectly detecting the degree of wear on the injection cylinder's check valve, without stopping production to remove the screw from the injection cylinder or similar measures. Furthermore, this diagnostic method identifies the anomaly by detecting torque on the screw or the presence of resin backflow towards the screw.

[0005] For example, JP H01-168421 A discloses a method in which a torque acting on a screw is measured and an anomaly is identified if the measured value exceeds a tolerance range. Furthermore, JP 2017-030221 A and JP 2017-202632 A disclose methods in which an anomaly is diagnosed by supervised learning of a drive component load, resin pressure, and the like. In addition, JP 2018-097616 A discloses a learning method in which machine learning is performed on multiple time-series data elements and feature vector clustering is carried out.

[0006] However, in industrial machines such as injection molding machines, whose drive unit has components with different specifications, there are various devices that make up the machine in question, and elements that are processed within the machine. Therefore, the problem arises that the difference between measured values ​​obtained from the machine and the numerical values ​​of training data input during machine learning is so large that a diagnosis by machine learning cannot be performed correctly.For example, if the type of equipment of the moving part that forms the injection molding machine, the type of resin as raw material of molded parts produced by the injection molding machine, or the types of mold, mold temperature control, resin dryer, and the like as auxiliary equipment of the injection molding machine differ from the learning conditions during the machine learning model generation, the measured values ​​obtained from the machine will deviate from the measured values ​​used during the machine learning model generation, so that the state determination for the anomaly by machine learning may sometimes not be carried out correctly.

[0007] To increase the diagnostic accuracy of machine learning, one approach is to prepare a wide variety of learning conditions when generating the machine learning model. However, machine learning based on a wide range of injection molding machines, resins, and auxiliary equipment is costly. Furthermore, operating the machine requires the preparation of raw materials such as resins and workpieces, and the costs of the raw materials needed to acquire the training data are also high. Moreover, the process of acquiring the training data is time-consuming. Consequently, the problem arises that the training data cannot be collected efficiently.

[0008] Document WO 2020 / 136836A1 relates to a fault diagnosis device and procedure that extracts a group of data records from a large number of acquired data records to generate training data for fault diagnosis in a machine system. The group of data records is obtained by sorting, i.e., rearranging them according to their temporal proximity to a reference point at which the machine system is diagnosed or delivered.

[0009] DE 101 19 853 A1 relates to a neural network and a method for determining properties with regard to the manufacture of an injection molded part.

[0010] CN 1 05 751 470 A relates to a real-time temperature control method for an injection molding machine.

[0011] The invention is defined by the main claim and the dependent claim. Further embodiments of the invention are described by the dependent claims. BRIEF SUMMARY OF THE INVENTION

[0012] Therefore, there is a need for a condition determination device and a condition determination procedure that are able to efficiently perform machine learning based on measured values ​​obtained from an industrial machine without incurring high costs, and to support the maintenance of various industrial machines using the result of the learning.

[0013] Now, a state determination device and a method according to the present invention solve the above problems by generating multiple learning data elements in units of a certain number of data points or of time in the direction of a time axis by shifting (or relocating) time series data (current, velocity, etc.) obtained from an industrial machine, and by performing machine learning on multiple learning data elements generated from individual time series data, thereby introducing a general-purpose learning model that is free from excessive learning during machine learning and performing a highly accurate estimation of the operating state and the degree of anomaly.

[0014] A state determination device according to one embodiment of the present invention is configured to determine an operating state of an industrial machine and comprises a data acquisition unit configured to acquire data relating to the industrial machine, a learning data extraction unit configured to generate, on the basis of the data relating to the industrial machine acquired by the data acquisition unit, several partial time series data elements obtained by shifting time series data relating to physical quantities from the data relating to the industrial machine in the direction of a time axis, and extracts several data elements for learning containing the several partial time series data, and a learning unit configured to perform machine learning using the learning data extracted by the learning data extraction unit.and thereby creates a learning model.

[0015] The state determination device may further include an estimation unit configured to make an estimate of the operating state of the industrial machine using the learning model generated by the learning unit.

[0016] The state determination device may further include an extraction condition storage unit configured to store conditions for the training data extraction unit for extracting the multiple training data elements containing the multiple partial time series data obtained by shifting the time series data in the time axis direction, as the number of data elements in a range of a predetermined time duration or of time series data.

[0017] The industrial machine can be an injection molding machine, and the time series data obtained by the data acquisition unit can include at least one set of information elements for identifying a mold closing process, a mold clamping process, an injection process, a pressure holding process, a measuring process, a mold opening process, an ejection process, a cycle start and a cycle end as molding processes of the injection molding machine, and at least one set of information elements including the current, voltage, torque, position, speed, and acceleration of a motor for driving the injection molding machine, and a pressure, temperature, flow rate, and flow velocity related to the molding operation of the injection molding machine.

[0018] The learning unit can be supervised learning, unsupervised learning, and / or reinforcement learning.

[0019] The physical quantities of the time series data obtained by the data acquisition unit can be at least one of the physical quantities of several industrial machines connected via a wired / wireless network.

[0020] The estimating unit can estimate the degree of anomaly related to the operating condition of the industrial machine, and the condition monitoring device can display a warning message on a display device if the degree of anomaly estimated by the estimating unit has exceeded a predetermined threshold.

[0021] The estimating unit can estimate the degree of anomaly related to the operating condition of the industrial machine, and the condition monitoring device can display a warning icon on a display device when the degree of anomaly estimated by the estimating unit has exceeded a predetermined threshold.

[0022] The estimation unit can estimate the degree of anomaly related to the operating state of the industrial machine, and the state determination device can issue at least one of the following commands to the industrial machine: stopping operation, slowing down, and limiting the torque of a motor.

[0023] A motor to drive the industrial machine can be an electric motor, an oil-hydraulic cylinder, an oil-hydraulic motor or an air motor, and a transmission mechanism to drive the industrial machine can include a ball screw, gears, pulleys and / or a belt.

[0024] A state determination method according to another embodiment of the present invention serves to determine an operating state of an industrial machine and comprises a data acquisition step for obtaining data relating to the industrial machine, a training data extraction step for generating several partial time series data elements, which are obtained by shifting time series data relating to physical quantities from the data relating to the industrial machine in the direction of a time axis, based on the data relating to the industrial machine obtained in the data acquisition step, and for extracting several data elements for learning, which contain the several partial time series data elements, and a training step for performing machine learning using the training data extracted in the training data extraction step, and thereby generating a training model.

[0025] The state determination procedure may further include an estimation step to make an estimate of the operating state of the industrial machine using the learning model that was generated in the learning step.

[0026] The present invention, with the setup described above, can reduce the effort required to collect a wide variety of time series data by effectively using individual time series data points, thereby enabling efficient data collection for training. Furthermore, generating multiple training data elements for machine learning from individual time series data points can be expected to improve the accuracy of the machine learning process. BRIEF DESCRIPTION OF THE DRAWINGS Fig. Figure 1 is a schematic hardware diagram of a state determination device according to one embodiment; Fig. 2 is a schematic functional block diagram of the state determination device according to one embodiment; Fig. 3 is a diagram that illustrates the generation of data for learning by a learning data extraction unit; Fig. 4 is a diagram that illustrates an example of the processing to generate shifted time series data by the training data extraction unit; Fig. Figure 5 is a diagram that illustrates another example of the processing used to generate the shifted time series data by the training data extraction unit; and Fig. Figure 6 is a diagram showing an example of how an anomaly condition is displayed. DETAILED DESCRIPTION OF PREFERRED EXECUTION FORMS

[0027] Fig. Figure 1 is a schematic hardware layout diagram showing the main parts of a state determination device comprising a machine learning device according to one embodiment.

[0028] A state-determining device 1 of the present embodiment can, for example, be installed in a control unit for controlling industrial machines, or be implemented as a personal computer attached to the control unit for controlling industrial machines, as an administration device 3 connected to the control unit via a wired / wireless network, or as a computer such as an edge computer, a fog computer, or a cloud server. The following description provides an example of an embodiment of the state-determining device 1 of the present embodiment as a computer connected via the network to the control unit for controlling injection molding machines as industrial machines.

[0029] Although in each of the embodiments described below an injection molding machine will be described as an industrial machine, the industrial machines include, as possible objects for state determination, an injection molding machine, a machine tool, a robot, a mining machine, a woodworking machine, an agricultural machine, a construction machine, and the like.

[0030] A CPU 11 of the state determination device 1 according to the present embodiment is a processor for generally controlling the state determination device 1. The CPU 11 reads system programs stored in a ROM 12 via a bus 20 and controls the entire state determination device 1 according to these system programs. Temporary calculation data, various data entered by a worker via an input device 71, and the like are temporarily loaded into a RAM 13.

[0031] A non-volatile memory 14 consists, for example, of a battery-backed memory (not shown) or a solid-state drive, and its memory state can be retained even when the state-determining device 1 is switched off. The non-volatile memory 14 stores a setting area in which setting information regarding the operation of the state-determining device 1 is loaded, data entered by the input device 71, and static data (machine type, mass and material of a mold, resin type, etc.) received from the injection molding machine 2 via a network 7, time-series data regarding physical quantities (the temperature of a nozzle, the position, speed, acceleration, current, voltage, and torque of a motor for driving the nozzle, the temperature of the mold, the flow rate, flow velocity, and pressure of the resin, etc.).), which were detected during the forming operations of the injection molding machine 2, data read from other computers via external storage devices (not shown) or the network 7, and the like. The programs and the various data stored in the non-volatile memory 14 can be developed in RAM 13 during execution and use. In addition, the system programs, including a conventional analysis program for analyzing the various data, a program for controlling the exchange with a machine learning device 100 (described later), and the like, were pre-written in ROM 12.

[0032] The state-determining device 1 is connected to the wired / wireless network 7 via an interface 16. The network 7 is connected to at least one of the injection molding machines 2, the management device 3 for managing the manufacturing activity by the injection molding machine 2, and the like, and exchanges data with the state-determining device 1.

[0033] Each injection molding machine 2 is a machine designed to produce molded parts from a resin, such as plastic. The injection molding machine 2 melts the resin and injects it into the mold to create the part. The injection molding machine 2 has various components, including the nozzle, motor, transmission mechanism, reduction gear, and moving parts. The status of these components is detected by sensors or similar devices, and their operation is controlled by the control unit. For example, an electric motor, an oil-hydraulic cylinder, an oil-hydraulic motor, or an air motor can be used as the motor for the injection molding machine 2. Furthermore, a ball screw, gears, pulleys, a belt, and similar components can be used for the transmission mechanism of the injection molding machine 1.

[0034] Data read into memory, data obtained as a result of program execution, data output by the machine learning device (described later), and the like are output via an interface 17 and displayed on a display device 70. Furthermore, the input device 71, consisting of a keyboard, a pointing device, and the like, delivers commands, data, and the like to the CPU 11 via an interface 18 based on operator input.

[0035] An interface 21 serves to connect the state determination device 1 and the machine learning device 100. The machine learning device 100 has a processor 101, a ROM 102, a RAM 103, and non-volatile memory 104. The processor 101 is used to control the entire machine learning device 100. The ROM 102 stores the system programs and the like. The RAM 103 is used for temporary storage in each step of the processing related to machine learning. The non-volatile memory 104 is used to store learning models and the like. The machine learning device 100 can store various information elements (e.g.,Various data, such as the type of injection molding machine 2, the mass and material of the mold, and the type of resin, as well as time-series data regarding various physical quantities (such as the nozzle temperature, the position, speed, acceleration, current, voltage, and torque of the motor driving the nozzle, the mold temperature, and the resin flow rate, flow velocity, and pressure), which can be obtained by the state-determining device 1, are monitored via interface 21. Furthermore, the state-determining device 1 obtains the processing result output by the machine learning device 100, stores the obtained result, displays it, and transmits it to other devices via network 7 or the like.

[0036] Fig. Figure 2 is a schematic functional block diagram of the state determination device 1 and the machine learning device 100 according to one embodiment.

[0037] The state determination device 1 of the present embodiment has a structure that is required when the machine learning device 100 performs the learning process. Each of the functional blocks that are in Fig. The process shown in 2 is implemented when the CPU 11 of the state determination device 1 and the processor 101 of the machine learning device 100, which are in Fig. 1 are shown, execute their respective system programs and control the operation of the individual parts of the state determination device 1 and the machine learning device 100.

[0038] The state determination device 1 of the present embodiment comprises a data acquisition unit 30, a learning data extraction unit 32, a preprocessing unit 34, and the machine learning device 100. The machine learning device 100 comprises a learning unit 110 and an estimation unit 120. Furthermore, a data acquisition storage unit 50 and an extraction condition storage unit 52 are provided in the non-volatile memory 14 of the state determination device 1. The data acquisition storage unit 50 stores data acquired from external machines or the like. The extraction condition storage unit 52 stores conditions for extracting learning data from the acquired data. A learning model storage unit 130 is provided in the non-volatile memory 104 of the machine learning device 100. The learning model storage unit stores learning models built by machine learning through the learning unit 110.

[0039] The data acquisition unit 30 acquires various data inputs from the injection molding machine 2, the input device 71, and the like. For example, the data acquisition unit 30 acquires static data such as the type of injection molding machine 2, the mass and material of the mold, and the type of resin; time-series data regarding various physical quantities such as the nozzle temperature, the position, speed, acceleration, current, voltage, and torque of the motor driving the nozzle, the mold temperature, and the resin flow rate, flow velocity, and pressure; and various data such as information (a type of time-series data acquired in conjunction with time) for identifying a mold closing process, a mold clamping process, an injection process, a pressure holding process, a measuring process, a mold opening process, and an ejection process.The data acquisition unit 30 acquires the start and end of a cycle, as forming processes of the injection molding machine 2, and information regarding maintenance activities on the injection molding machine entered by the operator, and stores this data in the acquired data storage unit 50. When acquiring the time series data, the data acquisition unit 30 considers the time series data acquired within a predetermined time range (e.g., the range of forming processes of a cycle) based on changes in the signal data acquired from the injection molding machine 2 and other time series data as individual time series data and then stores them in the acquired data storage unit 50. The data acquisition unit 30 can be configured to acquire the data from the management device 3 or other computers via external storage devices (not shown) or the wired / wireless network 7.

[0040] During the machine learning stage, the learning unit 110 extracts data to be used for learning from the acquired data obtained by the data acquisition unit 30 (and stored in the acquired data storage unit 50), based on extraction conditions stored in the extraction condition storage unit 52. A time width Wd (e.g., a time equivalent to the range of the forming processes of a cycle) of individual time series data (sub-time series data) to be extracted, and a displacement magnitude Dt for shifting (or relocating) the time series data, were predefined in the extraction condition storage unit 52.The defined value of the displacement amplitude Dt can, for example, be a numerical value smaller than the time width Wd, or a time corresponding to the mold closing process, mold clamping process, injection process, pressure holding process, measuring process, mold opening process, and ejection process as forming processes of injection molding machine 2. The displacement amplitude Dt can be defined in time units or as the number of elements of acquired data.

[0041] As in Fig. As shown in Figure 3, the training data extraction unit 32 generates several time series data elements obtained by shifting the time series data contained in the respective acquired data stored in the acquired data storage unit 50, and extracts several elements from acquired data, each containing the multiple elements of the generated time series data, as training data. Here, the generation of the time series data obtained by shifting the time series data on a time axis means as shown in Figure 3. Fig. Figure 4 shows the generation of partial time series data, which are obtained by shifting the start time by the predetermined shift amount Dt to a time with the time width Wd, for a series of target time series data.

[0042] The acquired data includes, for example, static data that does not change over time and time series data that record changes over time. The training data extraction unit 32 generates several elements of sub-time series data from the time series data, which are shifted along the time axis, and extracts several elements of acquired data obtained by combining these sub-time series data with the static data.

[0043] If acquisition data (FN-1, Re1, ECi), which as static data contain a model designation FN-1 and a resin type RE1, and as time series data contain a current ECi, are accepted as the object of data extraction for learning, and it has been determined that, under extraction conditions stored in the extraction condition memory unit 52, partial time series data, shifted by the displacement amplitude Dt at a time with time width Wd, are to be generated, the learning data extraction unit 32 generates partial time series data ECi1, ECi2, ... ECi n with the time width Wd, which is obtained by shifting the time series data ECi along the time axis by Dt, and extracts n elements from (FN-1, RE1, ECi1), (FN-1, RE1, ECi2), ... (FN-1, RE1, ECi n ), which are obtained by combining these n elements of partial time series data and the static data FN-1 and RE1 as data for learning.

[0044] If, as another example, acquisition data (FN-1, Re1, ECi, PR), which as static data contain a model designation FN-1 and a resin type RE1, and as time series data contain a current ECi and a pressure PR, are assumed to be the object of data extraction for learning, and it has been determined that, under the extraction conditions stored in the extraction condition storage unit 52, partial time series data, shifted at a time with time width Wd by the displacement extent Dt, are to be generated, the training data extraction unit 32 (1) generates the partial time series data ECi1, ECi2, ... ECi n With the time width Wd, which is obtained by shifting the time series data ECi on the time axis by Dt, it generates (2) partial time series data PR1 to PR nwith the time width Wd, which is obtained by shifting the time series data PR on the time axis by Dt, and extracts them (3) n elements from (FN-1, RE1, ECi1, PR1), (FN-1, RE1, ECi2, PR2), ... (FN-1, RE1, ECi n , PR n ), which are obtained by combining the n elements of partial time series data and the static data FN-1 and RE1, as data for learning.

[0045] Therefore, if the acquired data contains multiple time series elements, the data for learning is generated in such a way that the sub-time series data generated based on the respective time series data are combined in a single set with the time series data shifted by the same amount. The reason for this is that it is important to learn the changes in each time series data point simultaneously when multiple time series elements are present.

[0046] As in Fig. As shown in Figure 5, the extraction condition storage unit 52 can further contain an extraction start position St for the partial time series data generated from the time series data contained in the acquired data. For example, the extraction start position can be determined using a predetermined process during the operation of the injection molding machine 2 or a cycle start time, or it can be determined using the predetermined process or cycle start time plus a predetermined time width Dt. d be determined.

[0047] By defining the extraction start position St for the partial time series data together with the time width Wd and the displacement amplitude Dt of the partial time series data, several elements of partial time series data can be extracted from the time series data stored in the obtained data storage unit 50. These elements might contain, for example, a waveform in which a predetermined process (e.g., the injection process, in which the waveform of a current value in Fig. (4 vertically fluctuating) appears as learning data is extracted for learning.

[0048] During the machine learning stage, the machine learning device 100's preprocessing unit 34 generates learning data for use by the machine learning device 100, based on the learning data extracted by the learning data extraction unit 32. The preprocessing unit 100 generates learning data obtained by converting (or quantifying or sampling) the data input by the learning data extraction unit 32 into a uniform format for handling by the machine learning device 100. If the machine learning device 100 performs unsupervised learning, the preprocessing unit 34 generates, for example, state data S with a predefined format as learning data. If the machine learning device 100 performs supervised learning, the preprocessing unit 34 generates, as learning data, a set of state data S and label data L with a predefined format as learning data.When the machine learning device 100 performs reinforcement learning, the preprocessing unit 34 generates a set of state data S and determination data D with a predetermined format as learning data.

[0049] Furthermore, during the estimation stage by the machine learning device 100, the preprocessing unit 34 converts the acquired data obtained by the data acquisition unit 30 (and stored in the acquired data storage unit 50) into a uniform form for handling in the machine learning device 100 (or quantifies or samples it), thereby generating the state data S with a predetermined format, which are used for estimation by the machine learning device 100.

[0050] The learning unit 110 of the machine learning device 100 performs machine learning using the learning data generated for learning by the preprocessing unit 34 based on the data extracted by the learning data extraction unit 32. The learning unit 110 generates a learning model by performing machine learning using the data obtained from the injection molding machine 2, based on a conventional machine learning method such as unsupervised learning, supervised learning, or reinforcement learning, and stores the generated learning model in the learning model storage unit 130.The unsupervised learning method used in learning unit 110 can be represented, for example, by the autoencoder method or the k-means method, while the supervised learning method can be represented, for example, by the multilayer perceptron method, the recurrent neural network method, the long-short-time memory method, or the convolutional neural network method. The reinforcement learning method can be represented, for example, by the Q-learning method.

[0051] The learning unit 110 can perform unsupervised learning, for example, based on training data obtained when the data acquired from the injection molding machine 2 in a normal operating state is processed by the training data extraction unit 32 and the preprocessing unit 34, and generate the distribution of the data acquired in a normal state as a training model. Using training models generated in this way, the estimation unit 120 (described later) can estimate the extent of the deviation S of the state data obtained when the data acquired from the injection molding machine 2 is processed by the preprocessing unit 34 from the state data acquired during normal operation, and calculate an anomaly level as the result of the estimate.

[0052] Furthermore, the learning unit 110 can, for example, perform supervised learning using training data by processing the acquired data through the training data extraction unit 32 and the preprocessing unit 34 in such a way that the acquired data obtained from a normally functioning injection molding machine is assigned a label “normal” and the acquired data obtained from the injection molding machine 2 before and after the occurrence of an anomaly is assigned a label “abnormal”, thereby generating distinction boundaries between the normal and the abnormal data as training models.Using the learning models generated in this way, the estimation unit 120 (described later) can input the state data S obtained when the data obtained from the injection molding machine 2 are processed by the preprocessing unit 34 into the learning models, estimate whether the state data S belong to the normal data or the abnormal data, and calculate a label value (normal / abnormal) as a result of the estimation and the degree of its reliability.

[0053] Based on the state data S generated by the preprocessing unit 34, the estimating unit 120 of the machine learning device 100 estimates the state of the injection molding machine using the learning models stored in the learning model storage unit 130. In the estimating unit 120 of the present embodiment, by inputting the state data S obtained from the preprocessing unit 34 into the learning model generated by the learning unit 110 (or for which parameters were determined), the degree of anomaly related to the state of the injection molding machine is estimated and calculated, or the class (normal / abnormal, etc.) to which the operating state of the injection molding machine belongs is estimated and calculated. The result of the estimation by the estimating unit 120 (the degree of anomaly related to the state of the injection molding machine, the class to which the operating state of the injection molding machine belongs, etc.)) can be used to be output for display on the display unit 70 or to be transmitted via a wired / wireless network (not shown) to a host computer, a cloud computer, or the like. Furthermore, if the result of the estimation by the estimation unit 120 proves to be a predetermined condition (e.g., if the degree of anomaly estimated by the estimation unit 120 exceeds a predetermined threshold, or if the class to which the operating state of the injection molding machine estimated by the estimation unit 120 is "abnormal" is shown), a warning message and an icon can be output for display on the display unit 70, as shown in . Fig. 6 is shown, or a command to stop operation, slow down or limit the motor torque is issued to the injection molding machine.

[0054] In the state determination device with the above configuration, multiple data elements for learning are generated from individual acquired data points by the learning data extraction unit 32. This is achieved by shifting the time series data contained in the acquired data from the injection molding machine according to the extraction conditions stored in the extraction condition storage unit 52. In this way, a large number of learning data points can be generated from a predetermined number of acquired data points obtained from the limited operation of the injection molding machine 2.Therefore, the learning unit 110, which is included in the machine learning device 100, can effectively advance learning to support the maintenance of various industrial machines without incurring high costs, and generate learning models that can flexibly overcome waveform deviations in the time axis direction.

[0055] Although the state determination device 1, according to the present embodiment, is applicable to cases where states related to industrial machines such as robots and machine tools are to be determined, it can, for example, be appropriately applied to industrial machines that behave unstably when operation is started or when operating conditions are changed. In particular, the operation of injection molding machines can sometimes be delayed, even when operating under the same injection conditions, depending on the internal and external conditions. Even in such a case, the molding operation of the injection molding machine itself is normal, which is why data for learning such normal operation is required so that such data is not identified as abnormal.The state determination device 1 according to the present embodiment is particularly useful for state determination in an injection molding machine, since it can generate several data elements for learning by shifting the time series data from the data obtained conventionally, without having to specifically obtain data and the like for the case in which the machine operation is so delayed.

[0056] Although embodiments of the present invention have been described above, the invention is not limited to the embodiments described above, but can be appropriately modified and implemented in various forms.

[0057] Although the embodiments described above, for example, describe the state determination device 1 and the machine learning device 100 as devices comprising different CPUs (processors), the machine learning device 100 can alternatively be implemented by the CPU 11 of the state determination device 1 and the system programs stored in the ROM 12 of the state determination device 1. Furthermore, if several injection molding machines 2 are interconnected via the network, their respective operating states can be determined by a single state determination device 1, or the state determination device 1 can be installed in the control unit of the injection molding machine 1.

Claims

[1] State determination device (1) for determining an operating state of an industrial machine, wherein the state determination device (1) comprises: a data acquisition unit (30) configured to acquire data relating to the industrial machine; a training data extraction unit (32) configured to generate, on the basis of the data relating to the industrial machine obtained by the data acquisition unit (30), several partial time series data elements obtained by shifting time series data relating to physical quantities from the data relating to the industrial machine in the direction of a time axis, and to extract several training data containing the several partial time series data elements, and a learning unit (110) which is trained to perform machine learning using the learning data extracted by the learning data extraction unit (32), thereby generating a learning model. [2] State determination device (1) according to claim 1, further comprising an estimation unit (120) configured to make an estimate of the operating state of the industrial machine using the learning model generated by the learning unit (110). [3] State determination device (1) according to claim 1, further comprising an extraction condition storage unit (52) configured to store conditions for the training data extraction unit (32) for extracting the multiple training data elements containing the multiple partial time series data elements obtained by shifting the time series data in the time axis direction as the number of data elements in a range of a predetermined time duration or of time series data. [4] State determination device (1) according to claim 1 or 2, wherein the industrial machine is an injection molding machine (2), and the time series data obtained by the data acquisition unit (30) include at least one of information elements for identifying a mold closing process, a mold clamping process, an injection process, a pressure holding process, a measuring process, a mold opening process, an ejection process, a cycle start and a cycle end, as molding processes of the injection molding machine (2), and at least one of information elements including the current, voltage, torque, position, speed, and acceleration of a motor for driving the injection molding machine (2), and a pressure, temperature, flow rate, and flow velocity associated with the molding operation of the injection molding machine (2). [5] State determination device (1) according to claim 1, wherein the learning unit (110) is supervised learning, unsupervised learning, and / or reinforcement learning. [6] State determination device (1) according to claim 1 or 2, wherein the physical quantities of the time series data obtained by the data acquisition unit (30) are at least one of physical quantities of several industrial machines which are connected via a wired / wireless network. [7] State determination device (1) according to claim 2, wherein the estimating unit (120) estimates a degree of anomaly related to the operating state of the industrial machine, and the state determination device (1) displays a warning message on a display device when the degree of anomaly estimated by the estimating unit (120) has exceeded a predetermined threshold. [8] State determination device (1) according to claim 2, wherein the estimating unit (120) estimates a degree of anomaly related to the operating state of the industrial machine, and the state determination device (1) displays a warning icon on a display device when the degree of anomaly estimated by the estimating unit (120) has exceeded a predetermined threshold. [9] State determination device (1) according to claim 2, wherein the estimating unit (120) estimates a degree of anomaly related to the operating state of the industrial machine, and the state determination device (1) outputs at least one of commands to stop operation, to slow down, and to limit the torque of a motor to the industrial machine. [10] State determination device (1) according to claim 1 or 2, wherein a motor for driving the industrial machine is an electric motor, an oil-hydraulic cylinder, an oil-hydraulic motor or an air motor, and a transmission mechanism for driving the industrial machine comprises a ball screw, gears, discs and / or a belt. [11] State determination procedure for determining an operating state of an industrial machine, wherein the state determination procedure comprises: a data acquisition step to obtain data regarding the industrial machine; a training data extraction step for generating multiple sub-time series data elements, obtained by shifting time series data with respect to physical quantities from the data relating to the industrial machine in the direction of a time axis, based on the data relating to the industrial machine obtained in the data acquisition step, and for extracting multiple training data containing the multiple sub-time series data elements; and a learning step to perform machine learning using the learning data extracted in the learning data extraction step, thereby generating a learning model. [12] State determination method according to claim 11, further comprising an estimation step for making an estimate of the operating state of the industrial machine using the learning model generated in the learning step.

Images