Learning device, insulation diagnosis system, and program

Abstract

A learning device (1) according to the present disclosure uses operation information related to an operation history of a rotary electric machine to obtain a first trained model that infers non-destructive electrical characteristics related to dielectric strength in a main insulating layer of a stator coil of the rotary electric machine. The learning device (1) comprises: a learning data acquisition unit (11) that acquires learning data including the operation information and the non-destructive electrical characteristics obtained when the rotary electric machine is operated under conditions indicated by the operation information; and a model generation unit (12) that uses the learning data to generate the first trained model for inferring the non-destructive electrical characteristics from the operation information.

Description

Learning device, insulation diagnosis system, and program

[0001] The present disclosure relates to a learning device, an insulation diagnosis system, and a program.

[0002] Regarding the deterioration of the coil insulation of a rotating electrical machine or the like, due to various deteriorations such as electrical deterioration, thermal deterioration, mechanical deterioration, and environmental deterioration according to the operation history of the rotating electrical machine, a large number of small voids are generated in the entire insulation layer, and large voids may also be generated locally. When a high voltage is applied to the voids in this insulation layer, the void portions discharge and become conductors, resulting in a shortening of the insulation distance and a decrease in the breakdown voltage. In order to know such an insulation deterioration situation, non-destructive tests such as a dielectric loss tangent test, an alternating current test, and a partial discharge test of the insulator are generally carried out.

[0003] As a conventional method for estimating the deterioration of a coil insulator, using the operation history data and the data of the above non-destructive tests as inputs, converting this into reference data, obtaining a combined deterioration degree by fuzzy inference, estimating the insulation breakdown voltage remaining rate from the obtained combined deterioration degree, and learning the actual insulation breakdown voltage remaining rate obtained by a breakdown test to correct the insulation breakdown voltage remaining rate to be estimated, thereby estimating the remaining life is known.

[0004] Japanese Patent Laid-Open No. 5-264645

[0005] In the method for estimating the deterioration of an insulator disclosed in the above-mentioned Patent Document 1, since it is necessary to learn the actual insulation breakdown voltage remaining rate obtained by the breakdown voltage, for example, a breakdown test must be actually performed at the time of insulator replacement, and the actual insulation breakdown voltage remaining rate must be obtained from this result. For this reason, it is necessary to stop the rotating electrical machine during operation, and there is a problem that it takes time and effort to measure this insulation breakdown voltage remaining rate. The present disclosure has been made to solve the above problems, and an object thereof is to provide a learning device, an insulation diagnosis system, and a program that can diagnose the insulation performance of a rotating electrical machine while reducing the time and effort required to measure electrical characteristics of the rotating electrical machine such as the insulation breakdown voltage remaining rate.

[0006] The learning device relating to this disclosure is a learning device that obtains a first trained model for inferring non-destructive electrical characteristics related to dielectric strength in the main insulating layer of the stator coil of a rotating electric machine using operating information related to the operating history of the rotating electric machine, and comprises a learning data acquisition unit that acquires learning data including operating information and non-destructive electrical characteristics when the rotating electric machine is operated under the conditions indicated in the operating information, and a model generation unit that generates a first trained model for inferring non-destructive electrical characteristics from the operating information using the learning data.

[0007] The learning device described herein acquires training data including operating information and non-destructive electrical characteristics obtained when the machine is operated under the conditions indicated in the operating information. Using this training data, it generates a first trained model for inferring non-destructive electrical characteristics from the operating information. Therefore, it only needs to use operating information and data on non-destructive electrical characteristics that can be obtained even when the rotating electric machine is in operation. Consequently, compared to acquiring data on dielectric strength retention, the effort required to stop the rotating electric machine while it is in operation to acquire the data can be reduced.

[0008] This is a block diagram showing the configuration of the learning device according to Embodiment 1. This is a diagram showing an example of a neural network in the learning device according to Embodiment 1. This is a flowchart for explaining the processing of the learning device according to Embodiment 1. This is a block diagram showing the configuration of the insulation diagnostic system according to Embodiment 1. This is a flowchart for explaining the processing of the insulation diagnostic system according to Embodiment 1. This is a block diagram showing another configuration of the insulation diagnostic system in Embodiment 1. This is a block diagram showing the configuration of the learning device, which is an element constituting the learning device according to Embodiment 2. This is a flowchart for explaining the processing of the learning device shown in Figure 7. This is a block diagram showing the configuration of the insulation diagnostic system utilizing the trained model generated by the learning device shown in Figure 7. This is a flowchart for explaining the processing of the insulation diagnostic system shown in Figure 9. This is a block diagram showing another configuration of the insulation diagnostic system utilizing the trained model generated by the learning device shown in Figure 7. This is a block diagram showing the configuration of the learning device according to Embodiment 2. This is a flowchart for explaining the processing of the learning device according to Embodiment 2. This is a block diagram showing the configuration of the insulation diagnostic system according to Embodiment 2. This is a flowchart for explaining the processing of the insulation diagnostic system according to Embodiment 2. This is a diagram showing an example of the hardware of the learning device, trained model storage unit, inference unit, diagnostic unit and insulation diagnostic system.

[0009] The embodiments will be described in detail below with reference to the drawings. Note that the embodiments described below are illustrative examples. Furthermore, each embodiment can be combined as appropriate.

[0010] Embodiment 1. <Learning Phase> Figure 1 is a block diagram showing the configuration of the learning device 1 according to Embodiment 1. The learning device 1 obtains a first trained model for inferring the non-destructive electrical characteristics of the main insulating layer of the stator coil of a rotating electric machine in an insulation diagnostic system for a rotating electric machine. The learning device 1 comprises a learning data acquisition unit 11 and a model generation unit 12.

[0011] The learning data acquisition unit 11 acquires learning data, which is a combination of the first input data and the second input data. Here, the first input data is operation information relating to the operating history of the rotating electric machine. The operating information of the rotating electric machine may include, for example, at least one of the number of times the rotating electric machine has been started and stopped and the operating time of the rotating electric machine.

[0012] The second input data is a non-destructive electrical characteristic related to the dielectric strength of the main insulating layer of a stator coil in a rotating electric machine operated under the conditions indicated by the first input data. The non-destructive electrical characteristic of the main insulating layer may include at least one of the increase in dielectric loss tangent of the main insulating layer, the first current surge voltage of the main insulating layer, and the maximum partial discharge charge of the main insulating layer.

[0013] As insulation degradation progresses and the number of voids in the insulating layer increases, partial discharge occurs, causing an increase in the dielectric loss tangent, which serves as an indicator of insulation degradation. In AC current tests, when a voltage is applied to a good insulator, the voltage-current curve is linear. However, if the insulation absorbs moisture and the dielectric constant increases, or if partial discharge occurs, the rate of current increase increases sharply. When the AC current-voltage characteristics are taken, two points of sharp increases in the rate of current increase are observed, but here we will focus on the first voltage of sharp current increase. Furthermore, when measuring the relationship between the amount of partial discharge charge and the applied voltage, it is found that the amount of partial discharge charge increases as the applied voltage increases. However, the maximum amount of partial discharge charge at normal voltage (E / √3, E: rated voltage) represents the localized insulation degradation state. As described above, all non-destructive electrical characteristics used here are related to the dielectric strength of the main insulating layer, but other characteristics related to the dielectric strength of the main insulating layer may also be used.

[0014] The learning data acquisition unit 11 may acquire input data from a data input device or database (not shown). The data input device is, for example, a keyboard, mouse, keypad, or touch panel, and is operated by the user. The database is, for example, a storage device that stores design and operation information of a rotating electric machine.

[0015] The model generation unit 12 learns the non-destructive electrical characteristics of the main insulating layer of the stator coil in a rotating electric machine operated under the conditions indicated by the first input data, based on the training data, which is a combination of the first input data and the second input data output from the training data acquisition unit 11. That is, it generates a first trained model that infers the non-destructive electrical characteristics of the main insulating layer of the stator coil in a rotating electric machine operated under the conditions indicated by the first input data, from the training data, which is a combination of the first input data and the second input data. Here, the training data is data obtained by relating the first input data and the second input data to each other.

[0016] The learning algorithm used by the model generation unit 12 can be any known algorithm, such as supervised learning, unsupervised learning, reinforcement learning, or semi-supervised learning. As an example, the case in which a neural network is applied will be described. The model generation unit 12 learns the non-destructive electrical characteristics of the main insulating layer by so-called supervised learning, following a neural network model, for example. Here, supervised learning is a method in which a set of input and result (label) data is provided to the learning device 1, the features in that learning data are learned, and the result is inferred from the input.

[0017] A neural network consists of an input layer made up of multiple neurons, an intermediate layer (hidden layer) made up of multiple neurons, and an output layer made up of multiple neurons. The intermediate layer may be one layer or two or more layers. Figure 2 is a diagram showing an example of a neural network of a learning device 1 according to Embodiment 1. For example, in a three-layer neural network as shown in Figure 2, when multiple inputs are input to the input layer (X1 to X3), the values ​​are multiplied by weights w1 (w11 to w16) set according to the transmission path and input to the intermediate layer (Y1 to Y2), and the result is further multiplied by weights w2 (w21 to w26) set according to the transmission path and output from the output layer (Z1 to Z3). This output result changes depending on the values ​​of weights w1 and w2.

[0018] In the learning device 1 of Embodiment 1, the neural network learns the non-destructive electrical characteristics of the main insulating layer of the stator coil in a rotating electric machine operated under the operating conditions of the rotating electric machine indicated by the first input data, using so-called supervised learning, based on learning data created based on the combination of first input data and second input data acquired by the learning data acquisition unit 11. That is, the neural network learns by inputting the first input data, which is the operating information of the rotating electric machine, into the input layer and adjusting the weights w1 and w2 so that the result output from the output layer approaches the non-destructive electrical characteristics of the main insulating layer of the stator coil in a rotating electric machine operated under the conditions indicated by the first input data. The model generation unit 12 generates and outputs a first trained model by performing the learning described above. The trained model storage unit 2 stores the first trained model output from the model generation unit 12.

[0019] Next, the learning process performed by the learning device 1 will be described. Figure 3 is a flowchart illustrating the processes of the learning device 1 and the learned model storage unit 2 according to Embodiment 1. Step S01 is the learning data acquisition step, step S02 is the learning processing step, and step S03 is the learned model storage step.

[0020] In step S01, the training data acquisition unit 11 acquires training data, outputs it to the model generation unit 12, and proceeds to step S02. The training data acquisition unit 11 may acquire the first input data and the second input data included in the training data simultaneously, or it may acquire the first input data and the second input data at different times. In short, it is sufficient that the first input data and the second input data are acquired in association with each other.

[0021] In step S02, the model generation unit 12 uses training data created based on the combination of first and second input data acquired by the training data acquisition unit 11 to learn the non-destructive electrical characteristics of the main insulating layer of the stator coil in a rotating electric machine operated under the conditions indicated by the first input data, using so-called supervised learning. Furthermore, the model generation unit 12 generates a first trained model, outputs the generated first trained model to the trained model storage unit 2, and proceeds to step S03. In step S03, the trained model storage unit 2 stores the first trained model acquired from the model generation unit 12 and terminates processing.

[0022] <Application Phase> Next, the insulation diagnostic system for a rotating electric machine in Embodiment 1 will be described. Figure 4 is a block diagram showing the configuration of the insulation diagnostic system 100 for a rotating electric machine according to Embodiment 1. The insulation diagnostic system 100 for a rotating electric machine diagnoses the insulation performance of a rotating electric machine. The insulation diagnostic system 100 for a rotating electric machine comprises a trained model storage unit 2, an inference device 3, and a diagnostic unit 4. The inference device 3 consists of a data acquisition unit 31 and a first inference unit 32. The trained model storage unit 2, the inference device 3, and the diagnostic unit 4 may be configured on the same hardware, or they may be separate devices connected via a network. Furthermore, the trained model storage unit 2, the inference device 3, and the diagnostic unit 4 may reside on a cloud server.

[0023] The data acquisition unit 31 of the inference device 3 acquires operating information of a rotating electric machine as input data when diagnosing the insulation performance of the rotating electric machine. The operating information of the rotating electric machine may include, for example, at least one of the number of times the rotating electric machine has been started and stopped and the operating time of the rotating electric machine. The data acquisition unit 31 may acquire input data from a data input device or database (not shown). The data input device is, for example, a keyboard, mouse, keypad, or touch panel, and is operated by the user. The database is, for example, a storage device that stores the operating information of the rotating electric machine.

[0024] The first inference unit 32 uses the first trained model stored in the trained model storage unit 2 to infer the non-destructive electrical characteristics of the main insulating layer under the conditions indicated by the input data. That is, by inputting the input data acquired by the data acquisition unit 31 into this first trained model, the non-destructive electrical characteristics of the main insulating layer of the stator coil in a rotating electric machine operated under the conditions indicated by the input data are inferred. Specifically, at least one of the increase in dielectric loss tangent of the main insulating layer, the first current surge voltage of the main insulating layer, and the maximum partial discharge charge amount of the main insulating layer can be inferred, and the inference results can be output.

[0025] In this embodiment, the first inference unit 32 was described as outputting the non-destructive electrical characteristics of the main insulating layer using the first trained model learned by the model generation unit 12. However, the first trained model may be obtained from an external source such as another learning device, and the non-destructive electrical characteristics of the main insulating layer may be output based on this first trained model.

[0026] The diagnostic unit 4 receives at least one of the outputs from the first inference unit 32: the increase in the dielectric loss tangent of the main insulating layer, the first current surge voltage of the main insulating layer, and the maximum partial discharge charge amount of the main insulating layer. It determines whether the non-destructive electrical properties of the main insulating layer meet predetermined criteria and outputs the determination result. The method for determining the criteria values ​​depends on the type of insulating material, but for example, if the increase in the dielectric loss tangent of the main insulating layer exceeds 6.5%, it can be determined to be abnormal. The first current surge voltage and the maximum partial discharge charge amount can also be determined quantitatively in a similar manner.

[0027] Figure 5 is a flowchart illustrating the processing of the insulation diagnostic system 100 according to Embodiment 1. Step S11 is a data acquisition step, step S12 is an inference processing step, and step S13 is a diagnostic processing step.

[0028] In step S11, the data acquisition unit 31 acquires operating information of the rotating electric machine as input data when diagnosing the insulation performance of the rotating electric machine and outputs it to the first inference unit 32, proceeding to step S12. In step S12, the first inference unit 32 inputs the input data acquired from the data acquisition unit 31 into the first trained model stored in the trained model storage unit 2, acquires the non-destructive electrical characteristics of the main insulation layer of the stator coil in the rotating electric machine operated under the conditions indicated by the input data, outputs it to the diagnosis unit 4, and proceeds to step S13. In step S13, the diagnosis unit 4 determines whether the non-destructive electrical characteristics of the main insulation layer, which are the output data acquired from the first inference unit 32, meet predetermined criteria, outputs the determination result, and terminates the process. Non-destructive electrical characteristics can be acquired without stopping the operation of the rotating electric machine. Therefore, it is not necessary to stop the rotating electric machine even in the stage of acquiring non-destructive electrical characteristics and generating the first trained model in the learning phase. Accordingly, the insulation performance of the rotating electric machine can be diagnosed without stopping the rotating electric machine while it is in operation, including the learning phase and the utilization phase.

[0029] Figure 6 is a block diagram showing another configuration of the insulation diagnostic system 100 for a rotating electric machine in Embodiment 1. The insulation diagnostic system 100 for a rotating electric machine shown in Figure 6 includes a learning device 1, a learned model storage unit 2, an inference device 3, and a diagnostic unit 4. In the insulation diagnostic system 100 for a rotating electric machine shown in Figure 6, the operation of the learning device 1, the learned model storage unit 2, the inference device 3, and the diagnostic unit 4 is the same as that of the learning device 1, the learned model storage unit 2, the inference device 3, and the diagnostic unit 4 shown in Figures 1 and 4.

[0030] The learning device 1, the trained model storage unit 2, the inference device 3, and the diagnostic unit 4 may be configured on the same hardware, or they may be separate devices connected via a network. Alternatively, the learning device 1, the trained model storage unit 2, the inference device 3, and the diagnostic unit 4 may reside on a cloud server. The learning device 1 and the inference device 3 may be implemented by a single device; in this case, the training data acquisition unit 11 may also function as the data acquisition unit 31.

[0031] The model generation unit 12 may learn the non-destructive electrical characteristics of the main insulating layer according to the training data created for the insulation diagnostic systems of multiple rotating electric machines. The model generation unit 12 may acquire training data from the insulation diagnostic systems of multiple rotating electric machines used in the same area, or it may learn the non-destructive electrical characteristics of the main insulating layer using training data collected from the insulation diagnostic systems of multiple rotating electric machines operating independently in different areas.

[0032] Furthermore, it is possible to add or remove rotating electrical machine insulation diagnostic systems from the target midway through the process, as they collect training data. In addition, a learning device that has learned the non-destructive electrical properties of the main insulation layer for one rotating electrical machine insulation diagnostic system may be applied to another rotating electrical machine insulation diagnostic system, and the non-destructive electrical properties of the main insulation layer for that other rotating electrical machine insulation diagnostic system may be relearned and updated.

[0033] As described above, the learning device 1 according to this embodiment is a learning device 1 that obtains a first trained model for inferring non-destructive electrical characteristics related to dielectric strength in the main insulating layer of the stator coil of a rotating electric machine using operating information related to the operating history of the rotating electric machine, and comprises a learning data acquisition unit 11 that acquires learning data including operating information and non-destructive electrical characteristics when the rotating electric machine is operated under the conditions indicated in the operating information, and a model generation unit 12 that generates a first trained model for inferring non-destructive electrical characteristics from the operating information using the learning data.

[0034] The learning device 1 according to this embodiment acquires learning data including operating information and non-destructive electrical characteristics obtained when the machine is operated under the conditions indicated in the operating information. Using this learning data, it generates a first trained model for inferring non-destructive electrical characteristics from the operating information. Therefore, it only needs to use operating information and non-destructive electrical characteristic data that can be obtained even when the rotating electric machine is in operation. Accordingly, compared to the case where dielectric strength remaining rate data was acquired, the effort of having to stop the rotating electric machine in operation to acquire data can be reduced.

[0035] Furthermore, the learning data acquisition unit 11 may acquire the learning data which includes not only operating information and non-destructive electrical characteristics obtained when the machine is operated under the conditions indicated in the operating information, but also design information related to the dielectric strength of the main insulating layer of the stator coil. The model generation unit 12 may then use the learning data to generate a third trained model for inferring the non-destructive electrical characteristics from the operating information and the design information. Here, the design information related to the dielectric strength of the main insulating layer of the stator coil may include, for example, at least one of the rated voltage of the rotating electric machine and the thickness of the main insulating layer.

[0036] Furthermore, the insulation diagnostic system 100 may include a data acquisition unit 31 that acquires operating information and design information, a first inference unit 32 that outputs non-destructive electrical characteristics obtained when operating information and design information are input using a third trained model, and a diagnostic unit 4 that determines whether or not the non-destructive electrical characteristics meet predetermined criteria and outputs the result of the determination.

[0037] By adding design information to the training data in this way, the same insulation diagnostic system can be applied to multiple models with different design conditions, thereby reducing training costs. Another advantage is that the insulation diagnostic system, which has been trained using previously acquired training data, can be reused for new models.

[0038] Embodiment 2. <Description of the learning device 1a and insulation diagnostic system 100a, which are components of the learning device and insulation diagnostic system> In describing the learning device and insulation diagnostic system according to Embodiment 2, we will first describe the configuration of the learning device 1a and insulation diagnostic system 100a, which are elements that constitute this learning device and insulation diagnostic system. Figure 7 is a block diagram showing the configuration of the learning device 1a, which is one of the elements that constitute the learning device according to Embodiment 2. The learning device 1a obtains a trained model for inferring the dielectric strength remaining rate of the main insulation layer of the stator coil of a rotating electric machine in the insulation diagnostic system 100a for rotating electric machines. The learning device 1a includes a learning data acquisition unit 11a and a model generation unit 12a.

[0039] The learning data acquisition unit 11a acquires learning data, which is a combination of first input data and second input data. Here, the first input data includes operating information of the rotating electric machine. The operating information of the rotating electric machine may include, for example, at least one of the number of times the rotating electric machine has been started and stopped, or the operating time of the rotating electric machine. The second input data is the dielectric strength remaining rate of the main insulation layer of the stator coil in the rotating electric machine operated under the conditions indicated by the first input data. The learning data acquisition unit 11a may acquire the input data from a data input device or database (not shown). The data input device is, for example, a keyboard, mouse, keypad, or touch panel, which is operated by the user. The database is, for example, a storage device that stores data including operating information of the rotating electric machine.

[0040] The model generation unit 12a learns the dielectric strength remaining rate of the main insulating layer of a stator coil in a rotating electric machine operated under the conditions indicated by the first input data, based on the training data, which is a combination of the first input data and the second input data output from the training data acquisition unit 11a. In other words, it generates a trained model that infers the dielectric strength remaining rate of the main insulating layer of a stator coil in a rotating electric machine operated under the conditions indicated by the first input data, from the training data, which is a combination of the first input data and the second input data. Here, the training data is data obtained by relating the first input data and the second input data to each other.

[0041] The learning algorithm used by the model generation unit 12a can be any known algorithm, such as supervised learning, unsupervised learning, reinforcement learning, or semi-supervised learning. As an example, the case in which a neural network is applied will be described. The model generation unit 12a learns the dielectric strength remaining rate of the main insulating layer by so-called supervised learning, following a neural network model, for example. Here, supervised learning is a method in which a pair of input and result (label) data is provided to the learning device 1a, the features of these learning data are learned, and the result is inferred from the input.

[0042] In this learning device 1a, the neural network learns the insulation withstand voltage survival rate of the main insulation layer of the stator coil in a rotating electrical machine operated under the conditions indicated by the first input data through so-called supervised learning according to the learning data created based on the combination of the first input data and the second input data acquired by the learning data acquisition unit 11a. That is, the neural network inputs the first input data including the operation information of the rotating electrical machine to the input layer, and learns by adjusting the weights so that the result output from the output layer approaches the insulation withstand voltage survival rate of the main insulation layer of the stator coil in the rotating electrical machine operated under the conditions indicated by the first input data. The model generation unit 12a generates and outputs a learned model by executing the above learning. The learned model storage unit 2a stores the learned model output from the model generation unit 12a.

[0043] Next, the process learned by the learning device 1a will be described. FIG. 8 is a flowchart for explaining the processes of the learning device 1a and the learned model storage unit 2a. Step S21 is a learning data acquisition step, step S22 is a learning process step, and step S23 is a learned model storage step.

[0044] In step S21, the learning data acquisition unit 11a acquires learning data and outputs it to the model generation unit 12a, and then proceeds to step S22. Note that the learning data acquisition unit 11a may acquire the first input data and the second input data included in the learning data simultaneously, as long as the first input data and the second input data can be acquired in association with each other, or the first input data and the second input data may be acquired at different timings.

[0045] In step S22, the model generation unit 12a learns the dielectric strength remaining ratio of the main insulating layer of the stator coil in a rotating electric machine operated under the conditions indicated by the first input data, using so-called supervised learning, according to the learning data created based on the combination of the first input data and the second input data acquired by the learning data acquisition unit 11a. It generates a trained model, outputs the generated trained model to the trained model storage unit 2a, and proceeds to step S23. In step S23, the trained model storage unit 2a stores the trained model acquired from the model generation unit 12a and terminates processing.

[0046] <Utilization Phase> Next, we will describe an insulation diagnostic system for a rotating electric machine that utilizes a trained model generated by the learning device 1a. Figure 9 is a block diagram showing the configuration of this rotating electric machine insulation diagnostic system 100a. The rotating electric machine insulation diagnostic system 100a diagnoses the insulation performance of a rotating electric machine. The rotating electric machine insulation diagnostic system 100a comprises a trained model storage unit 2a, an inference device 3a, and a diagnostic unit 4a. The inference device 3a comprises a data acquisition unit 31a and an inference unit 32a. The trained model storage unit 2a, the inference device 3a, and the diagnostic unit 4a may be configured on the same hardware, or they may be separate devices connected via a network. Furthermore, the trained model storage unit 2a, the inference device 3a, and the diagnostic unit 4a may reside on a cloud server.

[0047] The data acquisition unit 31a of the inference device 3a acquires data including operating information of a rotating electric machine as input data when diagnosing the insulation performance of the rotating electric machine. The operating information of the rotating electric machine may include, for example, at least one of the number of times the rotating electric machine has been started and stopped and the operating time of the rotating electric machine. The data acquisition unit 31a may acquire input data from a data input device or database (not shown). The data input device is, for example, a keyboard, mouse, keypad, or touch panel, and is operated by the user. The database is, for example, a storage device that stores data including design and operating information of the rotating electric machine.

[0048] The inference unit 32a uses the learned model stored in the learned model storage unit 2a to infer the insulation breakdown voltage survival rate of the main insulation layer under the conditions indicated by the input data. That is, by inputting the input data acquired by the data acquisition unit 31a into this learned model, it is possible to infer the insulation breakdown voltage survival rate of the main insulation layer of the stator coil in the rotating electrical machine operated under the conditions indicated by the input data, and output the inference result.

[0049] In addition, in the second embodiment, the inference unit 32a has been described as outputting the insulation breakdown voltage survival rate of the main insulation layer using the learned model learned by the model generation unit 12a. However, a learned model may be acquired from an external device such as another learning device, and the insulation breakdown voltage survival rate of the main insulation layer that is the output based on this learned model may be output. The diagnosis unit 4a determines whether or not the insulation breakdown voltage survival rate of the main insulation layer, which is the output of the inference unit 32a, satisfies a predetermined criterion, and outputs the determination result.

[0050] FIG. 10 is a flowchart for explaining the processing of the inference device 3a and the diagnosis unit 4a in the second embodiment. Step S31 is a data acquisition step, step S32 is an inference processing step, and step S33 is a diagnosis processing step.

[0051] In step S31, the data acquisition unit 31a acquires, as input data, data including the operation information of the rotating electrical machine when diagnosing the insulation performance of the rotating electrical machine, and outputs it to the inference unit 32a, and proceeds to step S32. In step S32, the inference unit 32a inputs the input data acquired from the data acquisition unit 31a into the learned model stored in the learned model storage unit 2a, and acquires the insulation breakdown voltage survival rate of the main insulation layer of the stator coil in the rotating electrical machine operated under the conditions indicated by this input data, and outputs it to the diagnosis unit 4a, and proceeds to step S33. In step S33, the diagnosis unit 4a determines whether or not the insulation breakdown voltage survival rate of the main insulation layer, which is the output data acquired from the inference unit 32a, satisfies a predetermined criterion, outputs the determination result, and ends the processing.

[0052] Figure 11 is a block diagram showing another configuration of the insulation diagnostic system 100a for a rotating electric machine that utilizes a trained model generated by the learning device 1a. The insulation diagnostic system 100a for a rotating electric machine shown in Figure 11 comprises a learning device 1a, a trained model storage unit 2a, an inference device 3a, and a diagnostic unit 4a. In the insulation diagnostic system 100a for a rotating electric machine shown in Figure 11, the operation of the learning device 1a, the trained model storage unit 2a, the inference device 3a, and the diagnostic unit 4a is the same as that of the learning device 1a, the trained model storage unit 2a, the inference device 3a, and the diagnostic unit 4a shown in Figures 7 and 9.

[0053] The learning device 1a, the trained model storage unit 2a, the inference device 3a, and the diagnostic unit 4a may be configured on the same hardware, or they may be separate devices connected via a network. Alternatively, the learning device 1a, the trained model storage unit 2a, the inference device 3a, and the diagnostic unit 4a may reside on a cloud server. When the learning device 1a and the inference device 3a are implemented by a single device, the training data acquisition unit 11a may also function as the data acquisition unit 31a.

[0054] The model generation unit 12a may learn the dielectric strength remaining rate of the main insulating layer according to the training data created for the insulation diagnostic systems of multiple rotating electric machines. The model generation unit 12a may acquire training data from the insulation diagnostic systems of multiple rotating electric machines used in the same area, or it may learn the dielectric strength remaining rate of the main insulating layer using training data collected from the insulation diagnostic systems of multiple rotating electric machines operating independently in different areas.

[0055] Furthermore, it is possible to add or remove rotating electrical machine insulation diagnostic systems from the target midway through the process, as they collect training data. In addition, a learning device that has learned the dielectric strength retention rate of the main insulation layer for one rotating electrical machine insulation diagnostic system may be applied to another rotating electrical machine insulation diagnostic system, and the dielectric strength retention rate of the main insulation layer for that other rotating electrical machine insulation diagnostic system may be relearned and updated.

[0056] As described above, the configuration and operation of the learning device 1a and insulation diagnostic system 100a, which are elements constituting the learning device and insulation diagnostic system according to this embodiment, have been explained. By combining these with the inference device 3 shown in Embodiment 1, the learning device 1b and insulation diagnostic system 100b according to Embodiment 2 can be constructed. The details thereof will be explained below.

[0057] <Learning Phase> Figure 12 is a block diagram showing the configuration of the learning device 1b according to Embodiment 2. Compared to the learning device 1a shown in Figure 7, the learning device 1b according to this embodiment shown in Figure 12 has the addition of a trained model storage unit 2 and an inference device 3. The other components of the learning device 1b in Embodiment 2, the training data acquisition unit 11a and the model generation unit 12a, have the same functions as the learning device 1a described earlier. Furthermore, the trained model storage unit 2 and the inference device 3 have the same functions as the trained model storage unit 2 and the inference device 3 in Embodiment 1.

[0058] The learning data acquisition unit 11a acquires learning data, which is a combination of first input data and second input data. Here, the first input data includes operation information relating to the operation history of the rotating electric machine, as in Embodiment 1, but may further include the non-destructive electrical characteristics of the main insulating layer of the stator coil of the rotating electric machine inferred from the operation information of the rotating electric machine. Here, these non-destructive electrical characteristics are obtained by performing inference from the operation information of the rotating electric machine using the first trained model stored in the trained model storage unit 2. The operation information of the rotating electric machine may include, for example, at least one of the number of starts and stops of the rotating electric machine and the operating time of the rotating electric machine. The non-destructive electrical characteristics of the main insulating layer may also include at least one of the increase in the dielectric loss tangent of the main insulating layer, the first current surge voltage of the main insulating layer, or the maximum partial discharge charge amount of the main insulating layer. When the inference device 3 and the learning device 1a are implemented by a single device, the data acquisition unit 31 may also serve as the learning data acquisition unit 11a.

[0059] On the other hand, the second input data is the dielectric strength remaining rate of the main insulation layer of the stator coil in a rotating electric machine operated under the conditions indicated by the first input data. The model generation unit 12a learns the dielectric strength remaining rate of the main insulation layer of the stator coil in a rotating electric machine operated under the conditions indicated by the operating information of the rotating machine in the first input data, based on the training data, which is a combination of the first input data and the second input data output from the training data acquisition unit 11a. That is, it generates a second trained model that infers the dielectric strength remaining rate of the main insulation layer of the stator coil in a rotating electric machine operated under the conditions indicated by the first input data, from the training data, which is a combination of the first input data and the second input data. Here, the training data is data that associates the first input data and the second input data with each other. The trained model storage unit 2a stores the second trained model output from the model generation unit 12a.

[0060] Next, the learning process of the learning device 1b will be explained. Figure 13 is a flowchart illustrating the process of the learning device 1b according to Embodiment 2. In step S41, the data acquisition unit 31 acquires the operating information of the rotating electric machine when diagnosing the insulation performance of the rotating electric machine as input data and outputs it to the first inference unit 32, and proceeds to step S42. In step S42, the first inference unit 32 inputs the input data acquired from the data acquisition unit 31 into the first learned model stored in the learned model storage unit 2, and infers and acquires the non-destructive electrical characteristics of the main insulating layer of the stator coil in the rotating electric machine operated under the conditions indicated by this input data.

[0061] In step S43, the learning data acquisition unit 11a acquires a combination of the operating information of the rotating electric machine, the dielectric strength remaining rate of the main insulation layer of the stator coil in the rotating electric machine operated under the conditions indicated by this operating information, and the non-destructive electrical characteristics acquired in step S42, as learning data, outputs it to the model generation unit 12, and proceeds to step S44. The learning data acquisition unit 11a may acquire the various data included in the learning data simultaneously, or it may acquire them at different times. In short, it is sufficient that the operating information, dielectric strength remaining rate, and non-destructive electrical characteristics are acquired in association with each other.

[0062] In step S44, the model generation unit 12a uses the training data acquired by the training data acquisition unit 11a to learn the dielectric strength remaining ratio in association with the operating information and non-destructive electrical characteristics through so-called supervised learning. Furthermore, the model generation unit 12a generates a second trained model and outputs the generated second trained model to the trained model storage unit 2a, proceeding to step S45. In step S45, the trained model storage unit 2a stores the second trained model acquired from the model generation unit 12a and terminates the process.

[0063] <Application Phase> Next, the insulation diagnostic system for a rotating electric machine in Embodiment 2 will be described. Figure 14 is a block diagram showing the configuration of the insulation diagnostic system 100b for a rotating electric machine in Embodiment 3. Comparing the insulation diagnostic system 100b for a rotating electric machine in Embodiment 2 shown in Figure 14 with the insulation diagnostic system 100a for a rotating electric machine shown in Figure 9, a trained model storage unit 2 and an inference device 3 have been added. Here, the inference device 3 has the same functions as the one shown in Figure 12 and consists of a data acquisition unit 31 and a first inference unit 32. The inference device 3 also has the same functions as the one shown in Figure 9 and consists of a data acquisition unit 31a and a second inference unit 32b. The trained model storage unit 2 stores the first trained model, and the trained model storage unit 2a stores the second trained model.

[0064] The trained model storage unit 2, the inference device 3, the trained model storage unit 2a, the inference device 3a, and the diagnostic unit 4a may be configured on the same hardware, or they may be separate devices connected via a network. Furthermore, the trained model storage unit 2, the inference device 3, the trained model storage unit 2a, the inference device 3a, and the diagnostic unit 4a may reside on a cloud server. When the inference device 3 and the inference device 3a are implemented by a single device, the data acquisition unit 31 may also function as the data acquisition unit 31a.

[0065] The data acquisition unit 31 of the inference device 3 acquires data including operating information of the rotating electric machine as input data. Based on this operating information, the first inference unit 32 uses the first trained model stored in the trained model storage unit 2 to infer the non-destructive electrical characteristics of the rotating electric machine.

[0066] On the other hand, the data acquisition unit 31a of the inference device 3a acquires data including operating information of the rotating electric machine as input data when diagnosing the insulation performance of the rotating electric machine. Here, the data including operating information of the rotating electric machine may further include the non-destructive electrical characteristics of the main insulating layer of the stator coil of the rotating electric machine inferred from the operating information of the rotating electric machine. The non-destructive electrical characteristics of the main insulating layer may include at least one of the increase in dielectric loss tangent of the main insulating layer, the first current surge voltage of the main insulating layer, or the maximum partial discharge charge amount of the main insulating layer.

[0067] The second inference unit 32b uses the second trained model stored in the trained model storage unit 2a to infer the dielectric strength remaining rate of the main insulating layer under the conditions indicated by the operating information of the rotating electric machine acquired by the data acquisition unit 31a. That is, by inputting the operating information of the rotating electric machine acquired by the data acquisition unit 31a into this second trained model, the dielectric strength remaining rate of the main insulating layer of the stator coil in a rotating electric machine operated under the conditions indicated by this operating information can be inferred and the inference result can be output.

[0068] In this embodiment 2, the second inference unit 32b is described as outputting the dielectric strength remaining rate of the main insulating layer using the second trained model learned by the model generation unit 12a. However, the second trained model may be obtained from an external source such as another learning device, and the dielectric strength remaining rate of the main insulating layer may be output based on this second trained model. The diagnostic unit 4a determines whether the dielectric strength remaining rate of the main insulating layer, which is the output of the second inference unit 32b, meets a predetermined standard, and outputs the determination result.

[0069] Figure 15 is a flowchart illustrating the processing of the insulation diagnostic system 100b according to Embodiment 2. In step S51, the data acquisition unit 31 acquires operating information of the rotating electric machine when diagnosing the insulation performance of the rotating electric machine as input data and outputs it to the first inference unit 32, and proceeds to step S52. In step S52, the first inference unit 32 inputs the input data acquired from the data acquisition unit 31 into the first learned model stored in the learned model storage unit 2, and infers (first) and acquires the non-destructive electrical characteristics of the main insulation layer of the stator coil in the rotating electric machine operated under the conditions indicated by the input data.

[0070] In step S53, the data acquisition unit 31a acquires the operation information and the non-destructive electrical characteristics acquired in step S52 as input data and outputs it to the second inference unit 32b, proceeding to step S54. In step S54, the second inference unit 32b inputs the input data acquired from the data acquisition unit 31a into the second trained model stored in the trained model storage unit 2, infers (second) the dielectric strength remaining rate of the main insulating layer of the stator coil in a rotating electric machine operated under the conditions indicated by the input data, acquires it, outputs it to the diagnostic unit 4a, and proceeds to step S55. In step S55, the diagnostic unit 4a determines whether the dielectric strength remaining rate of the main insulating layer, which is the output data acquired from the second inference unit 32b, meets a predetermined standard, outputs the determination result, and terminates the process.

[0071] As described above, the learning device 1b according to Embodiment 2 is a learning device 1b that obtains a second learned model for inferring the dielectric strength remaining rate of the main insulating layer using a first learned model generated by the learning device 1, and comprises a data acquisition unit 31 that acquires operating information, a first inference unit 32 that outputs non-destructive electrical characteristics obtained when operating information is input using the first learned model, a learning data acquisition unit 11a that acquires learning data including operating information, non-destructive electrical characteristics obtained by the first inference unit 32 when the conditions indicated in the operating information are input, and the dielectric strength remaining rate when the rotating electric machine is operated under the conditions indicated by the operating information, and a model generation unit 12a that generates a second learned model for inferring the dielectric strength remaining rate from the operating information using the learning data.

[0072] Furthermore, the insulation diagnostic system 100b according to Embodiment 2 is an insulation diagnostic system that diagnoses the insulation status of the main insulation layer using a first trained model generated by a learning device 1 and a second trained model generated by a learning device 1b, and comprises a data acquisition unit 31 that acquires operating information, a first inference unit 32 that outputs non-destructive electrical characteristics obtained when operating information is input using the first trained model, a second inference unit 32b that outputs the dielectric strength remaining rate of the main insulation layer obtained when operating information and non-destructive electrical characteristics are input using the second trained model, and a diagnostic unit 4a that determines whether the dielectric strength remaining rate of the main insulation layer meets a predetermined standard and outputs the result of the determination.

[0073] According to the insulation diagnostic system 100b of this embodiment, the inference device 3 can infer non-destructive electrical characteristics from operating information. In the conventional technology, non-destructive electrical characteristics were actually measured and used to predict the dielectric strength remaining rate. Therefore, in order to make this prediction, it was necessary to measure not only the dielectric strength remaining rate but also the non-destructive electrical characteristics and then construct a second trained model, which required the effort of measuring the non-destructive electrical characteristics. In contrast, with the insulation diagnostic system 100b of this embodiment, non-destructive electrical characteristics can be inferred by using the first trained model, thus eliminating the effort of actually measuring them, and making it possible to infer the dielectric strength remaining rate using operating information and inferred values ​​of non-destructive electrical characteristics. Furthermore, since the dielectric strength remaining rate is predicted using non-destructive electrical characteristics that have a high correlation with the dielectric strength remaining rate as input data, there is an advantage in that the dielectric strength remaining rate can be predicted with greater accuracy compared to the case where the dielectric strength remaining rate is predicted using only the operating information of the rotating electric machine without using non-destructive electrical characteristics.

[0074] Furthermore, similar to Embodiment 1, it is also possible to add design information for the rotating electric machine. That is, the learning data acquisition unit 11a acquires operating information, design information, non-destructive electrical characteristics obtained by inference using a third learned model based on the conditions indicated in the operating information and design information, and the dielectric strength remaining rate when the rotating electric machine is operated under the conditions indicated by the operating information and design information as learning data, and the model generation unit 12a may use this learning data to generate a fourth learned model for inferring the dielectric strength remaining rate from the operating information, design information and non-destructive electrical characteristics. Here, the design information related to the dielectric strength in the main insulating layer of the stator coil may include, for example, at least one of the rated voltage of the rotating electric machine and the thickness of the main insulating layer.

[0075] Furthermore, the insulation diagnostic system 100b according to Embodiment 2 is an insulation diagnostic system that diagnoses the insulation status of the main insulation layer using a third trained model generated by a learning device 1 and a fourth trained model generated by a learning device 1b, and comprises a data acquisition unit 31 that acquires operation information and design information, a first inference unit 32 that outputs non-destructive electrical characteristics obtained when the operation information and design information are input using the third trained model, a second inference unit 32b that outputs the dielectric strength remaining rate of the main insulation layer obtained when the operation information, design information and non-destructive electrical characteristics are input using the fourth trained model, and a diagnostic unit 4a that determines whether the dielectric strength remaining rate of the main insulation layer meets a predetermined standard and outputs the result of the determination.

[0076] As described above, by adding design information in addition to operating information to the training data, the same insulation diagnostic system can be applied to multiple models with different design conditions, thereby reducing training costs. Another advantage is that the insulation diagnostic system learned from previously acquired training data can be reused for new models.

[0077] In Embodiments 1 and 2, the learning devices 1, 1a, 1b, the learned model storage units 2, 2a, the inference devices 3, 3a, the diagnostic units 4, 4a, and the isolation diagnostic systems 100, 100a, 100b are described, and an example of the hardware is shown in Figure 14, consisting of a processor 200 and a storage device 201 located inside the computer. Although not shown, the storage device comprises a volatile storage device such as random access memory and a non-volatile auxiliary storage device such as flash memory. Alternatively, a hard disk may be provided as an auxiliary storage device instead of flash memory. The processor 200 executes the program input from the storage device 201. In this case, the program is input from the auxiliary storage device to the processor 200 via the volatile storage device. The processor 200 may also output data such as calculation results to the volatile storage device of the storage device 201, or it may store the data in the auxiliary storage device via the volatile storage device.

[0078] While this disclosure describes various exemplary embodiments and examples, the various features, aspects, and functions described in one or more embodiments are not limited to the application of a particular embodiment, but are applicable individually or in various combinations to the embodiments. Accordingly, countless variations not illustrated are envisioned within the scope of the art disclosed in this specification. For example, these include modifying, adding or omitting at least one component, or extracting at least one component and combining it with a component from another embodiment.

[0079] 1, 1a, 1b Learning device, 11, 11a Learning data acquisition unit, 12, 12a Model generation unit, 2, 2a Learned model storage unit, 3, 3a Inference device, 31, 31a Data acquisition unit, 32 First inference unit, 32a Inference unit, 32b Second inference unit, 4, 4a Diagnostic unit, 100, 100a, 100b Insulation diagnostic system, 200 Processor, 201 Storage device

Claims

1. A learning device for obtaining a first trained model for inferring non-destructive electrical characteristics related to dielectric strength in the main insulating layer of the stator coil of a rotating electric machine using operating information related to the operating history of the rotating electric machine, comprising: a learning data acquisition unit that acquires learning data including the operating information and the non-destructive electrical characteristics when the rotating electric machine is operated under the conditions indicated in the operating information; and a model generation unit that generates a first trained model for inferring the non-destructive electrical characteristics from the operating information using the learning data.

2. An insulation diagnostic system for diagnosing the insulation status of the main insulating layer using the first trained model generated by the learning device described in claim 1, comprising: a data acquisition unit for acquiring the operating information; a first inference unit for outputting the non-destructive electrical characteristics obtained when the operating information is input using the first trained model; and a diagnostic unit for determining whether the non-destructive electrical characteristics meet predetermined criteria and for outputting the result of the determination.

3. A learning device for obtaining a second learned model for inferring the dielectric strength remaining rate of the main insulating layer using the first learned model generated by the learning device according to claim 1, comprising: a data acquisition unit for acquiring the operating information; a first inference unit for outputting the non-destructive electrical characteristics obtained when the operating information is input using the first learned model; a learning data acquisition unit for acquiring learning data including the operating information, the non-destructive electrical characteristics obtained by the first inference unit when the conditions indicated in the operating information are input, and the dielectric strength remaining rate when the rotating electric machine is operated under the conditions indicated by the operating information; and a model generation unit for generating the second learned model for inferring the dielectric strength remaining rate from the operating information using the learning data.

4. An insulation diagnostic system for diagnosing the insulation status of the main insulating layer using the first learned model used by the learning device described in claim 3 and the second learned model generated by the learning device, comprising: a data acquisition unit for acquiring the operating information; a first inference unit for outputting the non-destructive electrical characteristics obtained when the operating information is input using the first learned model; a second inference unit for outputting the dielectric strength remaining rate obtained when the operating information and the non-destructive electrical characteristics are input using the second learned model; and a diagnostic unit for determining whether the dielectric strength remaining rate satisfies a predetermined standard and for outputting the result of the determination.

5. The learning device according to claim 1, characterized in that the learning data acquisition unit acquires learning data including the operation information and the non-destructive electrical characteristics obtained when the device is operated under the conditions indicated in the operation information, as well as design information related to dielectric strength, and the model generation unit generates a third trained model for inferring the non-destructive electrical characteristics from the operation information and the design information using the learning data.

6. An insulation diagnostic system for diagnosing the insulation status of the main insulating layer using the third learned model generated by the learning device described in claim 5, comprising: a data acquisition unit for acquiring the operation information and the design information; a first inference unit for outputting the non-destructive electrical characteristics obtained when the operation information and the design information are input using the third learned model; and a diagnostic unit for determining whether the non-destructive electrical characteristics meet predetermined criteria and for outputting the result of the determination.

7. A learning device for obtaining a fourth learned model for inferring the dielectric strength remaining rate of the main insulating layer using the third learned model generated by the learning device according to claim 5, comprising: a data acquisition unit for acquiring the operation information and the design information; a first inference unit for outputting the non-destructive electrical characteristics obtained when the operation information and the design information are input using the third learned model; a learning data acquisition unit for acquiring learning data including the operation information and the design information, the non-destructive electrical characteristics obtained in the first inference unit when the conditions indicated in the operation information and the design information are input, and the dielectric strength remaining rate when the rotating electric machine is operated under the conditions indicated by the operation information and the design information; and a model generation unit for generating the fourth learned model for inferring the dielectric strength remaining rate from the operation information and the design information using the learning data.

8. An insulation diagnostic system for diagnosing the insulation status of the main insulating layer using the third learned model used by the learning device described in claim 7 and the fourth learned model generated by the learning device, comprising: a data acquisition unit that acquires the operation information and the design information; a first inference unit that outputs the non-destructive electrical characteristics obtained when the operation information and the design information are input using the third learned model; a second inference unit that outputs the dielectric strength remaining rate obtained when the operation information, the design information and the non-destructive electrical characteristics are input using the fourth learned model; and a diagnostic unit that determines whether the dielectric strength remaining rate meets a predetermined standard and outputs the result of the determination.

9. The learning device according to any one of claims 1, 3, 5, and 7, characterized in that the operating information includes at least one of the number of starts and stops of the rotating electric machine and the operating time of the rotating electric machine.

10. The learning device according to any one of claims 1, 3, 5, and 7, characterized in that the nondestructive electrical characteristics include at least one of the increase in dielectric loss tangent of the main insulating layer, the first current surge voltage of the main insulating layer, and the maximum partial discharge charge amount of the main insulating layer.

11. The learning device according to claim 5 or 7, characterized in that the design information includes at least one of the rated voltage of the rotating electric machine and the thickness of the main insulating layer.

12. The insulation diagnostic system according to any one of claims 2, 4, 6, and 8, characterized in that the operating information includes at least one of the number of starts and stops of the rotating electric machine and the operating time of the rotating electric machine.

13. The insulation diagnostic system according to any one of claims 2, 4, 6, and 8, characterized in that the nondestructive electrical characteristics include at least one of the increase in dielectric loss tangent of the main insulating layer, the first current surge voltage of the main insulating layer, and the maximum partial discharge charge amount of the main insulating layer.

14. The insulation diagnostic system according to claim 6 or 8, characterized in that the design information includes at least one of the rated voltage of the rotating electric machine and the thickness of the main insulating layer.

15. A program for causing a computer to function as a learning device according to any one of claims 1, 3, 5, 7, 9, 10, and 11.

16. A program for causing a computer to function as an insulation diagnostic system according to any one of claims 2, 4, 6, 8, 12, 13, and 14.

Images