WO2020044477A1

WO2020044477A1 - Abnormality diagnosis system, method, and program

Info

Publication number: WO2020044477A1
Application number: PCT/JP2018/032026
Authority: WO
Inventors: 幸造伴野; 俊也高野; 友祐星野
Original assignee: 株式会社東芝; 東芝エネルギーシステムズ株式会社
Priority date: 2018-08-29
Filing date: 2018-08-29
Publication date: 2020-03-05
Also published as: JPWO2020044477A1; JP6957762B2

Abstract

Provided are an abnormality diagnosis system, method, and program capable of accurately diagnosing the classification of an abnormality cause under diagnosis. In the present invention, a diagnosis unit comprises a data processing unit for, upon receiving data output from an object of diagnosis, generating preprocessed data in which noise, which consists of features other than features that make it possible to identify the category of an abnormality cause from the data, has been attenuated and an abnormality cause identification unit for, upon receiving the preprocessed data, identifying the category of an abnormality cause under diagnosis on the basis of a diagnosis model for outputting the classification of an abnormality cause under diagnosis. The diagnosis model is a machine learning model. A learning unit comprises a diagnosis model learning unit for generating a diagnosis model through machine learning and a noise attenuation unit for generating attenuated data in which the aforementioned noise has been removed from learning data for the diagnosis model machine learning. The learning data is either simulated data simulating abnormalities in the object of diagnosis or real data indicating abnormalities in the object of diagnosis. The diagnosis model learning unit generates a diagnosis model through machine learning by using the attenuated data as the learning data and using abnormality cause classifications as teaching data.

Description

Abnormality diagnosis system, method and program

The embodiments of the present invention relate to an abnormality diagnosis system, method, and program.

機械 Machine learning, such as neural networks, is used to identify the causes of failures, accidents, etc. in equipment or equipment that supports social infrastructure such as power equipment, such as failures and accidents. In order to construct a classifier using machine learning, it is generally necessary to train a large number of abnormal data.However, failures and accidents of social infrastructure equipment and devices rarely occur, and the number of actual abnormal data Is less. Furthermore, such facilities and equipment are installed in various environments, and the data obtained from each facility and equipment is affected by noise and the like specific to each environment, but abnormal data covering all environments is collected. It is difficult to prepare.

診断 Regarding diagnosis by machine learning in a situation where the number of data is limited, a technique is known in which learning is performed using a large number of normal data and a small number of abnormal data to distinguish between normal and abnormal. For example, learning is performed using a small number of data when the object surface is frozen and a large number of data when the object surface is not frozen, and whether or not the object surface is frozen is identified.

However, according to the above-mentioned technology, it is possible to distinguish between two states of normal and abnormal based on a large number of normal data, but it is not possible to identify the type of abnormal data having a small number of data. That is, the type of the cause of the abnormality to be diagnosed indicated by the abnormality data cannot be identified.

JP-A-2017-125809

As a measure to identify the type of abnormal data, simulated data that simulates a failure is prepared in advance by simulation or desk experiment, and the simulated data is learned to construct a classifier. A method of identifying data and confirming the identification accuracy can be considered.

However, the environment in which the equipment and devices are installed cannot be completely reproduced by simulation or desk experiment, and the simulated data and the actual data may have different characteristics such as noise. In this case, in the classifier that has learned only the simulation data, the simulation data can be correctly identified, but there is a problem that the actual data cannot be identified correctly. In addition, if the simulation data or the actual data contains noise peculiar to the data and the data is biased, the noise is erroneously learned to be a feature indicating the cause of the abnormality, and the identification accuracy is deteriorated. was there.

An embodiment of the present invention has been made to solve the above-described problem, and has an object to provide an abnormality diagnosis system, method, and program that can accurately diagnose an abnormality cause type to be diagnosed. And

In order to achieve the above object, an abnormality diagnosis system according to the present embodiment is an abnormality diagnosis system that diagnoses an abnormality cause of a diagnosis target, and a diagnosis unit that identifies a type of the abnormality cause of the diagnosis target based on a model. And a learning unit that generates the model by machine learning, wherein the diagnostic unit is capable of identifying the type of the cause of the abnormality from the data when the data output from the diagnosis target is input. A data processing unit that generates pre-processed data in which noise, which is a characteristic part other than the part, is attenuated, and a diagnostic model that outputs the type of abnormality cause of the diagnosis target when the pre-processed data is input, An abnormality cause identification unit that identifies a type of the abnormality cause to be diagnosed, wherein the diagnosis model is a machine learning model, and the learning unit is configured to execute the diagnosis model by machine learning. A diagnostic model learning unit for generating, and a noise attenuating unit for generating attenuated data in which the noise is attenuated from learning data for machine learning of the diagnostic model, wherein the learning data includes the diagnostic data. Simulation data simulating an abnormality of the target or actual data indicating the abnormality of the diagnosis target, wherein the diagnosis model learning unit performs the diagnosis by machine learning using the attenuated data as learning data and the abnormality cause type as teacher data. The method is characterized in that a model is generated.

The present embodiment can also be understood as a method of realizing the processing of each unit by a computer or an electronic circuit, and a program for realizing the processing of each unit by a computer.

That is, the abnormality diagnosis method of the present embodiment is an abnormality diagnosis method for diagnosing an abnormality cause of a diagnosis target, and includes a diagnosis step of identifying a type of the abnormality cause of the diagnosis target based on a model; The diagnostic step is a characteristic part other than the characteristic part that, when the data output from the diagnosis target is input, enables the type of the abnormal cause to be identified from the data. A data processing step of generating pre-processed data in which noise has been attenuated, and a diagnosis model that outputs an abnormality cause type of the diagnosis target when the pre-processed data is input, based on a diagnosis model of the abnormality cause of the diagnosis target. An abnormality cause identification step of identifying a type, wherein the diagnostic model is a machine learning model, and A diagnostic model learning step generated by learning, and a noise attenuation step of generating attenuated data in which the noise is attenuated from learning data for machine learning of the diagnostic model, wherein the learning data includes: Simulation data simulating the abnormality of the diagnosis target or actual data indicating the abnormality of the diagnosis target, wherein the diagnostic model learning step uses the attenuated data as learning data and the abnormality cause type as teacher data by machine learning. The diagnostic model is generated.

The abnormality diagnosis program according to the present embodiment is an abnormality diagnosis program for diagnosing an abnormality cause of a diagnosis target.The computer includes: a diagnosis step for identifying a type of the abnormality cause of the diagnosis target based on a model; A learning step of generating by learning, wherein the diagnosis step is characterized in that, when data output from the diagnosis target is input, a characteristic part other than a characteristic part that enables the type of the abnormal cause to be identified from the data. A data processing step of generating pre-processed data in which noise has been attenuated, and an abnormality of the diagnosis target based on a diagnosis model that outputs an abnormality cause type of the diagnosis target when the pre-processed data is input. An abnormality cause identification step of identifying a cause type, wherein the diagnostic model is a machine learning model, and the learning step A diagnostic model learning step of generating the diagnostic model by machine learning, and a noise attenuation step of generating attenuated data in which the noise is attenuated from learning data for machine learning of the diagnostic model, The learning data is simulated data simulating the abnormality of the diagnosis target or actual data indicating the abnormality of the diagnosis target, and the diagnosis model learning step uses the attenuated data as learning data and sets the abnormality cause type. The diagnostic model is generated by machine learning as teacher data.

1 is a diagram illustrating a configuration of an abnormality diagnosis system according to a first embodiment. FIG. 5 is a diagram illustrating an example of data acquired by a data acquisition unit. 6 shows an example of a table of learning data stored in a learning data storage unit. It is a schematic diagram of a neural network of a data conversion unit. It is a schematic diagram of a neural network of a data identification unit. It is a schematic diagram of the said neural network when the machine learning model of an abnormality cause identification part is a neural network. 5 is an operation flowchart illustrating an example of an operation of the abnormality diagnosis system. It is a schematic diagram of the neural network of the data conversion part and the data identification part at the time of learning. FIG. 5 is a diagram illustrating a display example of a diagnosis result on a display unit according to the first embodiment. FIG. 6 is a diagram illustrating a configuration of an abnormality diagnosis system according to a second embodiment. FIG. 4 is a diagram illustrating an example of a screen displayed on a display unit by a display control unit. 9 is an operation flowchart illustrating an example of an operation of the abnormality diagnosis system according to the second embodiment. It is a figure showing an example of a screen which a display control part concerning modification 1 of a 2nd embodiment displays on a display part. It is a figure showing an example of a screen which a display control part concerning modification 2 of a 2nd embodiment displays on a display part. It is a figure showing an example of a screen which a display control part concerning modification 3 of a 2nd embodiment displays on a display part. It is a schematic diagram of the neural network of the data conversion part, the data identification part, and the abnormality cause identification part at the time of learning. 4 is an operation flowchart illustrating an example of an operation of the abnormality diagnosis system according to the present embodiment based on the first embodiment. It is an operation | movement flowchart which shows an example of operation | movement of the abnormality diagnosis system of this embodiment based on 2nd Embodiment. It is a figure for explaining an operation of a 3rd embodiment. 14 is an operation flowchart illustrating an example of an operation of the abnormality diagnosis system according to the fourth embodiment. It is a figure showing an example of a screen which a display control part displays on a display part in a 5th embodiment. It is an operation | movement flowchart which shows an example of operation | movement of the abnormality diagnosis system which concerns on 5th Embodiment. It is a figure showing the composition of the abnormality diagnosis system concerning a 7th embodiment. It is a figure showing an example of a screen which a display control part displays on a display part in a 7th embodiment. It is an operation | movement flowchart which shows an example of an operation | movement of the abnormality diagnosis system 1 of 7th Embodiment. It is an operation | movement flowchart which shows an example of operation | movement of the abnormality diagnosis system 1 which concerns on the modification of 7th Embodiment. It is a figure showing an example of a screen which a display control part concerning modification 1 of a 7th embodiment displays on a display part. It is a figure showing an example of a screen which a display control part concerning modification 2 of a 7th embodiment displays on a display part.

[First Embodiment]
(Schematic configuration)
FIG. 1 is a diagram illustrating a configuration of the abnormality diagnosis system according to the first embodiment. The abnormality diagnosis system 1 constructs a machine learning model, and diagnoses the cause of an abnormality in equipment or equipment to be diagnosed (hereinafter, also simply referred to as “diagnosis object”) using the machine learning model. The diagnosis target is, for example, equipment or equipment used in a power system, and the abnormality diagnosis system 1 is used in a power system monitoring system or the like. As shown in FIG. 1, a data acquisition unit 100 and a learning data storage unit 200 are connected to the abnormality diagnosis system 1.

The data acquisition unit 100 acquires data (hereinafter, simply referred to as “diagnosis target data”) diagnosed by the abnormality diagnosis system 1. The diagnosis target data is data measured when an abnormality occurs in the diagnosis target. Examples of the abnormality of the diagnosis target include a failure of the diagnosis target and a defect due to an accident. The diagnosis target data is, for example, waveform data as shown in FIG. However, the diagnosis target data is not limited to the waveform data, and may be image data.

The learning data storage unit 200 stores learning data used for learning of the abnormality diagnosis system 1. The learning data is actual data of past abnormalities of the actual diagnosis target, and simulated data imitating the abnormalities of the diagnosis target.

In the learning data storage unit 200, the actual data of the past actual abnormality of the diagnosis target and the simulated data simulating the abnormality of the diagnosis target are provided for each data, the information indicating the abnormality cause type, and the actual data or the simulated data. Information indicating the data type of the data is added and stored. For example, as shown in FIG. 3, actual data at the time of past actual equipment failure (“actual failure data” in the figure) and simulated data simulating a failure by a simulation or a desk test are included in each data. Information indicating the cause type is associated with information indicating the data type. The actual data stored in the learning data storage unit 200 is data indicating an abnormality of a diagnosis target measured in the past. In FIG. 3, the failure cause types are two types, failures A and B, but may be three or more types.

(4) The learning data may include noise specific to the simulation data or the real data. This noise is a part where the cause of abnormality is not reflected. That is, since the simulated data or the actual data is data indicating one of the abnormality cause types, the simulation data or the actual data includes a characteristic portion that makes it possible to identify the abnormality cause type and a noise portion that is another characteristic portion. This noise portion is a characteristic portion other than the characteristic portion that makes the abnormality cause type identifiable. For example, simulated data or real data is made into waveform data as shown in FIG. Assuming that the characteristic portion is a peak of a waveform, it is an offset value of the waveform. The specific noise of the actual data is, for example, noise that reflects the environment in which the diagnosis target is installed, and the specific noise of the simulation data is, for example, noise that reflects an experimental environment such as a desk experiment.

The abnormality diagnosis system 1 includes a single computer or a plurality of computers and a display device connected to a network. The abnormality diagnosis system 1 stores a program and a database in a storage such as an HDD or an SSD, and appropriately develops the program and the database in a memory such as a RAM, and processes the data by a CPU. Perform necessary calculations.

Specifically, the abnormality diagnosis system 1 includes a processing unit 2, a storage unit 3, an input unit 4, and a display unit 5. The storage unit 3 includes a memory or a storage, and stores an operation program of the processing unit 2, an operation result, and the like. The input unit 4 is an input interface by a user, and is, for example, a keyboard, a mouse, and a touch panel. The display unit 5 displays the calculation result of the processing unit 2. The display unit 5 is a display device such as an organic EL or a liquid crystal display.

The processing unit 2 includes a CPU and performs various operations described later. Specifically, the processing unit 2 includes a diagnosis unit 21, a learning unit 22, and a display control unit 23.

The diagnosis unit 21 includes a CPU, and identifies the type of abnormality cause to be diagnosed based on the model. The learning unit 22 generates the model by machine learning.

Specifically, the diagnosis unit 21 includes a data processing unit 211 and an abnormality cause identification unit 212. The data processing unit 211 is configured to include a CPU, and when data to be diagnosed is input, generates preprocessed data in which noise is attenuated from the data. That is, the data processing unit 211 acquires the data to be diagnosed from the data acquisition unit 100, and generates preprocessed data by converting the data to be diagnosed based on a noise attenuation model described later. Here, the noise refers to a portion of the diagnosis target data that is a feature other than a feature portion that makes it possible to identify the cause of the abnormality. For example, if the data to be diagnosed is waveform data as shown in FIG. 2 and the characteristic portion that makes it possible to identify the type of the cause of the abnormality is a waveform peak, it is the offset value of the waveform.

The abnormality cause identification unit 212 includes a CPU, and identifies the type of abnormality cause to be diagnosed based on a diagnostic model that outputs the type of abnormality cause to be diagnosed when preprocessed data is input. The diagnostic model is a machine learning model, for example, a neural network, a decision tree, a random forest, an SVM (support @ vector @ machine), or the like can be used. FIG. 6 is a schematic diagram of the neural network when the machine learning model of the abnormality cause identification unit 212 is a neural network. As illustrated in FIG. 6, the abnormality cause identification unit 212 receives the preprocessed data output from the data processing unit 211 as input and outputs the abnormality cause type.

The learning unit 22 includes a CPU and includes a noise attenuation unit 221, a noise attenuation model generation unit 222, and a diagnostic model learning unit 223.

The noise attenuator 221 is configured to include a CPU, and generates attenuated data in which noise has been removed from learning data for machine learning of a diagnostic model. The learning data is learning data acquired from the learning data storage unit 200. Here, the noise refers to a part of the learning data that has a characteristic other than a characteristic part that enables the cause of the abnormality to be identified. For example, if the learning data is waveform data as shown in FIG. 3 and the characteristic portion that makes it possible to identify the cause of the abnormality is a waveform peak, it is an offset value of the waveform.

More specifically, the noise attenuator 221 generates attenuated data based on a noise attenuation model that attenuates noise from input data. This noise attenuation model is used not only by the noise attenuation unit 221 in the learning stage for generating a diagnostic model, but also by the diagnostic unit 21 in the diagnostic stage. That is, the data processing unit 211 attenuates noise from the input diagnosis target data and generates preprocessed data based on the noise attenuation model.

The noise attenuation model generation unit 222 includes a CPU and generates a noise attenuation model used in the noise attenuation unit 221. The noise attenuation model generation unit 222 includes a data conversion unit 222a, a data identification unit 222b, a conversion model learning unit 222c, and an identification model learning unit 222d.

The data conversion unit 222a is configured to include a CPU, and converts the learning data based on a conversion model for converting the learning data and outputs converted data. The learning data is learning data acquired from the learning data storage unit 200. The transformation model is here a neural network. FIG. 4 is a schematic diagram of a neural network of the data conversion unit 222a. This neural network is, for example, an auto encoder.

The data identification unit 222b includes a CPU, and based on an identification model for identifying the data type of the learning data when the converted data is input, a data type indicating whether the learning data is simulated data or real data. Identify. The identification model is here a neural network. FIG. 5 is a schematic diagram of a neural network of the data identification unit 222b. The data identification unit 222b acquires the converted data from the data conversion unit 222a, and identifies, using this neural network, whether the learning data that is the source of the converted data is simulated data or real data.

The identification model learning unit 222d includes a CPU and generates an identification model by machine learning. That is, the identification model learning unit 222d generates the identification model so that the identification model is a model that correctly identifies the data type of the learning data from the converted data. Specifically, an identification model is generated by updating the model of the data identification unit 222b so that the error between the output result of the data identification unit 222b and the data type of the learning data serving as teacher data is reduced.

The conversion model learning unit 222c includes a CPU and generates a noise attenuation model by machine learning. That is, a noise attenuation model is generated by updating the conversion model of the data conversion unit 222a by machine learning. Specifically, the conversion model learning unit 222c generates a noise attenuation model that is similar to the learning data but outputs the converted data so that the data identification unit 222b cannot correctly identify the data. More specifically, the conversion model learning unit 222c reduces the error between the learning data and the converted data, and the error between the output result of the data identification unit 222b and the incorrect data type. To generate a noise attenuation model. Note that the incorrect data type is a data type obtained by inverting the data type of the learning data corresponding to the output result of the data identification unit 222b.

The diagnostic model learning unit 223 includes a CPU and generates a diagnostic model by machine learning. That is, the diagnostic model learning unit 223 generates a diagnostic model by a machine learning model using the attenuated data output from the noise attenuating unit 221 as learning data and the abnormality cause type as teacher data. The abnormality cause type used as the teacher data is the abnormality cause type of the learning data (input) corresponding to the attenuated data (output). Also, the diagnostic model learning unit 223 may update the diagnostic model once constructed using new teacher data.

The display control unit 23 includes a CPU, and causes the display unit 5 to display the identification result of the abnormality cause identification unit 212 (diagnosis model).

(Detailed configuration)
The processing of the abnormality diagnosis system 1 will be described in detail with reference to a flowchart. FIG. 7 is an operation flowchart illustrating an example of the operation of the abnormality diagnosis system 1.

As shown in FIG. 7, first, the conversion model learning unit 222c and the identification model learning unit 222d learn each neural network of the data conversion unit 222a and the data identification unit 222b (step S01).

That is, as shown in FIG. 8, the network is configured such that the output of the neural network of the data conversion unit 222a becomes the input of the neural network of the data identification unit 222b. The data conversion unit 222a acquires learning data from the learning data storage unit 200, and performs data conversion to output converted data. Then, the data identification unit 222b acquires the converted data from the data conversion unit 222a, and identifies the data type. The discrimination model learning unit 222d acquires the output result of the data discrimination unit 222b and the data type of the learning data input to the data conversion unit 222a, which is the source of the output result, as teacher data. Is used to update the weight of the neural network of the data identification unit 222b, and the conversion model learning unit 222c acquires the converted data from the data conversion unit 222a and the learning data from the learning data storage unit 200 as the teacher data. The weight of the neural network of the data conversion unit 222a is updated using the backpropagation method, taking into account the output result of the data identification unit 222b.

More specifically, the discrimination model learning unit 222d uses the neural network of the data discrimination unit 222b as a two-class using the output data of the data conversion unit 222a as an input variable and a data type indicating whether it is simulated data or real data as an output variable. Train as a classifier. For example, the k data type (0 or 1), one-hot representation of true label indicating a data type and _{t k} (i.e., only the index to be correct label 1, others are set to _0), data _{y D1k} When the output value of the k-th output node is used as the output value of the k-th output node by the identification unit 222b, the identification model learning unit 222d calculates Expression (1) indicating the cross entropy error as a loss function, and uses the error backpropagation method to calculate the data. the output _{y D1k} of 222b as close to the true label _{t k,} generating an identification model by updating the weights included in the neural network of the data identification unit 222b.

At the same time, the conversion model learning unit 222c receives the learning data obtained from the learning data storage unit 200 as an input to the neural network of the data conversion unit 222a, and is similar to the input data. A noise attenuation model is generated by learning as a data converter that converts and outputs input data so that it cannot be correctly identified.

That is, the conversion model learning unit 222c constructs a neural network of the data conversion unit 222a based on an auto encoder (Auto Encoder). Specifically, the index data of the l, a u _l of l-th input data value, when the output value of the l-th data with the y _ael to the data conversion section 222a, using the formula (2) as a loss function Thus, the data converter 222a and the noise attenuator 221 are constructed as an auto encoder, and output a value similar to the input data.

Further, the conversion model learning unit 222c sets the auto encoder so that the data type output by the data identification unit 222b is correct so that the data identification unit 222b outputs data that cannot be correctly identified as simulated data or real data. The learning is performed so as to be inverted with respect to the data type of the learning data. That is, the learning is performed so that the error between the data type output from the data identification unit 222b and the data type obtained by inverting the data type of the learning data as the correct answer is improved.

Specifically, the conversion model learning unit 222c updates the conversion model of the data conversion unit 222a based on the expression (3) as a loss function so as to improve the error E _{ae ′} of the expression (3). , And outputs data that is similar to the input data (learning data) and cannot be correctly identified by the data identification unit 222b, thereby generating a noise attenuation model.

Incidentally, the weighting parameter a ₁ of Learning of formula (3) may be a constant, it may be determined by trial and error. Further, for example, when the accuracy of the neural network identification of the data identification unit 222b is high, the output of the data conversion unit 222a is greatly changed. It is good also as a formula which makes a loss function a variable.

As described above, the learning unit 22 sets the neural network of the data identification unit 222b based on the loss function of the equation (1) and the neural network of the data conversion unit 222a based on the loss function of the equation (3). The process of updating using the back propagation method is repeated a fixed number of times or until the loss function value does not improve, thereby completing the learning. Note that the neural network of the data conversion unit 222a may perform pre-learning using Equation (2) before learning using Equation (3).

Next, the diagnostic model learning unit 223 makes the machine learning model learn (step S02). That is, the machine learning model of the abnormality cause identification unit 212 is learned as a multi-class classifier using the converted data (attenuated data) of the data conversion unit 222a trained in step S01 as an input variable and the abnormality cause type as an output variable. By doing so, a diagnostic model is generated. For this learning, only the simulation data among the learning data in the learning data storage unit 200 is used. That is, the input variable to the machine learning model of the abnormality cause identification unit 212 is data (attenuated data) obtained by converting the simulation data of the learning data storage unit 200 by the learned data conversion unit 222a. After learning only the simulation data, it is confirmed that the learning data in the learning data storage unit 200 can be correctly identified using actual data.

When used neural network as a machine learning model, m the abnormality cause type, one-hot representation of true label indicating the abnormality cause type of v _m (only the index which is a true label is 1 and the other is a 0), y _d2m Is the output value of the m-th output node by the abnormality cause identification unit 212, the diagnostic model learning unit 223 calculates Expression (4) indicating the cross entropy error as a loss function, and uses the error backpropagation method. the output y _d2m of abnormality cause identifying unit 212 so as to approach the true label v _m, updates the weight with the neural network. The diagnostic model learning unit 223 repeats this update a fixed number of times or until the loss function value does not improve, thereby completing the learning. Thereby, a diagnostic model is generated.

Then, the data to be diagnosed is diagnosed by the data processing unit 211 and the abnormality cause identification unit 212 (step S03). That is, the data processing unit 211 acquires the diagnosis target data from the data acquisition unit 100, converts the acquired data by the noise attenuation model generated by the conversion model learning unit 222c, and converts the acquired data as preprocessed data into the abnormality cause identification unit 212. Output to The abnormal cause identification unit 212 acquires the preprocessed data from the data processing unit 211, and outputs an abnormal cause type.

(5) The display control unit 23 acquires the abnormality cause type output from the abnormality cause identification unit 212 and causes the display unit 5 to display the abnormality cause type as a diagnosis result as shown in FIG. 9 (step S04).

(Action / Effect)
(1) The abnormality diagnosis system 1 of the present embodiment is an abnormality diagnosis system for diagnosing an abnormality cause of a diagnosis target. The diagnosis unit 21 for identifying a type of the abnormality cause of the diagnosis target based on a model, and a machine And a learning unit 22 that is generated by learning. The diagnostic unit 21 is a characteristic part other than the characteristic part that enables identification of the type of the cause of the abnormality from the data when the data output from the diagnosis target is input. Based on a data processing unit 211 that generates pre-processed data in which noise has been attenuated and a diagnostic model that outputs the type of an abnormal cause of the diagnostic target when the pre-processed data is input, the type of the abnormal cause of the diagnostic target is determined. The diagnostic model is a machine learning model. The learning unit 22 includes a diagnostic model learning unit 223 that generates a diagnostic model by machine learning. A noise attenuator 221 that generates attenuated data from which the noise has been removed from learning data for machine learning of the model, wherein the learning data is simulation data simulating an abnormality of a diagnosis target or a diagnosis target. The diagnostic model learning unit 223 generates the diagnostic model by machine learning using the attenuated data as learning data and the abnormality cause type as teacher data.

Thus, the diagnostic model is generated by using the noise attenuating unit 221 as input, attenuated data obtained by attenuating noise, which is a characteristic part other than the characteristic part that enables the type of the cause of the abnormality to be identified from the learning data. In addition, the type of abnormality cause to be diagnosed can be accurately diagnosed. That is, since the simulation data and the actual data include a characteristic portion that allows the type of abnormality to be identified and a noise portion that is a characteristic other than the characteristic portion, the noise portion is attenuated by the noise attenuator 221. Thus, the diagnostic model learning unit 223 allows the machine learning model to learn characteristic parts of the simulated data and the actual data that can identify the cause of the abnormality, and generate a diagnostic model. When identifying the cause of the abnormality using the diagnosis model, the data processing unit 211 attenuates a noise portion that is a feature other than a feature that allows the cause of the abnormality to be identified from the data to be diagnosed. A characteristic portion that makes the type identifiable is left, and is easily diagnosed by the abnormality cause identification unit 212. As a result, it is possible to accurately diagnose the type of abnormality cause to be diagnosed.

(2) The noise attenuation unit 221 generates attenuated data based on a noise attenuation model that removes noise from input data, and the learning unit 22 includes a noise attenuation model generation unit 222 that generates a noise attenuation model. I did it. Then, based on the conversion model that converts the learning data, the noise attenuation model generation unit 222 converts the learning data and outputs the converted data. The data conversion unit 222a receives the converted data. Based on an identification model that identifies the data type of the learning data, a data identification unit 222b that identifies a data type indicating whether the learning data is simulated data or real data, and a conversion model that is updated by machine learning to perform learning. Model learning unit 222c that generates a noise attenuation model that outputs attenuated data when input data is input, and an identification model learning unit 222d that updates an identification model by machine learning. Is a neural network, and the discrimination model learning unit 222d converts the discrimination model from the converted data to the learning data. The conversion model learning unit 222c generates an identification model so as to be a model that correctly identifies the data type. The conversion model learning unit 222c converts the conversion model into a model similar to the learning data, but converts the conversion model so that the data identification unit 222b cannot correctly identify it. By updating the data to be output, a noise attenuation model is generated, and the data processing unit 211 performs preprocessing on the data output from the diagnosis target based on the noise attenuation model generated by the conversion model learning unit 222c. Generated data.

Accordingly, the noise attenuation unit 221 is generated by updating the conversion model to be similar to the learning data but outputting the converted data so that the data identification unit 222b cannot correctly identify the noise attenuation model. Thus, the specific noise of the simulation data or the actual data is attenuated or removed from the learning data, and the data is converted into intermediate data that cannot be identified as the simulation data or the actual data by the data identification unit 222b. For example, if the simulation data and the actual data each include a specific noise, the data identification unit 222d can identify the data type by focusing on the noise portion such as an offset in the above example. The conversion model is updated so as to output converted data (for example, data having the same waveform but an offset value different from both the simulation data and the actual data) that cannot be identified as the simulation data or the actual data by the data identification unit 222b, and the noise attenuation model Is generated, intermediate data obtained by attenuating or removing noise peculiar to the simulation data and the real data can be obtained. That is, the characteristic part necessary for identifying the original abnormality cause type is left in the intermediate data.

Since the diagnostic model learning unit 223 learns the machine learning model and generates a diagnostic model using such intermediate data, it is possible to accurately identify the cause of the abnormality regardless of whether the original learning data is simulated data or real data. In addition to this, it is possible to identify the cause of the abnormality with high accuracy even for new diagnosis target data.

(3) The conversion model learning unit 222c updates the conversion model such that the data type of the simulated data or the actual data output from the data identification unit 222b is inverted from the data type of the input data input to the data conversion unit 222a. I did it.

This makes it possible to generate intermediate data that cannot be distinguished from real data and simulation data. That is, if the conversion model learning unit 222c only trains the conversion model to improve the error between the input data and the output data (the first term on the rightmost side of Expression (2) or Expression (3)) Although only the output data is learned so as to approach the input data, in the present embodiment, the data type of the output data of the conversion model is further input to the data conversion unit 222a by the conversion model learning unit 222c. Since the learning is performed so that the data type becomes the data type inverted from the data type of the input data (corresponding to the second term on the rightmost side of Expression (3)), the data output from the noise attenuation model is correctly recognized by the data identification unit 222b. It can be unidentifiable data. This makes it possible to generate intermediate data that cannot be distinguished from real data and simulation data.

(4) The display control unit 23 for displaying the identification result of the abnormality cause identification unit 212 on the display unit is provided. As a result, the user can obtain a result of identifying the cause of the abnormality displayed on the display unit 5, and can cope with the abnormality to be diagnosed.

(Second embodiment)
(Constitution)
A second embodiment will be described with reference to FIG. The second embodiment is the same as the basic configuration of the first embodiment. Hereinafter, only different points from the first embodiment will be described, and the same parts as those in the first embodiment will be denoted by the same reference numerals and detailed description thereof will be omitted.

FIG. 10 is a diagram illustrating the configuration of the abnormality diagnosis system according to the second embodiment. In the present embodiment, the input unit 4 receives a user's input of parameters related to learning by the learning unit 22. Parameters relating to the learning, the data conversion section 222a and a data discrimination unit 222b of the neural network, and the network structure of the machine learning model abnormality cause identifying unit 212, the number of times of learning, the formula (3) in the weight parameter a _1, etc. included is there. The parameters related to the learning are used for learning in the learning unit 22.

The display control unit 23 causes the display unit 5 to display the parameter reset button PR as shown in FIG. The parameter reset button PR is a button for adjusting parameters and causing the learning unit 22 to execute learning.

Further, as shown in FIG. 11, the display control unit 23 learns the data conversion unit 222a, the neural network of the data identification unit 222b, and the learning of the machine learning model of the abnormality cause identification unit 212 by the learning unit 22, and the noise attenuation model and the diagnostic model. Is generated, the display unit 5 displays on the display unit 5 the type of abnormality cause of the learning data, the correct answer rate of the learning data based on the diagnostic model, the data before and after noise attenuation by the noise attenuator 221 and the data number of the data, or the diagnosis based on the diagnostic model Display the result. This correct answer rate can be calculated by (number of diagnostic models matching the abnormality cause type indicated by the learning data / number of learning data) × 100. Here, the processing unit 2 has a correct answer rate calculating unit 24 including a CPU, and the correct answer rate calculating unit 24 calculates the correct answer rate. The correct answer rate is also referred to as a failure cause identification rate in the present specification or the drawings.

診断 A diagnosis result by the diagnosis model is also referred to as a determination result in the present specification or the drawings. The noise-attenuated data is equal to the converted data of the data conversion unit 222a after learning by the conversion model learning unit 222c, and the data before noise attenuation is the data conversion unit 222a after learning by the conversion model learning unit 222c. Is equal to the data before conversion, that is, the learning data. Therefore, the data before and after the noise attenuation by the noise attenuator 221 is also referred to as data before and after the conversion by the data converter 222a after learning (hereinafter, also simply referred to as “data before and after the conversion”) in this specification or the drawings.

In order to perform the display, the display control unit 23 uses the learning data from the learning data storage unit 200, the data after conversion from the data conversion unit 222a after learning, and the identification result from the abnormality cause identification unit 212 after learning. Acquires each type of abnormality cause.

(4) The display control unit 23 causes the display unit 5 to display the left and right buttons LR and the abnormality diagnosis start button S as shown in FIG. The left and right buttons LR switch the display of data before and after conversion. That is, when the left button is pressed once, the data before and after the conversion with the data number smaller by one is displayed, and when the right button is pressed once, the data before and after the conversion with the data number larger by one is displayed. The abnormality diagnosis start button S is a button for starting identification of an abnormality cause type. The abnormality diagnosis start button S is a failure diagnosis start button S when the abnormality is a failure.

(4) Each display on the display unit 5 by the display control unit 23 is performed before diagnosing the cause of the abnormality with respect to the diagnosis target data.

FIG. 12 is an operation flowchart illustrating an example of the operation of the abnormality diagnosis system 1 according to the second embodiment. Note that the description of the same operation as that of the first embodiment will be appropriately omitted.

As shown in FIG. 12, first, input of a parameter related to learning by the input unit 4 is received, and the parameter is set (step S11). Thereafter, the learning unit 22 causes the neural network of the data conversion unit 222a and the data identification unit 222b to learn (step S01), and causes the abnormality cause identification unit 212 to learn the machine learning model (step S02).

(4) After learning of each neural network and the machine learning model, the display control unit 23 displays a confirmation screen on the display unit 5 (step S12). That is, the correct answer rate, the data before and after the conversion and the data numbers thereof, the abnormality cause type of the learning data, and the identification result of the abnormality cause identification unit 212 are displayed. When the user confirms the display on the display unit 5 by pressing the left and right buttons LR or the like and resets the parameters (YES in step S13), the user returns to step S11 by pressing the parameter reset button PR, and returns to step S11. If the resetting is not necessary (NO in step S13), the diagnosis target data is diagnosed by the user pressing the abnormality diagnosis start button S (step S03), and the diagnosis result is displayed on the display unit 5 by the display control unit 23. (Step S04).

(Action / Effect)
In the present embodiment, after the learning of the data conversion unit 222a and the abnormality cause identification unit 212 by the learning unit 22, the display control unit 23 displays a diagnosis on the learning data on the display unit 5 before diagnosing the abnormality cause to be diagnosed. The correct answer rate of the model or the data before and after the conversion by the data conversion unit 222a is displayed.

This allows the user to confirm the validity of the noise attenuation model and the diagnostic model before diagnosing the cause of the abnormality. For example, referring to the example of FIG. 11, in the learning data of the waveform, a peak portion is a characteristic portion indicating an abnormality, and the waveform data before conversion is biased downward so that the signal intensity is reduced as a whole. Assuming that noise is included, the converted waveform data has the signal intensity increased as a whole, the noise is attenuated, and the peak portion of the characteristic portion remains, and the processing of the noise attenuation model is appropriate. It can be confirmed that the process has been performed.

As a first modification, as illustrated in FIG. 13, the display control unit 23 includes a correct answer rate, individual data information including a data number, data before and after conversion, an abnormal cause type of learning data, and an abnormal cause identification result. May be displayed on the display unit 5 side by side for the simulation data and the actual data. Thereby, the user can confirm the validity of the diagnostic model while comparing the validity of the learning using the simulated data with the validity of the learning using the actual data.

As a second modification, as shown in FIG. 14, the display control unit 23 causes the display unit 5 to display the correct answer rate and the individual data information side by side with the simulation data, the actual data, and the abnormality cause type. May be. As a result, the user can confirm the validity of the machine learning model by comparing the validity of the learning using the simulated data and the validity of the learning using the actual data for each abnormality cause type. A high degree of abnormality diagnosis can be performed.

As a third modification, as shown in FIG. 15, the display control unit 23 causes the display unit 5 to display the identification result of the abnormality cause identification unit 212 and the data before and after the conversion by the data conversion unit 222a in step S04. You may do it. Thereby, it is possible to confirm what kind of conversion has been performed by the data conversion unit 222a. That is, since the learning data of the abnormality cause identification unit 212 and the data on which the learning data is based can be confirmed, the reliability of the diagnosis result can be verified by the user.

(Third embodiment)
(Constitution)
A third embodiment will be described. The third embodiment is the same as the basic configuration of the first embodiment or the second embodiment. Hereinafter, only differences from the second embodiment will be described, and the same parts as those in the second embodiment will be denoted by the same reference numerals and detailed description thereof will be omitted.

機械 The machine learning model of the abnormality cause identification unit 212 of the present embodiment is a neural network. The learning unit 22 simultaneously learns the neural networks of the data conversion unit 222a, the data identification unit 222b, and the abnormality cause identification unit 212. As illustrated in FIG. 16, the learning unit 22 configures the network such that an output of the neural network of the data conversion unit 222a is an input of the data identification unit 222b and the abnormality cause identification unit 212. Update weights simultaneously.

Specifically, the data conversion unit 222a acquires learning data from the learning data storage unit 200, performs data conversion, and outputs the converted data to the data identification unit 222b and the abnormality cause identification unit 212. Then, the data identification unit 222b identifies the data type of the converted data, and the abnormality cause identification unit 212 identifies the abnormality cause type from the converted data. The learning unit 22 receives the identification result of the data identification unit 222b and the identification result of the abnormality cause identification unit 212, and updates the weight of the neural network of each of the

units

211, 212, and 23 using the backpropagation method.

In other words, the diagnostic model learning unit 223 sets the neural network of the abnormal cause identifying unit 212 to the abnormal cause type that is the identification result of the abnormal cause identifying unit 212 and the abnormal cause type of the learning data from which the identification result is based. In order to improve the error, the error is back-propagated, and the identification model learning unit 222d converts the neural network of the data identification unit 222b into a data type that is the identification result of the data identification unit 222b, The data is updated using the backpropagation method so that the error between the original learning data and the data type is improved.

More specifically, the diagnostic model learning unit 223 trains the neural network of the abnormality cause identification unit 212 using Expression (4) as a loss function, and the identification model learning unit 222d calculates Expression (1) as a loss function. To learn the neural network of the data identification unit 222b. The learning of the abnormality cause identification unit 212 can be performed using only the simulation data among the learning data of the learning data storage unit 200.

Further, the conversion model learning unit 222c sets the neural network of the data conversion unit 222a to data which is similar to the input data, but which cannot be correctly identified by the data identification unit 222b and which can be identified correctly by the abnormality cause identification unit 212. The data conversion unit 222a is trained to output. In other words, the conversion model learning unit 222c uses the neural network of the data conversion unit 222a to determine the error between the learning data and the data after conversion by the data conversion unit 222a, the data type output by the data identification unit 222b, and the learning that is the correct answer. The error is updated so that the error between the data type obtained by inverting the data type of the use data and the error between the error cause type output from the error cause identification unit 212 and the error cause type of the learning data as the correct answer are improved. .

More specifically, the conversion model learning unit 222c repeats the process of updating using the error back propagation method a fixed number of times, or until the loss function value does not improve, based on Equation (5) as the loss function, and performs learning. .

Note that the weighting parameters a ₁ and a ₂ relating to the learning of Expression (5) may be constants or may be determined by trial and error. For example, a ₁ of the neural network of the data identification unit 222b may be used for a1. When the identification accuracy is high, in order to greatly change the output of the data conversion unit 222a, an expression using a loss function such as the identification rate (correct answer rate) of the neural network of the data identification unit 222b or equation (1) as a variable may be used. For the varying increase the output of the data converter 222a when the neural network identification accuracy of the abnormality cause identifying unit 212 for a ₂ is low, the identification rate of the neural network (percentage of correct answers) and loss function, such as equation (4) May be a mathematical expression having as a variable.

FIG. 17 is an operation flowchart showing an example of the operation of the abnormality diagnosis system 1 of the present embodiment based on the first embodiment. FIG. 18 is an operation flowchart illustrating an example of an operation of the abnormality diagnosis system 1 according to the present embodiment based on the second embodiment. Note that the description of the same operation as that of the first embodiment and the second embodiment will be appropriately omitted.

In the operation of the present embodiment, as shown in FIGS. 17 and 18, instead of steps S01 and S02 of FIGS. 7 and 12, the learning unit 22 executes a neural network of the

units

222a, 222b, and 212 as step S1a. Let them learn.

(Action / Effect)
In the present embodiment, the conversion model learning unit 222c converts the conversion model into an error between the learning data and the converted data, an error between the output result of the data identification unit 222b and the incorrect data type, and an abnormality cause identification. The noise attenuation model is generated by updating so that the error between the output result of the unit 212 and the correct abnormality cause type becomes small.

(4) This makes it possible to emphasize features necessary for identifying the cause of the abnormality in the data that has been attenuated by the noise attenuator 221, and to improve the accuracy of identifying the cause of the abnormality even with the same number of learnings. In addition, the accuracy of abnormality cause identification can be improved with a small number of learnings.

That is, as shown in FIG. 19, in the learning of the neural network by the data conversion unit 222a, the conversion model learning unit 222c outputs the error between the learning data and the data after the conversion by the data conversion unit 222a, and outputs the data identification unit 222b. Not only the error between the data type obtained and the data type obtained by inverting the data type of the learning data to be the correct answer, but also the error cause type output by the error cause identifying unit 212, and the error cause type of the learning data to be the correct answer Is improved, the neural network of the data conversion unit 222a is trained so that the output result of the abnormality cause identification unit 212 is correct, so that the features necessary for identifying the abnormality cause are emphasized. The converted data can be output by the data conversion unit 222a. It is possible to improve the degree.

(Fourth embodiment)
(Constitution)
A fourth embodiment will be described. The fourth embodiment is the same as the basic configuration of the first embodiment, the second embodiment, or the third embodiment. Hereinafter, only different points from the third embodiment will be described, and the same portions as those in the third embodiment will be denoted by the same reference numerals and detailed description thereof will be omitted.

(4) The learning unit 22 according to the present embodiment generates two or more diagnostic models of the abnormality cause identification unit 212. For example, the diagnostic model according to the first or second embodiment (hereinafter, also referred to as a first model) and the diagnostic model according to the third embodiment (hereinafter, also referred to as a second model) are respectively used. Respectively.

{Circle around (1)} The correct answer rate calculating unit 24 calculates the correct answer rate of the abnormal cause identifying unit 212 based on the first model and the correct answer rate of the abnormal cause identifying unit 212 based on the second model. As described above, this correct answer rate can be calculated by (the number of the identification results of the abnormality cause identification unit 212 after learning that matches the abnormality cause type indicated by the learning data / the number of learning data) × 100.

(4) The abnormality cause identification unit 212 receives the correct answer rate of each model from the correct answer rate calculation unit 24, and sets a model having high diagnostic accuracy among the models as its own diagnostic model. That is, a diagnostic model that maximizes the correct answer rate is adopted.

FIG. 20 is an operation flowchart illustrating an example of the operation of the abnormality diagnosis system 1 according to the fourth embodiment. Note that the description of the same operation as that of the first, second, and third embodiments will be appropriately omitted.

As shown in FIG. 20, the learning unit 22 first learns the neural network of the data conversion unit 222a and the data identification unit 222b (step S01), and learns the machine learning model of the abnormality cause identification unit 212 (step S02). . Next, the learning unit 22 causes the neural networks of the

units

222a, 222b, and 223 to learn (step S1a). Steps S01 and S02 and step S1a may be performed simultaneously.

Next, the correct answer rate calculation unit 24 calculates the correct answer rate of the abnormality cause identification unit 212 based on each machine learning model (step S21). The abnormal cause identification unit 212 acquires the correct answer rate of each diagnostic model from the correct answer rate calculation unit 24, selects the model with the highest correct answer rate, and sets the model as the diagnostic model for identifying the type of the abnormal cause (step S22). Then, the diagnosis target data is diagnosed by the abnormality cause identification unit 212 (step S03), and the diagnosis result is displayed on the display unit 5 by the display control unit 23 (step S04).

(Action / Effect)
In the present embodiment, the diagnostic model learning unit 223 generates two or more diagnostic models, and the diagnostic model of the abnormality cause identifying unit 212 is a model with high diagnostic accuracy from the two or more generated diagnostic models.

Thereby, the diagnostic accuracy can be improved. Which diagnostic model of the abnormality cause identification unit 212 has the higher identification accuracy depends on various parameters such as the number of times of learning, and cannot be known unless actually verified. Since the verification is performed by the abnormality cause identification unit 212, the accuracy of diagnosis can be improved.

(Fifth embodiment)
(Constitution)
A fifth embodiment will be described. The fifth embodiment is the same as the basic configuration of the fourth embodiment. Hereinafter, only different points from the fourth embodiment will be described, and the same portions as those in the fourth embodiment will be denoted by the same reference numerals and detailed description thereof will be omitted.

In the present embodiment, as shown in FIG. 21, the display control unit 23 determines, for the diagnostic model generated by the diagnostic model learning unit 223, the correct answer rate of the diagnostic model and the data (before and after the attenuation of the noise attenuation unit 221). (Data before and after the conversion) are displayed on the display unit 5. The display control unit 23 acquires the correct answer rate from, for example, the correct answer rate calculating unit 24. Further, the input unit 4 accepts a user's selection of the machine learning model displayed on the display unit 5. The display control unit 23 causes the display unit 5 to display a selection unit SL that receives the selection.

FIG. 22 is an operation flowchart illustrating an example of the operation of the abnormality diagnosis system 1 according to the fifth embodiment. The description of the same operation as the fourth operation will be appropriately omitted.

In the present embodiment, as shown in FIG. 22, instead of step S22 in FIG. 20, for each diagnostic model generated by the diagnostic model learning unit 223, the display control unit 23 calculates the correct answer rate and the data before and after the conversion. Is displayed on the display unit 5 (step S31). Then, the input unit 4 accepts the selection of the diagnostic model by the user (step S32), sets the diagnostic model of the abnormality cause identifying unit 212 as the selected diagnostic model, and diagnoses the diagnosis target data by the abnormal cause identifying unit 212. (Step S03), and the display control unit 23 causes the display unit 5 to display the diagnosis result (Step S04).

(Action / Effect)
In the present embodiment, the diagnostic model learning unit 223 generates two or more diagnostic models, and the display control unit 23 displays, on the display unit 5, the correct answer rate of the diagnostic model and before and after attenuation by the noise attenuation unit 221 for each diagnostic model. Displayed with data. This allows the user to examine the validity of the diagnostic model. Then, a more appropriate diagnostic model can be adopted to diagnose the cause of the abnormality.

(Sixth embodiment)
A sixth embodiment will be described. The sixth embodiment is the same as the basic configuration of any of the first to fifth embodiments. Hereinafter, only different points from the fifth embodiment will be described, and the same portions as those in the fifth embodiment will be denoted by the same reference numerals and detailed description thereof will be omitted.

The input unit 4 of the present embodiment receives an input of a hyperparameter used in the generation of the noise attenuation model in the conversion model learning unit 222c. This hyper parameter is a parameter of the neural network constituting the conversion model, where is the weighting parameter a ₁ of the formula (3) or Formula (5) used in converting the model learning unit 222c. The input unit 4 receives an input of the weighting parameter a ₁ depending on the data type of either the learning data is simulated data or actual data. Transformation model learning unit 222c sets the weighting parameter a ₁ input from the input unit 4 separately.

As described above, in the present embodiment, the conversion model learning unit 222c multiplies the error between the output result of the data identification unit 222b and the correct teacher data by the hyperparameter received by the input unit 4, and generates a noise attenuation model. Was generated.

This makes it possible to adjust whether the converted data of the data conversion unit 222a is closer to the output of the simulation data or the actual data. For example, when noise is included only in the actual data and noise is not included in the simulation data, the weighting parameter for the actual data is set to 0, so that the data conversion unit 222a functions as a noise filter and includes noise. The data type is not identified by the data identification unit 222b to which the data after the actual data is converted by the data conversion unit 222a is input, thereby preventing erroneous learning of noise. That is, the noise included in the actual data does not need to be reflected in the learning of the machine learning model of the abnormality cause identification unit 212. Therefore, the accuracy of identifying the cause of the abnormality can be improved.

Further, since the display control unit 23 displays the data before and after the attenuation of the noise attenuating unit 221 (data before and after the conversion) on the display unit 5, it is possible to visually confirm that noise has been removed from the data before and after the conversion. Therefore, it can be determined that the diagnosis model of the abnormality cause identification unit 212 is identified without being affected by noise, and the reliability of the diagnosis result can be improved.

(Seventh embodiment)
(Constitution)
A seventh embodiment will be described. The seventh embodiment is the same as the basic configuration of the sixth embodiment. Hereinafter, only different points from the sixth embodiment will be described, and the same parts as those in the sixth embodiment will be denoted by the same reference numerals and detailed description thereof will be omitted.

FIG. 23 is a diagram showing the configuration of the abnormality diagnosis system according to the seventh embodiment. The abnormality diagnosis system 1 according to the present embodiment includes an adjustment unit 25. The adjustment unit 25 includes a CPU, and adjusts a hyper parameter used for generating a noise attenuation model. The conversion model learning unit 222c generates a noise attenuation model by updating the conversion model using the hyperparameter adjusted by the adjustment unit 25.

The display control unit 23 displays, on the display unit 5, an adjustment reception image for receiving the adjustment of the adjustment unit 25 and an adjustment result including a correct answer rate of the diagnostic model or data before and after attenuation by the noise attenuation unit 221. It is displayed on the same display screen of the unit 5. As shown in FIG. 24, the adjustment reception image includes, for example, a slide bar 51 and a knob 52 on the slide bar 51, and the user places the knob 52 on the slide bar 51 via the input unit 4 such as a mouse. slide adjusts the weighting parameter a _1. When the position of the knob 52 is x (0 ≦ x ≦ 1), the adjustment unit 25 multiplies the weight parameter a ₁ when the learning data is the simulation data by x, and the weight parameter a ₁ when the learning data is the real data. Let ₁ be (1-x) times.

FIG. 25 is an operation flowchart illustrating an example of the operation of the abnormality diagnosis system 1 according to the seventh embodiment. Note that the description of the same operation as that of the sixth embodiment will be appropriately omitted.

As shown in FIG. 25, the adjustment unit 25 adjusts the weighting parameters _{a 1} (step S41). Learning unit 22, each unit using the weight parameters _{a 1} after adjustment 222a, train the neural network 222b (step S42), further, after completion of the learning, the converted data in the data conversion unit 222a after learning, That is, the diagnostic model is generated by learning the machine learning model of the abnormality cause identification unit 212 using the attenuated data of the noise attenuation unit 221 (step S43).

Then, the display control unit 23 displays, on the same display screen as the screen on which the adjustment reception screen of the display unit 5 is displayed, the correct answer rate of the diagnostic model or the data before and after the attenuation by the noise attenuation unit 221 (by the data conversion unit 222a). The adjustment result including the data before and after the conversion is displayed (step S44).

Next, if the abnormality diagnosis start button S displayed on the display unit 5 is not pressed by the input unit 4 (NO in step S45), that is, if the user determines that the adjustment result is not appropriate, without starting a diagnosis, the process returns to step S41, adjusting the weighting parameter _{a 1.} On the other hand, when the abnormality diagnosis start button S displayed on the display unit 5 is pressed by the input unit 4 (YES in step S45), that is, when the adjustment result is determined to be valid by the user, the abnormality diagnosis is started. Then, the diagnosis target data is diagnosed by the abnormality cause identification unit 212 (step S03), and the diagnosis result is displayed on the display unit 5 by the display control unit 23 (step S04).

As shown in FIG. 26, instead of the steps S42 and S43 in FIG. 25, similarly to the third embodiment, the learning unit 22, each unit using the weight parameters _{a 1} after adjustment by the adjustment portion 25 222a , 222b, and 223 (step S45a).

(Action / Effect)
Abnormality diagnosis system 1 of this embodiment includes a controller 25 for adjusting the weighting parameters a _1, transformation model learning unit 222c includes a diagnostic model by updating the conversion model using the weighting parameters a _1, which is adjusted Generated. Then, the display control unit 23 outputs the adjustment reception image for receiving the adjustment of the adjustment unit 23 and the correct answer rate of the diagnostic model or the adjustment result including the data before and after the attenuation by the noise attenuation unit 221 to the display unit 5. Display on the same display screen.

Accordingly, while viewing the adjustment result, users because it can adjust the weighting parameters a _1, thereby improving convenience. That is, complicated operations to adjust the screen for adjusting the weighting parameters a _1, a weighting parameter a ₁ by switching the display screen of the data before and after the adjustment result correct rate and an attenuation (data before and after conversion) You don't have to.

As a first modification of the present embodiment, as shown in FIG. 27, the display control unit 23 determines the correct answer rate, the data number, the data before and after the attenuation (data before and after the conversion), the abnormality cause type of the learning data, and the abnormality cause. The individual data information including the identification result may be arranged for each of the simulation data and the actual data, and may be displayed on the display screen of the display unit 5 on which the adjustment reception image is displayed. Thereby, the user can confirm the validity of the diagnostic model while comparing the validity of the learning using the simulated data with the validity of the learning using the actual data.

As a second modification, as shown in FIG. 28, the display control unit 23 arranges the correct answer rate and the individual data information for each of the simulation data, the actual data, and the abnormality cause type, and adjusts the adjustment reception image of the display unit 5. May be displayed on a display screen on which is displayed. As a result, the user can confirm the validity of the diagnosis model by comparing the validity of learning using the simulated data and the validity of learning using the actual data for each type of abnormality cause, and can improve the reliability. Abnormal diagnosis can be performed.

(Other embodiments)
In the present specification, a plurality of embodiments according to the present invention have been described. However, these embodiments are presented as examples and are not intended to limit the scope of the invention. The embodiments described above can be implemented in other various forms, and various omissions, replacements, and changes can be made without departing from the scope of the invention. These embodiments and modifications thereof are included in the scope and gist of the invention, and are also included in the invention described in the claims and their equivalents.

In the first to seventh embodiments, the abnormality diagnosis system 1 includes the display unit 5, but the display unit 5 does not necessarily have to include the display unit 5. For example, the abnormality diagnosis system 1 outputs an identification result of the abnormality cause identification unit 212, a correct answer rate, data before and after conversion by the data conversion unit 222a, and the like in response to a request from the outside, and causes the external display device to display the data. You may do it. Such an abnormality diagnosis system 1 is, for example, a single server or a server configured by a computer.

1 abnormality diagnosis system 2 processing unit 21 diagnosis unit 211 data processing unit 212 abnormality cause identification unit 22 learning unit 221 noise attenuation unit 222 noise attenuation model generation unit 222a data conversion unit 222b data identification unit 222c conversion model learning unit 222d identification model learning unit 223 diagnostic model learning unit 23 display control unit 24 correct answer rate calculation unit 25 adjustment unit 3 storage unit 4 input unit 5 display unit 51 slide bar 52 knob 100 data acquisition unit 200 learning data storage unit

Claims

An abnormality diagnosis system for diagnosing an abnormality cause to be diagnosed,
A diagnosis unit that identifies a type of an abnormal cause of the diagnosis target based on a model,
A learning unit that generates the model by machine learning,
With
The diagnostic unit includes:
When data output from the diagnosis target is input, a data processing unit that generates preprocessed data in which noise, which is a characteristic part other than a characteristic part that enables the type of the cause of the abnormality to be identified from the data, is attenuated. When,
An abnormal cause identification unit that identifies the type of the abnormal cause of the diagnostic target based on a diagnostic model that outputs the abnormal cause type of the diagnostic target when the preprocessed data is input,
Has,
The diagnostic model is a machine learning model,
The learning unit includes:
A diagnostic model learning unit that generates the diagnostic model by machine learning,
A noise attenuator that generates attenuated data in which the noise is attenuated from learning data for machine learning of the diagnostic model,
Has,
The learning data is simulated data simulating the abnormality of the diagnosis target or actual data indicating the abnormality of the diagnosis target,
The diagnostic model learning unit,
The diagnostic data is generated by machine learning using the attenuated data as learning data and the abnormality cause type as teacher data,
Abnormal diagnosis system.
The noise attenuator generates the attenuated data based on a noise attenuation model that removes the noise from the input data,
The learning unit includes:
Having a noise attenuation model generation unit that generates the noise attenuation model,
The noise attenuation model generator,
A data conversion unit that converts the learning data and outputs converted data based on a conversion model that converts the learning data;
A data identification unit that identifies a data type indicating whether the learning data is the simulated data or the actual data, based on an identification model that identifies a data type of the learning data when the converted data is input; ,
By updating the conversion model by machine learning, a conversion model learning unit that generates a noise attenuation model that outputs the attenuated data when the learning data is input,
An identification model learning unit that updates the identification model by machine learning,
Has,
The identification model and the conversion model are neural networks,
The identification model learning unit,
Generate the identification model so that the identification model is a model that correctly identifies the data type of the learning data from the converted data,
The conversion model learning unit,
The noise reduction model is generated by updating the conversion model to be similar to the learning data but outputting the converted data so that the data identification unit cannot correctly identify the data.
The data processing unit includes:
Based on the noise attenuation model generated by the conversion model learning unit, to generate the pre-processed data from data output from the diagnosis target,
The abnormality diagnosis system according to claim 1.
The conversion model learning unit updates the conversion model so that the data type output by the data identification unit is inverted from the data type of the learning data.
The abnormality diagnosis system according to claim 2.
The conversion model learning unit,
The conversion model,
An error between the learning data and the converted data, and
Generating the noise attenuation model by updating the output result of the data identification unit and the error between the data type that is incorrect, so as to reduce the error,
The abnormality diagnosis system according to claim 2 or 3.
The conversion model learning unit,
The conversion model,
An error between the learning data and the converted data,
An error between the output result of the data identification unit and the data type that is incorrect, and
The noise attenuation model is generated by updating the output result of the abnormality cause identification unit and the error between the abnormality cause type that is the correct answer so as to be small,
An abnormality diagnosis system according to any one of claims 2 to 4.
The diagnostic model learning unit generates two or more diagnostic models,
The diagnostic model of the abnormality cause identification unit is a model with good diagnostic accuracy from the two or more generated diagnostic models.
An abnormality diagnosis system according to any one of claims 2 to 5.
An input unit that receives an input of a hyper parameter used in generating the noise attenuation model,
The hyper parameter is a parameter corresponding to a data type of the learning data, the simulation data or the actual data,
The conversion model learning unit,
The hyperparameter is multiplied by the error between the output result of the data identification unit and the correct teacher data to generate the noise attenuation model,
The diagnostic system according to any one of claims 2 to 6.
A display control unit that displays an identification result of the abnormality cause identification unit on a display unit,
An abnormality diagnosis system according to any one of claims 2 to 7.
The display control unit, before diagnosing the cause of the abnormality of the diagnosis target, on the display unit, the correct answer rate of the diagnostic model, or to display data before and after attenuation by the noise attenuation unit,
An abnormality diagnosis system according to claim 8.
The display control unit causes the display unit to display the data before and after the attenuation by the identification result and the noise attenuation unit,
The abnormality diagnosis system according to claim 8.
The diagnostic model learning unit generates two or more diagnostic models,
The display control unit causes the display unit to display, for each of the diagnostic models, a correct answer rate of the diagnostic model and data before and after attenuation by the noise attenuation unit.
The abnormality diagnosis system according to any one of claims 8 to 10.
An adjustment unit that adjusts a hyper parameter used in generating the noise attenuation model,
The conversion model learning unit generates a diagnostic model by updating the conversion model using the adjusted hyperparameter,
The display control unit,
An adjustment reception image for receiving the adjustment of the adjustment unit,
Correction rate of the diagnostic model, or adjustment results including data before and after attenuation by the noise attenuation unit,
Displaying on the same display screen of the display unit,
An abnormality diagnosis system according to any one of claims 8 to 11.
An abnormality diagnosis method for diagnosing an abnormality cause to be diagnosed,
A diagnosis step of identifying a type of an abnormal cause of the diagnosis target based on a model,
A learning step of generating the model by machine learning,
With
The diagnosis step includes:
A data processing step of, when data output from the diagnosis target is input, generating pre-processed data in which noise, which is a characteristic portion other than a characteristic portion that enables the type of the cause of the abnormality to be identified from the data, is attenuated; When,
An abnormal cause identification step of identifying the type of the abnormal cause of the diagnostic target based on a diagnostic model that outputs the abnormal cause type of the diagnostic target when the preprocessed data is input,
Has,
The diagnostic model is a machine learning model,
The learning step includes:
Diagnostic model learning step of generating the diagnostic model by machine learning,
A noise attenuation step of generating attenuated data in which the noise is attenuated from learning data for machine learning of the diagnostic model,
Has,
The learning data is simulated data simulating the abnormality of the diagnosis target or actual data indicating the abnormality of the diagnosis target,
The diagnostic model learning step includes:
The diagnostic data is generated by machine learning using the attenuated data as learning data and the abnormality cause type as teacher data,
Abnormal diagnosis method.
An abnormality diagnosis program for diagnosing the cause of the abnormality to be diagnosed,
On the computer,
A diagnosis step of identifying a type of an abnormal cause of the diagnosis target based on a model,
A learning step of generating the model by machine learning,
And execute
The diagnosis step includes:
A data processing step of, when data output from the diagnosis target is input, generating pre-processed data in which noise, which is a characteristic portion other than a characteristic portion that enables the type of the cause of the abnormality to be identified from the data, is attenuated; When,
An abnormal cause identification step of identifying the type of the abnormal cause of the diagnostic target based on a diagnostic model that outputs the abnormal cause type of the diagnostic target when the preprocessed data is input,
Has,
The diagnostic model is a machine learning model,
The learning step includes:
Diagnostic model learning step of generating the diagnostic model by machine learning,
A noise attenuation step of generating attenuated data in which the noise is attenuated from learning data for machine learning of the diagnostic model,
Has,
The learning data is simulated data simulating the abnormality of the diagnosis target or actual data indicating the abnormality of the diagnosis target,
The diagnostic model learning step includes:
The diagnostic data is generated by machine learning using the attenuated data as learning data and the abnormality cause type as teacher data,
Abnormal diagnosis program.