Information processing apparatus, information processing method, and information processing program

Abstract

An information processing apparatus (100) includes a normal / abnormal diagnosis unit (130), a parameter estimation unit (140), and a synthesis unit (160). The normal / abnormal diagnosis unit (130) obtains an object output data and an estimated value of an object physical parameter by inputting, to a normal / abnormal diagnosis model, a feature quantity corresponding to an object input data corresponding to time series data representing a frequency, and determines whether or not an abnormality has occurred in a diagnosis target device based on the object output data. The parameter estimation unit (140) obtains the estimated value of the object physical parameter by inputting, to a parameter estimation model, the feature quantity corresponding to the object input data. The synthesis unit (160) outputs information corresponding to a difference between the estimated value of the object physical parameter based on the normal / abnormal diagnosis model and the estimated value of the object physical parameter based on the parameter estimation model, and a result of determining whether or not an abnormality has occurred in the diagnosis target device.

Description

Technical Field

[0001] This disclosure relates to information processing apparatus, information processing method, and information processing procedure. Background Technology

[0002] Numerous sensors are used in the diagnosis of mechanical anomalies. However, the accuracy of anomaly diagnosis decreases due to deviations in sensor measurements, thus improving the accuracy of anomaly diagnosis has become a research topic.

[0003] Patent document 1 discloses a technology for detecting sensors with a high number of false alarms and false negatives.

[0004] Existing technical documents

[0005] Patent documents

[0006] Patent Document 1: Japanese Patent Application Publication No. 2016-133944 Summary of the Invention

[0007] The problem that the invention aims to solve

[0008] Depending on changes in the characteristics of various physical parameters (such as temperature and operating speed) related to the device being diagnosed, the operating sound and vibration of the device may sometimes change. Furthermore, the characteristics of sensor measurement results vary depending on the changes in the characteristics of these physical parameters, which can sometimes lead to a decrease in the accuracy of anomaly diagnosis. Therefore, in estimating the causes of decreased accuracy in anomaly diagnosis of machinery, it is preferable to consider the characteristics of each physical parameter as well as the operating sound and vibration of the machinery. However, in the technology disclosed in Patent Document 1, the characteristics of each physical parameter and the operating sound and vibration of the device being diagnosed are not considered in estimating the causes of decreased accuracy in anomaly diagnosis of the device being diagnosed.

[0009] The purpose of this disclosure is to take into account the characteristics of each physical parameter corresponding to the physical quantity in the diagnostic object and the mechanical operating noise or vibration in estimating the cause of the reduced accuracy of the abnormal diagnosis of the diagnostic object.

[0010] Methods for solving problems

[0011] The information processing apparatus of this disclosure includes: a normal / abnormal diagnosis unit, which, when setting each physical parameter of a group of physical parameters, each corresponding to a physical quantity in a diagnostic target device, as an object physical parameter, obtains object output data for diagnosing the diagnostic target device and an estimated value of the object physical parameter by inputting a feature quantity corresponding to object input data into a normal / abnormal diagnosis model, and determines whether the diagnostic target device has experienced an abnormality based on the object output data, wherein the object input data is data obtained in the diagnostic target device and corresponds to time series data representing frequency; a parameter estimation unit, which obtains an estimated value of the object physical parameter by inputting a feature quantity corresponding to the object input data into a parameter estimation model corresponding to the object physical parameter; and a synthesis unit, which outputs information corresponding to the difference between the estimated value of the object physical parameter based on the normal / abnormal diagnosis model and the estimated value of the object physical parameter based on the parameter estimation model corresponding to the object physical parameter, and a result determining whether the diagnostic target device has experienced an abnormality.

[0012] Invention Effects

[0013] According to this disclosure, the comprehensive department outputs the following results, which correspond to the difference between the estimated values ​​of the object's physical parameters based on the normal / abnormal diagnostic model and the estimated values ​​of the object's physical parameters based on the parameter estimation model corresponding to the object's physical parameters, as well as the result of determining whether the diagnostic object device has experienced an abnormality. Here, feature quantities corresponding to object input data are input to the normal / abnormal diagnostic model and each parameter estimation model. This object input data is data obtained from the diagnostic object device and corresponds to time series data representing frequency.

[0014] Therefore, according to this disclosure, in estimating the cause of reduced accuracy in the diagnosis of abnormality of the diagnostic device, the characteristics of each physical parameter corresponding to the physical quantity in the diagnostic device, as well as the mechanical operating noise or vibration, can be taken into account. Attached Figure Description

[0015] Figure 1 This is a diagram showing an example of the structure of the information processing device 100 during learning according to Embodiment 1.

[0016] Figure 2 This is a diagram showing an example of the structure of the information processing device 100 for diagnosis according to Embodiment 1.

[0017] Figure 3 This is a diagram illustrating the normal and abnormal diagnosis model of Implementation Method 1.

[0018] Figure 4 This is a diagram illustrating the parameter estimation model of Implementation Method 1.

[0019] Figure 5 This is a diagram illustrating the processing of the integration unit 160 in Embodiment 1.

[0020] Figure 6 This is a diagram illustrating an example of the hardware structure of the information processing apparatus 100 according to Embodiment 1.

[0021] Figure 7 This is a flowchart illustrating the operation of the information processing device 100 during learning in Embodiment 1.

[0022] Figure 8 This is a flowchart illustrating the operation of the information processing device 100 during diagnosis in Embodiment 1.

[0023] Figure 9 This is a diagram illustrating a hardware structure example of the information processing apparatus 100, a variation of Embodiment 1.

[0024] Figure 10 This is a diagram showing an example of the structure of the information processing device 100 for prior data collection in Embodiment 2.

[0025] Figure 11 This is a diagram showing an example of the structure of the information processing device 100 during learning in Embodiment 2.

[0026] Figure 12 This is a flowchart illustrating the operation of the information processing device 100 during the prior data collection in Embodiment 2.

[0027] Figure 13 This is a flowchart illustrating the operation of the information processing device 100 during learning in Embodiment 2. Detailed Implementation

[0028] In the description of the embodiments and the accompanying drawings, the same elements and corresponding elements are labeled with the same reference numerals. The descriptions of elements labeled with the same reference numerals are appropriately omitted or simplified. The arrows in the figures mainly indicate data flow or processing flow. In addition, "part" may be appropriately replaced with "circuit", "process", "step", "processing" or "line".

[0029] Implementation method 1.

[0030] Hereinafter, this embodiment will be described in detail with reference to the accompanying drawings.

[0031] In this embodiment, the subject is the possibility that the operating sound and vibration of a machine may change based on variations in multiple physical parameters (such as temperature and operating speed) related to the machine's operation. The changes in the operating sound and vibration of a machine caused by these multiple physical parameters correspond to domain variations. Furthermore, this embodiment can be applied to both sound data and vibration data related to machinery. However, for the sake of simplicity, sound data and vibration data are sometimes collectively referred to as "sound" or "sound data." Additionally, the term "parameter" sometimes refers to the value of a parameter.

[0032] When collecting learning data, in addition to the sound data obtained from the microphone, data representing various physical quantities such as temperature, voltage, microphone, and distance between devices can also be obtained as physical parameters.

[0033] When collecting evaluation data, it is set to only obtain the microphone's sound data and not to obtain additional data related to each physical parameter.

[0034] When using a normal / abnormality diagnosis model as the generative model to diagnose normal / abnormal conditions under the aforementioned circumstances, misclassification of normal and abnormal conditions may occur due to domain shifts caused by changes in various physical parameters. Therefore, this embodiment provides a method for estimating whether changes in physical parameters are within acceptable limits and which physical parameters have changed.

[0035] Furthermore, to address the problem, the difference between the estimated physical parameters output by the normal / abnormality diagnostic model and the estimated physical parameters output by each specialized model is used. The normal / abnormality diagnostic model is equivalent to an estimator for multiple physical parameters. As specialized models, regression models such as linear regression or symbolic regression are used. Each specialized model is equivalent to an estimator for each physical parameter. Sound data is transformed into the frequency domain using methods such as FFT (Fast Fourier Transform) and used as feature quantities. Then, estimators for multiple physical parameters are constructed using training data.

[0036] When inputting evaluation data, each estimator estimates each physical parameter. Here, if a difference exceeding a threshold occurs in the estimation results between the normal / abnormal diagnostic model and each specialized model, it indicates a change in the physical parameter being evaluated. This threshold can be a pre-set value or a value derived using statistical tests, etc.

[0037] Domain offset is a state where the characteristics of the input data differ from the characteristics of the normal data input during learning. Domain offset can also refer to a state where the characteristics of the physical quantities associated with the input data differ from the characteristics of the physical quantities associated with the normal data input during learning.

[0038] Normal data corresponds to a normal state. Specifically, normal data consists of data obtained under normal conditions and data calculated based on that data. A normal state is a state where no anomalies have been identified.

[0039] When domain shift occurs, data with characteristics different from those of the data input during learning may be received. Therefore, domain shift contributes to reduced diagnostic accuracy in normal anomaly diagnostic models. Specific examples of changes in data characteristics include variations in temperature, voltage, and sensor placement. For instance, changes in temperature may cause the diagnostic device to expand or shrink, sometimes altering the timbre of its operating sound. Changes in voltage may alter the operating speed of the diagnostic device, sometimes changing the cycle of the rotational sound from motors, etc. Changes in sensor placement may alter the volume of the operating sound due to the change in the distance between the diagnostic device and the microphone.

[0040] In this embodiment, the changes in sound and vibration data obtained from the diagnostic device when changes occur in temperature, voltage, sensor location, and humidity and air pressure obtained by other sensors are treated as domain offsets.

[0041] Structural description

[0042] Figure 1 An example of the structure of the information processing apparatus 100 during learning in Embodiment 1 is shown. The information processing apparatus 100 may be part of a diagnostic device or a device communicatively connected to the diagnostic device. Figure 1 As shown, the information processing device 100 includes an input unit 110, a feature calculation unit 120, a normal / abnormal diagnosis unit 130, a parameter estimation unit 140, and a parameter difference calculation unit 150. Furthermore, the information processing device 100 stores a normal data baseline value DB191, a parameter estimation model DB192, and a parameter difference DB193. DB is an abbreviation for database.

[0043] In addition, the feature quantity calculation unit 120, the normal and abnormal diagnosis unit 130, the parameter estimation unit 140, and the parameter difference calculation unit 150 are collectively referred to as the learning unit.

[0044] Figure 2 An example of the structure of the information processing device 100 during diagnosis is shown. Compared with the information processing device 100 during learning, the information processing device 100 also includes an integration unit 160 and an output unit 170. In addition, the input unit 110 only includes a microphone and a vibration sensor.

[0045] Furthermore, the feature quantity calculation unit 120, the normal / abnormal diagnosis unit 130, the parameter estimation unit 140, the parameter difference calculation unit 150, and the integration unit 160 are collectively referred to as the diagnosis unit. Additionally, the information processing device 100 for diagnosis may also possess the elements possessed by the information processing device 100 for learning. However, among the elements possessed by the information processing device 100 for learning... Figure 2 The elements not shown are not used in the diagnosis.

[0046] The input unit 110 has the function of managing data input. Specifically, the input unit 110 consists of a microphone, a vibration sensor, various sensors, and an input device. The sensors constituting the input unit 110 are divided into sensors used for normal and abnormal diagnosis and sensors used for estimating various physical parameters. The sensors included in the input unit 110 may also be sensors installed independently of the information processing device 100.

[0047] The input device is used to input the measurement results of physical parameters that are difficult to measure automatically, such as the distance between the microphone and the object.

[0048] As a specific example, various sensors consist of thermometers, voltmeters, barometers, etc.

[0049] The sensors included in the input unit 110 perform different operations during learning and diagnosis, as described below. Furthermore, the diagnostic operation assumes situations where sensors used in estimating various physical parameters cannot be used due to cost-cutting measures during sensor installation.

[0050] During learning: Data is acquired from all sensors in the input unit 110.

[0051] During diagnosis: Among the sensors provided in the input unit 110, only the sensors used in normal and abnormal diagnosis are used to obtain data.

[0052] The feature calculation unit 120 calculates frequency-related feature quantities. As a specific example, the feature calculation unit 120 calculates frequency domain data, such as RMS (Root Mean Square), based on sound data or vibration data, using methods such as FFT or Wavelet, and calculates feature quantities related to the calculated frequency domain data.

[0053] The feature calculation unit 120 performs the same operation during both learning and evaluation.

[0054] The normal / abnormality diagnosis unit 130 learns a normal / abnormality diagnosis model based on input data and physical quantities acquired by sensors that obtain physical quantities corresponding to the physical parameters of the object. The input data is data acquired from the device being diagnosed, and it corresponds to time-series data representing frequency. The normal / abnormality diagnosis unit 130 can also learn the normal / abnormality diagnosis model through unsupervised learning.

[0055] The normal / abnormality diagnosis unit 130 uses a normal / abnormality diagnosis model to perform normal / abnormality diagnosis. Specifically, when using each physical parameter of the physical parameter group as the object's physical parameters, the normal / abnormality diagnosis unit 130 inputs feature quantities corresponding to the object's input data into the normal / abnormality diagnosis model to obtain object output data and estimated values ​​of the object's physical parameters corresponding to the object's input data. Furthermore, the normal / abnormality diagnosis unit 130 determines whether the object device has malfunctioned based on the object's output data. The normal / abnormality diagnosis unit 130 can also determine whether the object device has malfunctioned based on the difference between the object's input data and the object's output data. The object's output data is data used to diagnose the object device. The object's output data can also be data corresponding to the object's input data.

[0056] The physical parameter group consists of multiple physical parameters, each corresponding to a physical quantity in the device being diagnosed.

[0057] Object input data is data acquired from the diagnostic device and corresponds to time-series data representing frequencies. Object input data can also be any data representing sound or vibration.

[0058] A normal / abnormal diagnostic model is used to determine whether the state of a device being diagnosed is normal or abnormal, and it estimates various physical parameters. The normal / abnormal diagnostic model takes as input the feature quantities corresponding to the time-series data representing the frequency obtained from the device being diagnosed, and outputs the data used to determine whether the state of the device is normal or abnormal, and the estimated values ​​of the device's physical parameters. A normal / abnormal diagnostic model should satisfy the following conditions.

[0059] Condition 1: The model can be learned using only normal data.

[0060] Condition 2: While learning about anomalies, it is possible to learn models that estimate various physical parameters.

[0061] Specific examples of normal / anomaly diagnostic models include generative models such as AE (Auto Encoder), VAE (Variational Auto Encoder), or flow. In particular, when using AE or VAE as a normal / anomaly diagnostic model, the difference between the input and output data is output as the anomaly degree.

[0062] The normal and abnormal diagnosis unit 130 performs different actions during learning and diagnosis as follows.

[0063] (During study)

[0064] The normal / abnormality diagnosis unit 130 uses only normal data to learn the normal / abnormality diagnosis model. After learning, the normal / abnormality diagnosis unit 130 saves the baseline values ​​required for normal / abnormality diagnosis in the normal data baseline value DB191. As a specific example, the normal / abnormality diagnosis unit 130 uses the anomalousness obtained during learning and determines the baseline values ​​using methods based on Hotelling's theory or quartiles, etc. At the same time, the normal / abnormality diagnosis unit 130 also saves the estimation results of each physical parameter in the normal data baseline value DB191.

[0065] (During diagnosis)

[0066] The normal / abnormality diagnosis unit 130 inputs data that is unsure whether it corresponds to a normal or abnormal state into the normal / abnormality diagnosis model. The model then outputs the anomalousness level and estimates of various physical parameters.

[0067] The normal / abnormality diagnosis unit 130 obtains a baseline value from the normal data baseline value DB191. If the output abnormality exceeds the obtained baseline value, it is determined to be abnormal; otherwise, it is determined to be normal.

[0068] Here, when using VAE as a normal anomaly diagnosis model, as shown in [Mathematical Formula 1], it is possible to define an objective function corresponding to the loss value.

[0069] [Mathematical Expression 1]

[0070] The following explains the variables shown in [Mathematical Formula 1].

[0071] x represents sound or vibration data. x is input into the normal / abnormal diagnostic model and the parameter estimation models. x sometimes also represents the feature quantity corresponding to each data point.

[0072] x' represents the sound or vibration data output by the normal / abnormal diagnostic model. x' sometimes also represents the characteristic quantity corresponding to each data point.

[0073] p1 to p n These represent the true values ​​of each physical parameter. Here, subscripts are used to distinguish between different physical parameters such as temperature, voltage, and distance. Furthermore, let p = (p1, p2, ..., p...). n ).

[0074] p1' to p n 'Represents the physical parameters output by the normal / abnormal diagnostic model or each parameter estimation model. Here, let p = (p1, p2, ..., p... n ).

[0075] σ1 to σ n These represent the variance values ​​output in the intermediate layers of the VAE. Here, let σ = (σ1, σ2, ..., σ...). n ).

[0076] D represents the entire dataset. N D N represents the number of data points contained in D. x Let x represent the dimension.

[0077] Furthermore, as shown in [Mathematical Formula 2], when learning, the normal / abnormality diagnosis unit 130 sets a threshold related to the abnormality degree A based on Hotelling's theory or quartiles, etc., for abnormality detection.

[0078] [Mathematical Expression 2]

[0079] Figure 3 An outline of a normal / abnormal diagnostic model for VAEs is shown. Figure 3 The following scenario is illustrated: if sound data x is input into the normal abnormality diagnosis model, then the physical parameters and sound data x' are estimated. Figure 3 The normal / abnormal diagnostic model shown is a model learned using x and p as normal data.

[0080] During evaluation, the normal / abnormality diagnosis unit 130 inputs a mixture of normal and abnormal sound data into the VAE to perform normal / abnormality diagnosis. At this time, various physical parameters are estimated using the normal / abnormality diagnosis model.

[0081] The parameter estimation unit 140 learns each parameter estimation model. Each parameter estimation model is a model that predicts the physical parameters obtained from each sensor used for physical parameter estimation based on data obtained from sensors used for normal and abnormal diagnosis. One parameter estimation model corresponds to one sensor used for physical parameter estimation. As a specific example, the parameter estimation model is a regression model corresponding to linear regression or symbolic regression. The parameter estimation models constitute a parameter estimation model group.

[0082] The parameter estimation model corresponding to the physical parameters of the object is as follows: the model takes the feature quantity corresponding to the time series data representing the frequency obtained in the diagnostic object equipment as input and the estimated value of the physical parameters of the object as output.

[0083] Figure 4 An overview of the parameter estimation models is shown. Figure 4 The following scenario is illustrated: if sound data x is input to the parameter estimation model, then the physical parameters p are estimated. i (i=1,2,…,n). The parameter estimation models use x and p as normal data. i The model has been learned.

[0084] The parameter estimation unit 140 performs different actions during learning and diagnosis as follows.

[0085] (During study)

[0086] The parameter estimation unit 140 learns a parameter estimation model corresponding to the physical parameters of the object based on the input data and the physical quantities obtained by the sensor that obtains the physical quantities corresponding to the physical parameters of the object.

[0087] As a specific example, the parameter estimation unit 140 uses only normal data to learn models for estimating each physical parameter. Here, the parameter estimation unit 140 creates a number of parameter estimation models that are the same number of sensors used for physical parameter estimation.

[0088] (During diagnosis)

[0089] The parameter estimation unit 140 estimates each physical parameter using the parameter estimation models and data obtained from sensors used for normal and abnormal diagnosis. Specifically, the parameter estimation unit 140 obtains estimated values ​​of the object's physical parameters by inputting feature quantities corresponding to the object's input data into the parameter estimation model corresponding to the object's physical parameters.

[0090] As a specific example, the input data for the parameter estimation model is sound data.

[0091] The parameter difference calculation unit 150 calculates the difference between each physical parameter calculated by the normal / abnormal diagnosis model and each physical parameter calculated by each parameter estimation model, and uses this difference calculation result. The parameter difference calculation unit 150 performs different operations during learning and diagnosis as follows.

[0092] (During study)

[0093] The parameter difference calculation unit 150 saves the difference calculation results for each physical parameter in the parameter difference DB193.

[0094] (During diagnosis)

[0095] The parameter difference calculation unit 150 calculates the difference between the estimation results of the normal and abnormal diagnostic model and the estimation results of each parameter estimation model for each physical parameter, and uses the difference calculation result as the result.

[0096] Furthermore, the parameter difference calculation unit 150 obtains the difference from the normal data in the parameter difference DB193 for each physical parameter as a benchmark value, and determines whether the difference calculation result exceeds the obtained benchmark value. The benchmark value corresponding to the object's physical parameter is equivalent to the difference threshold corresponding to the object's physical parameter.

[0097] As a specific example, the difference associated with the i-th physical parameter is calculated as shown in [Mathematical Formula 3].

[0098] [Mathematical Expression 3]

[0099] The integration unit 160 integrates the judgment results of the normal / abnormal diagnosis unit 130 and the judgment results of the parameter difference calculation unit 150. Specifically, the integration unit 160 outputs information corresponding to the difference between the estimated value of the object's physical parameters based on the normal / abnormal diagnosis model and the estimated value of the object's physical parameters based on the parameter estimation model corresponding to the object's physical parameters, as well as the result of determining whether the diagnostic object device has malfunctioned. The integration unit 160 may also output information indicating that the physical characteristics during the learning of the normal / abnormal diagnosis model differ from the physical characteristics when the object's input data was acquired, for the physical quantity corresponding to the object's physical parameters, if the difference between the estimated value of the object's physical parameters based on the normal / abnormal diagnosis model and the estimated value of the object's physical parameters based on the parameter estimation model corresponding to the object's physical parameters exceeds a difference threshold corresponding to the object's physical parameters. The integration unit 160 can also, for any object physical parameter, if the difference between the estimated value of the object physical parameter based on the normal anomaly diagnosis model and the estimated value of the object physical parameter based on the parameter estimation model corresponding to the object physical parameter is below the difference threshold corresponding to the object physical parameter, for the physical quantity corresponding to the object physical parameter, not output information indicating that the physical characteristics at the time of learning the normal anomaly diagnosis model are different from the physical characteristics at the time of obtaining the object input data, but output information indicating the result of determining whether the diagnostic object device has an anomaly.

[0100] As a specific example, the integration unit 160 appropriately outputs the result of determining whether the condition is normal or abnormal, as well as the physical parameters that may be the cause of the domain offset.

[0101] Figure 5 The outputs of the integration unit 160 and specific examples of the conditions corresponding to each output are shown. The following is an explanation of these examples.

[0102] When the normal / abnormality diagnosis unit 130 determines that the device is normal, and all differential calculation results related to physical parameters are below the reference value, the integration unit 160 outputs information indicating that the device being diagnosed is normal.

[0103] When the normal / abnormal diagnosis unit 130 determines that the condition is normal, if there are differential calculation results in the differential calculation results related to physical parameters that exceed the reference value, the integration unit 160 outputs the following information: the information indicates that a domain offset has occurred, and the physical parameters corresponding to the differential calculation results that exceed the reference value may be the cause of the domain offset.

[0104] When the normal / abnormal diagnosis unit 130 determines that the device is abnormal, if all differential calculation results related to physical parameters are below the reference value, the integration unit 160 outputs information indicating that the device being diagnosed is abnormal.

[0105] When the normal / abnormal diagnosis unit 130 determines an abnormality, and if there are differential calculation results exceeding the reference value in the differential calculation results related to physical parameters, the integration unit 160 outputs the following information: this information indicates that the abnormality level has increased due to the generation of domain offset or the abnormality of the device being diagnosed. Furthermore, the integration unit 160 outputs the following information: this information indicates that the physical parameters corresponding to the differential calculation results exceeding the reference value may be the cause of the domain offset.

[0106] The output unit 170 outputs the output of the integration unit 160 to the outside of the information processing device 100.

[0107] Figure 6 An example of the hardware structure of the information processing apparatus 100 according to this embodiment is shown. The information processing apparatus 100 is configured as a computer. The information processing apparatus 100 may also be configured as multiple computers.

[0108] Various sensors, input devices, and output devices are connected to the information processing device 100.

[0109] As shown in this figure, the information processing device 100 is a computer equipped with a processor 11, a main storage device 12, and an auxiliary storage device 13. These hardware components are appropriately connected via signal lines.

[0110] Processor 11 is an IC (Integrated Circuit) that performs computational processing and controls the hardware of the computer. As a specific example, processor 11 is a CPU (Central Processing Unit), DSP (Digital Signal Processor), or GPU (Graphics Processing Unit).

[0111] The information processing device 100 may also have multiple processors instead of processor 11. The multiple processors share the functions of processor 11.

[0112] The main storage device 12 is typically a volatile storage device, specifically RAM (Random Access Memory). The main storage device 12 is also referred to as the main storage device or main memory. Data stored in the main storage device 12 is stored in the auxiliary storage device 13 as needed.

[0113] The auxiliary storage device 13 is typically a non-volatile storage device, such as ROM (Read Only Memory), HDD (Hard Disk Drive), or flash memory. Data stored in the auxiliary storage device 13 is loaded into the main storage device 12 as needed.

[0114] The main storage device 12 and the auxiliary storage device 13 can also be configured as a single unit.

[0115] Auxiliary storage device 13 stores an information processing program. The information processing program is a program that enables the computer to perform the functions of each component of the information processing device 100. The information processing program is loaded into main storage device 12 and executed by processor 11. The functions of each component of the information processing device 100 are implemented through software.

[0116] Data used during the execution of information processing programs and data obtained through the execution of information processing programs are appropriately stored in storage devices. Each part of the information processing apparatus 100 appropriately utilizes storage devices. Specifically, the storage device comprises at least one of a main storage device 12, an auxiliary storage device 13, a register within the processor 11, and a cache memory within the processor 11. Furthermore, the terms "data" and "information" sometimes have equivalent meanings. Storage devices can also be independent of the computer. Databases are implemented using storage devices.

[0117] The functions of the main storage device 12 and the auxiliary storage device 13 can also be implemented through other storage devices.

[0118] Information processing programs can also be recorded on computer-readable, non-volatile recording media. Specific examples of non-volatile recording media include optical discs or flash memory. Information processing programs can also be provided as program products.

[0119] Description of the action

[0120] The operational steps of the information processing device 100 correspond to the information processing method. Furthermore, the program that implements the operation of the information processing device 100 corresponds to the information processing program.

[0121] Figure 7 This is a flowchart illustrating an example of the operation of the information processing device 100 during learning. Figure 7 Explain the action.

[0122] (Step S101)

[0123] The input unit 110 acquires sound data through a microphone and various physical parameters through various sensors.

[0124] (Step S102)

[0125] The input unit 110 saves the acquired data as normal data.

[0126] (Step S103)

[0127] The feature calculation unit 120 calculates the feature based on the sound data in the stored normal data.

[0128] (Step S104)

[0129] The normal / abnormality diagnosis unit 130 uses the stored normal data and the feature quantities calculated by the feature quantity calculation unit 120 to learn the normal / abnormality diagnosis model.

[0130] (Step S105)

[0131] The normal / abnormal diagnosis unit 130 sets a reference value for determining whether the device to be diagnosed is normal or abnormal, and stores the set reference value in the normal data reference value DB191.

[0132] (Step S106)

[0133] The parameter estimation unit 140 learns a parameter estimation model for each physical parameter using the stored normal data and the feature quantities calculated by the feature quantity calculation unit 120.

[0134] The parameter estimation unit 140 stores the learned parameter estimation models in the parameter estimation model DB192.

[0135] (Step S107)

[0136] The parameter difference calculation unit 150 calculates the difference between the estimation results of the normal and abnormal diagnostic model and the estimation results of the estimation model for each physical parameter, and sets a benchmark value based on the calculated difference.

[0137] The parameter difference calculation unit 150 stores the reference values ​​corresponding to each set physical parameter in the parameter difference DB193.

[0138] Figure 8 This is a flowchart illustrating an example of the operation of the information processing device 100 during diagnosis. Figure 8 Explain the action.

[0139] (Step S121)

[0140] The input unit 110 acquires sound data via a microphone.

[0141] (Step S122)

[0142] The feature calculation unit 120 calculates the feature based on the sound data obtained from the input unit 110.

[0143] (Step S123)

[0144] The normal / abnormality diagnosis unit 130 uses the feature quantity calculated by the feature quantity calculation unit 120 and the normal / abnormality diagnosis model to calculate the degree of abnormality. Then, the normal / abnormality diagnosis unit 130 determines whether the degree of abnormality calculated with reference to the normal data benchmark value DB191 is equivalent to normal or equivalent to abnormal.

[0145] (Step S124)

[0146] The normal and abnormality diagnosis unit 130 uses the normal and abnormality diagnosis model to estimate various physical parameters.

[0147] (Step S125)

[0148] The parameter estimation unit 140 uses the parameter estimation models to estimate the physical parameters.

[0149] (Step S126)

[0150] The parameter difference calculation unit 150 calculates the difference between the estimation results of the normal and abnormal diagnostic model and the estimation results of the parameter estimation model for each physical parameter.

[0151] (Step S127)

[0152] The parameter difference calculation unit 150 refers to the parameter difference DB193. Regarding the differences related to each calculated physical parameter, if there is a difference exceeding the reference value, it proceeds to step S128; otherwise, it proceeds to step S131.

[0153] (Step S128)

[0154] If the diagnosis result of the normal / abnormal diagnosis unit 130 is normal, the integration unit 160 proceeds to step S129. Otherwise, the integration unit 160 proceeds to step S130.

[0155] (Step S129)

[0156] The output of the integrated unit 160 indicates that a domain offset may have occurred, as well as information on the physical parameters corresponding to the domain offset.

[0157] (Step S130)

[0158] The integration unit 160 outputs information indicating a malfunction in the target device. Furthermore, the integration unit 160 outputs information indicating a possible domain offset and the corresponding physical parameters.

[0159] (Step S131)

[0160] The integrated department 160 outputs the diagnostic results of the normal and abnormal diagnostic model.

[0161] Explanation of the effects of Implementation Method 1

[0162] As described above, according to this embodiment, by effectively utilizing the normal / abnormal diagnostic model and the parameter estimation model, it is possible to detect whether the values ​​of each physical parameter have changed during learning. Furthermore, according to this embodiment, through the output of the integration unit 160, it is possible to obtain data indicating a potential decrease in the accuracy of normal / abnormal diagnostics related to the target device, and the physical factors believed to be the cause of the domain shift.

[0163] Therefore, according to this embodiment, the time required for data parsing operations can be shortened in the normal and abnormal diagnosis of the target device.

[0164] Other structures

[0165] <Variation Example 1>

[0166] Figure 9 This illustrates a hardware structure example of the information processing apparatus 100 of this modified example.

[0167] The information processing device 100 includes a processing circuit 18 to replace the processor 11, the processor 11 and the main storage device 12, the processor 11 and the auxiliary storage device 13, or the processor 11, the main storage device 12 and the auxiliary storage device 13.

[0168] The processing circuit 18 is hardware that implements at least a portion of the components of the information processing device 100.

[0169] The processing circuit 18 can be dedicated hardware, or it can be a processor that executes the program stored in the main storage device 12.

[0170] In the case where the processing circuit 18 is dedicated hardware, as a specific example, the processing circuit 18 is a single circuit, a composite circuit, a programmable processor, a parallel programmable processor, an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or a combination thereof.

[0171] The information processing device 100 may also have multiple processing circuits that replace the processing circuit 18. The multiple processing circuits share the functions of the processing circuit 18.

[0172] In the information processing device 100, some functions may be implemented by dedicated hardware, while the remaining functions may be implemented by software or firmware.

[0173] As a specific example, the processing circuit 18 is implemented by hardware, software, firmware, or a combination thereof.

[0174] The processor 11, main storage device 12, auxiliary storage device 13, and processing circuit 18 are collectively referred to as "processing lines". That is, the functions of each functional component of the information processing device 100 are realized through the processing lines.

[0175] The information processing apparatus 100 in other embodiments may also have the same structure as this modified example.

[0176] Implementation method 2.

[0177] Hereinafter, the differences from the embodiments described above will be explained with reference to the accompanying drawings.

[0178] Structural description

[0179] Figure 10 An example of the structure of the information processing apparatus 100 for prior data collection in Embodiment 2 is shown. The information processing apparatus 100 includes an input unit 110, a feature calculation unit 120, and a parameter estimation unit 140. In addition, the information processing apparatus 100 stores a parameter estimation model DB192.

[0180] In addition, the feature quantity calculation unit 120 and the parameter estimation unit 140 are collectively referred to as the advance data collection unit.

[0181] In this embodiment, prior to the implementation of learning processes related to each model, data related to each physical parameter is collected in each device equipped with various sensors. Specifically, each device is of the same type as or similar to the device being diagnosed. Each device corresponds to other devices that are analogous to the device being diagnosed.

[0182] After data collection, similar to Implementation 1, for each physical parameter, the parameter estimation unit 140 creates a parameter estimation model based on the previously collected data and saves the created parameter estimation model in the parameter estimation model DB192.

[0183] Figure 11 The following is a structural example of the information processing device 100 for learning according to Embodiment 2. The information processing device 100 does not include a parameter estimation unit 140, but instead includes a tag assignment unit 210. Furthermore, the input unit 110 does not include various sensors.

[0184] In addition, the feature quantity calculation unit 120, the normal and abnormal diagnosis unit 130, the parameter difference calculation unit 150 and the label assignment unit 210 are collectively referred to as the learning unit.

[0185] The input device takes in data representing the changes in the values ​​of various physical parameters as the learning data is collected. At this time, the user can operate the input device, or the input device can automatically select the physical parameters to be input.

[0186] The tag assignment unit 210 calls the parameter estimation models corresponding to the data input by the input device from the parameter estimation model DB192. Then, for each normal data point, the tag assignment unit 210 estimates each physical parameter using the normal data and the called parameter estimation models, and assigns the estimated physical parameters to the normal data. Normal data is sound data or vibration data.

[0187] In this embodiment, the normal / abnormal diagnosis unit 130 uses the estimated value obtained by inputting input data to a parameter estimation model corresponding to the object's physical parameters as the physical quantity obtained by a sensor that acquires the physical quantity corresponding to the object's physical parameters. Specifically, the normal / abnormal diagnosis unit 130 uses each normal data point and each physical parameter assigned to each normal data point to learn a normal / abnormal diagnosis model.

[0188] In this embodiment, the sensor that obtains the physical quantity corresponding to the physical parameters of the object is a sensor possessed by other devices corresponding to the diagnostic object device.

[0189] The information processing device 100 for diagnosis is as described above.

[0190] In addition, in this embodiment, the objective function used when VAE is used as the normal and abnormal diagnosis model is shown in [Mathematical Formula 4].

[0191] In [Mathematical Expression 4], p1” to p n "Each represents a physical parameter output by the parameter estimation model when normal data is used as input. During the learning of the normal / abnormal diagnostic model, p1" to p n "These are treated as the true values ​​of each physical parameter. Here, let p" = (p1", p2", ..., p n (”).

[0192] [Mathematical Expression 4] Description of the action

[0193] Figure 12 This is a flowchart illustrating an example of the operation of the information processing device 100 during prior data collection. Figure 12 Explain the action.

[0194] (Step S201)

[0195] The input unit 110 acquires sound data through a microphone and various physical parameters through various sensors.

[0196] (Step S202)

[0197] The feature calculation unit 120 calculates the feature based on the sound data obtained from the input unit 110.

[0198] (Step S203)

[0199] The parameter estimation unit 140 learns a parameter estimation model for each physical parameter using the physical parameters obtained from various sensors and the feature quantities calculated by the feature quantity calculation unit 120.

[0200] (Step S204)

[0201] The parameter estimation unit 140 stores the learned parameter estimation models as references in the parameter estimation model DB192.

[0202] Figure 13 This is a flowchart illustrating an example of the operation of the information processing device 100 during learning. Figure 13 Explain the action.

[0203] (Step S221)

[0204] The input unit 110 acquires sound data via a microphone.

[0205] (Step S222)

[0206] The label assignment unit 210 reads in the parameter estimation model DB192, which corresponds to the parameter estimation model of the diagnostic target device, i.e., the reference parameter estimation model.

[0207] (Step S223)

[0208] The tag assignment unit 210 uses the feature quantities calculated by the feature quantity calculation unit 120 and the input parameter estimation models to estimate each physical parameter.

[0209] (Step S224)

[0210] The normal / abnormality diagnosis unit 130 uses the feature quantities calculated by the feature quantity calculation unit 120 and the physical parameters estimated by the label assignment unit 210 to learn a normal / abnormality diagnosis model. At this time, the normal / abnormality diagnosis unit 130 processes the physical parameters estimated by the label assignment unit 210 as the true values ​​of each physical parameter.

[0211] Explanation of the effects of Implementation Method 2

[0212] As described above, according to this embodiment, the true values ​​of each physical parameter in the diagnostic device are generated by using parameter estimation models based on measurement results from devices other than the diagnostic device. Therefore, according to this embodiment, even when the diagnostic device does not have sensors to acquire each physical parameter, it is possible to perform diagnosis of the diagnostic device and estimation related to domain offset.

[0213] Other implementation methods

[0214] It is possible to freely combine the above-described embodiments, modify any constituent element of each embodiment, or omit any constituent element of each embodiment.

[0215] Furthermore, the implementation method is not limited to the manner shown in Implementation Methods 1 and 2, and various modifications can be made as needed. The steps described using flowcharts and other similar methods can also be appropriately modified.

[0216] Label Explanation

[0217] 11 Processor, 12 Main storage device, 13 Auxiliary storage device, 18 Processing circuit, 100 Information processing device, 110 Input unit, 120 Feature quantity calculation unit, 130 Normal and abnormal diagnosis unit, 140 Parameter estimation unit, 150 Parameter difference calculation unit, 160 Synthesis unit, 170 Output unit, 191 Normal data baseline value DB, 192 Parameter estimation model DB, 193 Parameter difference DB, 210 Label assignment unit.

Claims

1. An information processing device comprising: The normal / abnormality diagnosis unit, when setting each physical parameter of a physical parameter group (composed of multiple physical parameters corresponding to physical quantities in the target device) as the target physical parameter, obtains the target output data and the estimated value of the target physical parameter for diagnosing the target device by inputting feature quantities corresponding to the target input data into the normal / abnormality diagnosis model. Based on the target output data, it determines whether the target device has experienced an abnormality. The object input data is data obtained from the diagnostic object device and is data corresponding to time series data representing frequency; The parameter estimation unit obtains estimated values ​​of the object's physical parameters by inputting feature quantities corresponding to the object's input data into a parameter estimation model corresponding to the object's physical parameters; and The integration department outputs information corresponding to the difference between the estimated value of the object's physical parameters based on the normal / abnormal diagnostic model and the estimated value of the object's physical parameters based on the parameter estimation model corresponding to the object's physical parameters, and the result of determining whether the diagnostic object device has experienced an abnormality.

2. The information processing apparatus according to claim 1, wherein, The object output data is the data corresponding to the object input data. The normal / abnormal diagnosis unit determines whether the diagnostic object device has malfunctioned based on the difference between the object input data and the object output data.

3. The information processing apparatus according to claim 1 or 2, wherein, The parameter estimation unit learns a parameter estimation model corresponding to the physical parameters of the object based on input data and physical quantities obtained by sensors that acquire physical quantities corresponding to the physical parameters of the object. The input data is data obtained in the diagnostic object device and is data corresponding to time series data representing frequency.

4. The information processing apparatus according to claim 3, characterized in that, The normal / abnormality diagnosis unit learns the normal / abnormality diagnosis model based on input data and physical quantities obtained by sensors that acquire physical quantities corresponding to the physical parameters of the object. The input data is data obtained in the diagnostic object device and is data corresponding to time series data representing frequency.

5. The information processing apparatus according to claim 4, wherein, The sensor that acquires the physical quantities corresponding to the physical parameters of the object is a sensor possessed by other devices corresponding to the diagnostic object device. The normal / abnormality diagnosis unit uses the estimated value obtained by inputting the input data into the parameter estimation model corresponding to the physical parameters of the object as the physical quantity obtained by the sensor that obtains the physical quantity corresponding to the physical parameters of the object.

6. The information processing apparatus according to claim 4 or 5, wherein, The normal and abnormal diagnosis unit learns the normal and abnormal diagnosis model through unsupervised learning.

7. The information processing apparatus according to any one of claims 1 to 6, wherein, When the difference between the estimated value of the object's physical parameters based on the normal / abnormal diagnostic model and the estimated value of the object's physical parameters based on the parameter estimation model corresponding to the object's physical parameters exceeds a difference threshold corresponding to the object's physical parameters, the integration unit outputs information indicating that the physical characteristics of the normal / abnormal diagnostic model during learning are different from the physical characteristics of the object when the input data is obtained.

8. The information processing apparatus according to any one of claims 1 to 7, wherein, For any physical parameter of the object, if the difference between the estimated value of the physical parameter of the object based on the normal / abnormal diagnosis model and the estimated value of the physical parameter of the object based on the parameter estimation model corresponding to the physical parameter of the object is below the difference threshold corresponding to the physical parameter of the object, the integrated unit does not output information indicating that the physical characteristics during the learning of the normal / abnormal diagnosis model are different from the physical characteristics when the object input data is obtained, but instead outputs information indicating whether the diagnostic object device has an abnormality.

9. The information processing apparatus according to any one of claims 1 to 8, wherein, The input data for the object is any data that represents sound or vibration.

10. An information processing method, wherein, When a computer sets each physical parameter of a physical parameter group (consisting of multiple physical parameters corresponding to physical quantities in the diagnostic target device) as the object's physical parameters, it obtains object output data and estimated values ​​of the object's physical parameters by inputting feature quantities corresponding to the object's input data into the normal / abnormal diagnostic model. Based on the object's output data, it determines whether the diagnostic target device has experienced an abnormality. The object's input data is data obtained from the diagnostic target device and corresponds to time-series data representing frequency. The computer obtains estimated values ​​of the object's physical parameters by inputting feature quantities corresponding to the object's input data into a parameter estimation model corresponding to the object's physical parameters. The computer outputs information corresponding to the difference between the estimated value of the object's physical parameters based on the normal / abnormal diagnostic model and the estimated value of the object's physical parameters based on the parameter estimation model corresponding to the object's physical parameters, and a result determining whether the diagnostic object device has experienced an abnormality.

11. An information processing program that causes an information processing device, which is a computer, to perform the following processes: In the normal anomaly diagnosis process, when each physical parameter of a physical parameter group (consisting of multiple physical parameters corresponding to physical quantities in the target device) is set as the target physical parameter, the characteristic quantity corresponding to the target input data is input to the normal anomaly diagnosis model to obtain the target output data and the estimated value of the target physical parameter for diagnosing the target device. Based on the target output data, it is determined whether the target device has experienced an anomaly. The object input data is data obtained from the diagnostic object device and is data corresponding to time series data representing frequency; The parameter estimation process obtains the estimated values ​​of the object's physical parameters by inputting the feature quantities corresponding to the object's input data into the parameter estimation model corresponding to the object's physical parameters. as well as The comprehensive processing outputs the following information, which corresponds to the difference between the estimated value of the object's physical parameters based on the normal / abnormal diagnostic model and the estimated value of the object's physical parameters based on the parameter estimation model corresponding to the object's physical parameters, as well as the result of determining whether the diagnostic object device has experienced an abnormality.

Images