Information processing device, information processing method, and information processing program

Abstract

An information processing device (100) is provided with a normality / abnormality diagnosis unit (130), a parameter estimation unit (140), and an integration unit (160). The normality / abnormality diagnosis unit (130) inputs a feature amount corresponding to target input data, which corresponds to time-series data indicating frequency, into a normality / abnormality diagnosis model to obtain target output data and an estimated value of a target physical parameter, and determines whether or not an abnormality is present in equipment to be diagnosed on the basis of the target output data. The parameter estimation unit (140) inputs the feature amount corresponding to the target input data into a parameter estimation model to obtain an estimated value of the target physical parameter. The integration unit (160) outputs information in accordance with: the difference between the estimated value of the target physical parameter according to the normality / abnormality diagnosis model and the estimated value of the target physical parameter according to the parameter estimation model; and the result of determining whether or not an abnormality is present in the equipment to be diagnosed.

Description

Information processing device, information processing method, and information processing program

[0001] The present disclosure relates to an information processing device, an information processing method, and an information processing program.

[0002] A large number of sensors are used in the diagnosis of machine anomalies. However, the accuracy of the anomaly diagnosis decreases due to variations in the measurement results of the sensors, so improving the accuracy of the anomaly diagnosis is an issue. Patent Document 1 discloses a technology for detecting sensors that generate many false alarms and missed alarms.

[0003] JP 2016-133944 A

[0004] Changes in the characteristics of each physical parameter (such as temperature and operating speed) related to the equipment to be diagnosed can change the operating noise and vibration of the equipment to be diagnosed. Furthermore, changes in the characteristics of each physical parameter can change the characteristics of the sensor measurement results, which can result in a decrease in the accuracy of abnormality diagnosis. Therefore, it is desirable to consider the characteristics of each physical parameter and the operating noise and vibration of the machine when estimating the cause of a decrease in the accuracy of abnormality diagnosis for the equipment to be diagnosed. However, the technology disclosed in Patent Document 1 does not consider the characteristics of each physical parameter and the operating noise and vibration of the equipment to be diagnosed when estimating the cause of a decrease in the accuracy of abnormality diagnosis for the equipment to be diagnosed. The present disclosure aims to consider the characteristics of each physical parameter corresponding to a physical quantity in the equipment to be diagnosed and the operating noise or vibration of the machine when estimating the cause of a decrease in the accuracy of abnormality diagnosis for the equipment to be diagnosed.

[0005] The information processing device according to the present disclosure comprises: a normal / abnormality diagnosis unit that, when each physical parameter of a physical parameter group consisting of a plurality of physical parameters each corresponding to a physical quantity in a device to be diagnosed is defined as a target physical parameter, acquires target output data used to diagnose the device to be diagnosed and an estimated value of the target physical parameter by inputting, into a normal / abnormality diagnosis model, feature values ​​corresponding to target input data, which is data acquired in the device to be diagnosed and corresponds to time series data indicating frequency, and determines whether or not an abnormality has occurred in the device to be diagnosed based on the target output data; a parameter estimation unit that acquires an estimated value of the target physical parameter by inputting, into a parameter estimation model corresponding to the target physical parameter; and an integration unit that outputs information corresponding to the difference between the estimated value of the target physical parameter obtained by the normal / abnormality diagnosis model and the estimated value of the target physical parameter obtained by the parameter estimation model corresponding to the target physical parameter, and the result of determining whether or not an abnormality has occurred in the device to be diagnosed.

[0006] According to the present disclosure, the integrating unit outputs a result corresponding to the difference between the estimated value of the target physical parameter according to the normal / abnormality diagnosis model and the estimated value of the target physical parameter according to the parameter estimation model corresponding to the target physical parameter, and the result of determining whether an abnormality has occurred in the equipment to be diagnosed. Here, the normal / abnormality diagnosis model and each parameter estimation model receive as input feature quantities corresponding to target input data, which is data acquired in the equipment to be diagnosed and corresponds to time-series data indicating frequency. Therefore, according to the present disclosure, the characteristics of each physical parameter corresponding to a physical quantity in the equipment to be diagnosed and the operating noise or vibration of the machine can be taken into consideration when estimating the cause of a decrease in the accuracy of an abnormality diagnosis for the equipment to be diagnosed.

[0007] 1 is a diagram showing an example of the configuration of the information processing device 100 during learning according to embodiment 1. FIG. 2 is a diagram showing an example of the configuration of the information processing device 100 during diagnosis according to embodiment 1. FIG. 3 is a diagram explaining a normal / abnormal diagnosis model according to embodiment 1. FIG. 4 is a diagram explaining each parameter estimation model according to embodiment 1. FIG. 5 is a diagram explaining the processing of an integrating unit 160 according to embodiment 1. FIG. 6 is a flowchart showing the operation of the information processing device 100 during learning according to embodiment 1. FIG. 7 is a flowchart showing the operation of the information processing device 100 during diagnosis according to embodiment 1. FIG. 8 is a diagram showing an example of the hardware configuration of the information processing device 100 according to a modification of embodiment 1. FIG. 9 is a diagram showing an example of the configuration of the information processing device 100 during advance data collection according to embodiment 2. FIG. 10 is a diagram showing an example of the configuration of the information processing device 100 during learning according to embodiment 2. FIG. 11 is a flowchart showing the operation of the information processing device 100 during advance data collection according to embodiment 2. FIG. 12 is a flowchart showing the operation of the information processing device 100 during learning according to embodiment 2.

[0008] In the description of the embodiments and the drawings, the same elements and corresponding elements are given the same reference numerals. The description of elements given the same reference numerals will be omitted or simplified as appropriate. Arrows in the drawings mainly indicate the flow of data or the flow of processing. Furthermore, "unit" may be read as "circuit," "step," "procedure," "process," or "circuitry" as appropriate.

[0009] Embodiment 1. This embodiment will be described in detail below with reference to the drawings. This embodiment deals with a case where the operating sound and vibration of a machine can change due to changes in multiple physical parameters related to machine operation (such as temperature and operating speed). Changes in the operating sound and vibration of a machine caused by multiple physical parameters correspond to domain changes. This embodiment can be applied to both sound data and vibration data related to a machine. However, for the sake of simplicity, sound data and vibration data may be collectively referred to as "sound" or "sound data." The term "parameter" may also refer to the value of a parameter.

[0010] When collecting training data, in addition to sound data acquired from a microphone, data indicating various physical quantities, such as temperature, voltage, and the distance between the microphone and the device, can be acquired as physical parameters. When collecting evaluation data, only sound data from the microphone can be acquired, and no additional data regarding the physical parameters can be acquired. When performing normal / abnormal diagnosis using a generative model, a normal / abnormality diagnosis model, in the above-described situation, a domain shift caused by changes in the physical parameters may result in erroneous determination of normality and abnormality. Therefore, this embodiment provides a method for estimating whether changes in physical parameters are within an acceptable range and which physical parameters have changed. To solve this problem, the difference between the estimated results of each physical parameter output by the normal / abnormality diagnosis model and the estimated results of each physical parameter output by each dedicated model for estimating each physical parameter is utilized. The normal / abnormality diagnosis model corresponds to an estimator that estimates multiple physical parameters. A regression model such as linear regression or symbolic regression is used as each dedicated model. Each dedicated model corresponds to an estimator for each physical parameter. The sound data is transformed into the frequency domain using FFT (Fast Fourier Transform) or the like and used as features. Then, an estimator is created that estimates multiple physical parameters using the training data. When the evaluation data is input, each estimator estimates each physical parameter. Here, if there is a difference in the estimation results between the normal / abnormal diagnosis model and each dedicated model that is equal to or greater than a threshold, it is displayed that the target physical parameter has changed. The threshold may be a preset value or may be a value derived using a statistical test or the like.

[0011] A domain shift is a state in which the characteristics of input data differ from the characteristics of normal data input during training. A domain shift may also be a state in which the characteristics of each physical quantity related to the input data differ from the characteristics of each physical quantity related to the normal data input during training. Normal data is data corresponding to a normal state. For example, normal data includes data acquired in a normal state and data calculated based on the data acquired in the normal state. The normal state is a state in which no abnormality is detected. When a domain shift occurs, data having characteristics different from the characteristics of the data input during training is input. Therefore, a domain shift can cause a decrease in the diagnostic accuracy of a normal / abnormal diagnosis model. Specific examples of causes of changes in data characteristics include changes in temperature, voltage, and sensor installation position. For example, a change in temperature can cause the expansion or contraction of the device to be diagnosed, which can change the tone of the operating sound. A change in voltage can change the operating speed of the device to be diagnosed, which can change the period of the rotation sound of a motor or the like. A change in the sensor installation position can change the distance between the device to be diagnosed and the microphone, which can change the volume of the operating sound. In this embodiment, when the temperature, voltage, sensor installation position, humidity and air pressure that can be acquired by other sensors change, the change in sound data and vibration data that can be acquired from the equipment to be diagnosed is treated as a domain shift.

[0012] *** Description of Configuration *** FIG. 1 shows an example configuration of an information processing device 100 during learning according to the first embodiment. The information processing device 100 may be part of the device to be diagnosed, or may be a device communicatively connected to the device to be diagnosed. As shown in FIG. 1 , 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. The information processing device 100 also stores a normal data reference value DB 191, a parameter estimation model DB 192, and a parameter difference DB 193. DB is an abbreviation for database. The feature calculation unit 120, the normal / abnormal diagnosis unit 130, the parameter estimation unit 140, and the parameter difference calculation unit 150 are collectively referred to as a learning unit.

[0013] FIG. 2 shows an example configuration of the information processing device 100 during diagnosis. Compared to the information processing device 100 during learning, the information processing device 100 further includes an integration unit 160 and an output unit 170. The input unit 110 includes only a microphone and a vibration sensor. The feature calculation unit 120, normal / abnormal diagnosis unit 130, parameter estimation unit 140, parameter difference calculation unit 150, and integration unit 160 are collectively referred to as a diagnosis unit. The information processing device 100 during diagnosis may include each of the elements included in the information processing device 100 during learning. However, elements not shown in FIG. 2 among the elements included in the information processing device 100 during learning are not used during diagnosis.

[0014] The input unit 110 has a function of managing data input. Specifically, the input unit 110 includes a microphone, a vibration sensor, various sensors, and an input device. The sensors included in the input unit 110 are divided into sensors used for normal / abnormal diagnosis and sensors used for estimating each physical parameter. The sensors included in the input unit 110 may be sensors installed independently of the information processing device 100. The input device is used to input measurement results of physical parameters that are difficult to measure automatically, such as the distance between the microphone and an object. Specific examples of the various sensors include a thermometer, a voltmeter, and a barometer. The sensors included in the input unit 110 perform different operations during learning and diagnosis, as described below. Note that the operation during diagnosis is intended to assume a situation in which the sensors used for estimating each physical parameter cannot be used due to reasons such as cost reduction when installing the sensors.

[0015] During learning: Data is acquired from all sensors included in the input unit 110. During diagnosis: Of the sensors included in the input unit 110, only the sensors used in normal / abnormal diagnosis acquire data.

[0016] The feature calculation unit 120 calculates feature quantities related to frequency. Specifically, the feature calculation unit 120 calculates frequency domain data from sound data or vibration data using FFT, wavelet, or the like, and calculates feature quantities such as RMS (Root Mean Square) related to the calculated frequency domain data. The feature calculation unit 120 performs the same operation both during learning and evaluation.

[0017] The normality / abnormality diagnosis unit 130 learns a normality / abnormality diagnosis model based on input data and physical quantities acquired by a sensor that acquires physical quantities corresponding to target physical parameters. The input data is data acquired by the device to be diagnosed and corresponds to time-series data indicating frequency. The normality / abnormality diagnosis unit 130 may learn the normality / abnormality diagnosis model through unsupervised learning. The normality / abnormality diagnosis unit 130 performs normality / abnormality diagnosis using the normality / abnormality diagnosis model. Specifically, when each physical parameter of the physical parameter group is set as a target physical parameter, the normality / abnormality diagnosis unit 130 inputs feature quantities corresponding to the target input data into the normality / abnormality diagnosis model to acquire target output data corresponding to the target input data and an estimated value of the target physical parameter. The normality / abnormality diagnosis unit 130 also determines whether an abnormality has occurred in the device to be diagnosed based on the target output data. The normality / abnormality diagnosis unit 130 may determine whether an abnormality has occurred in the device to be diagnosed based on the difference between the target input data and the target output data. The target output data is data used to diagnose the device to be diagnosed. The target output data may be data corresponding to the target input data. The physical parameter group consists of multiple physical parameters, each corresponding to a physical quantity in the equipment to be diagnosed. The target input data is data acquired from the equipment to be diagnosed and corresponds to time-series data indicating frequency. The target input data may be either data indicating sound or data indicating vibration. The normal / abnormal diagnosis model is a model for determining whether the state of the equipment to be diagnosed is normal or abnormal, and is also a model for estimating each physical parameter. The normal / abnormal diagnosis model is a model that receives as input feature quantities corresponding to data acquired from the equipment to be diagnosed and corresponding to time-series data indicating frequency, and outputs data used to determine whether the state of the equipment to be diagnosed is normal or abnormal, and estimated values ​​of the target physical parameters. A model that satisfies the following conditions is used as the normal / abnormal diagnosis model.

[0018] Condition 1: It is possible to train a model using only normal data. Condition 2: It is possible to train a model that estimates each physical parameter simultaneously with training on the degree of anomaly.

[0019] Specific examples of the normal / abnormality diagnosis model include generative models such as AE (Auto Encoder), VAE (Variational Auto Encoder), and flow. In particular, when AE or VAE is used as the normal / abnormality diagnosis model, the difference between input data and output data is output as the degree of anomaly. The normal / abnormality diagnosis unit 130 performs different operations during learning and during diagnosis, as described below.

[0020] (During Learning) The normal / abnormality diagnosis unit 130 learns a normal / abnormality diagnosis model using only normal data. After learning, the normal / abnormality diagnosis unit 130 stores the reference values ​​required for normal / abnormal diagnosis in the normal data reference value DB 191. As a specific example, the normal / abnormality diagnosis unit 130 determines the reference values ​​using the abnormality levels obtained during learning through a method such as Hotelling's theory or quartiles. At the same time, the normal / abnormality diagnosis unit 130 also stores the estimation results of each physical parameter in the normal data reference value DB 191.

[0021] (During diagnosis) The normality / abnormality diagnosis unit 130 inputs data into the normality / abnormality diagnosis model, without knowing whether the data corresponds to a normal state or an abnormal state. The normality / abnormality diagnosis model then outputs the degree of abnormality and also outputs the estimation results of each physical parameter. The normality / abnormality diagnosis unit 130 acquires a reference value from the normal data reference value DB 191, and determines an abnormality if the output degree of abnormality exceeds the acquired reference value, and determines a normality otherwise.

[0022] Here, when VAE is used as the normal / abnormal diagnosis model, an objective function corresponding to the loss value can be defined as shown in [Equation 1].

[0023]

[0024] Each variable shown in [Equation 1] will be explained below. x indicates sound data or vibration data. x is input to the normal / abnormal diagnosis model and each parameter estimation model. x may also indicate a feature value corresponding to each data. x' indicates sound data or vibration data output by the normal / abnormal diagnosis model. x' may also indicate a feature value corresponding to each data. p １ From p ｎ Each of these indicates the true value of each physical parameter. Here, different physical parameters such as temperature, voltage, and distance are distinguished by subscripts. Also, p = (p １ , p ２ , ..., p ｎ ) p １ ' to p ｎ Each of p' indicates a physical parameter output by the normal / abnormal diagnosis model or each parameter estimation model. １ , p ２ , ..., p ｎ ) σ １ From σ ｎ Each of these represents the variance value output in the hidden layer of the VAE. １ , σ ２ , …, σ ｎ ) where D denotes the entire data set. N Ｄ indicates the number of data included in D. ｘ indicates the number of dimensions of x. Furthermore, the degree of anomaly A is defined as shown in [Equation 2], and the normal / abnormality diagnosis unit 130 sets a threshold value for the degree of anomaly A for anomaly detection during learning using the Hotelling theory, quartiles, or the like.

[0025]

[0026] FIG. 3 shows an overview of the normal / abnormality diagnosis model, which is a VAE. FIG. 3 shows how, when sound data x is input to the normal / abnormality diagnosis model, each physical parameter and sound data x' are estimated. The normal / abnormality diagnosis model shown in FIG. 3 is a model trained using normal data x and p. During evaluation, the normal / abnormality diagnosis unit 130 inputs data that is a mixture of sound data corresponding to normality and sound data corresponding to abnormality into the VAE to perform normal / abnormal diagnosis. At this time, each physical parameter is estimated by the normal / abnormality diagnosis model.

[0027] The parameter estimation unit 140 trains each parameter estimation model. Each parameter estimation model is a model that predicts physical parameters obtained from each sensor for estimating physical parameters from data obtained from a sensor for normal / abnormal diagnosis. One parameter estimation model corresponds to one sensor for estimating physical parameters. Specific examples of the parameter estimation model include a regression model corresponding to linear regression or symbolic regression. Each parameter estimation model constitutes a parameter estimation model group. The parameter estimation model corresponding to the target physical parameter is a model that receives as input a feature value corresponding to data corresponding to time-series data indicating frequency, which is data acquired in the equipment to be diagnosed, and outputs an estimated value of the target physical parameter. Figure 4 shows an overview of each parameter estimation model. In Figure 4, when sound data x is input to each parameter estimation model, each physical parameter p ｉ (i = 1, 2, ..., n) are estimated. Each parameter estimation model uses normal data x and p ｉ The parameter estimation unit 140 performs different operations during learning and diagnosis as follows.

[0028] (During Learning) The parameter estimation unit 140 learns a parameter estimation model corresponding to the target physical parameter based on the input data and physical quantities acquired by a sensor that acquires the physical quantities corresponding to the target physical parameter. As a specific example, the parameter estimation unit 140 learns a model that estimates each physical parameter using only normal data. Here, the parameter estimation unit 140 creates parameter estimation models in the same number as the number of sensors used to estimate the physical parameters.

[0029] (During diagnosis) The parameter estimation unit 140 estimates each physical parameter using each parameter estimation model and data acquired by a sensor for normal / abnormal diagnosis. Specifically, the parameter estimation unit 140 acquires an estimate of the target physical parameter by inputting a feature corresponding to the target input data into the parameter estimation model corresponding to the target physical parameter. A specific example of the input data for each parameter estimation model is sound data.

[0030] 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 as a difference calculation result. The parameter difference calculation unit 150 operates differently during learning and during diagnosis as described below.

[0031] (During Learning) The parameter difference calculation section 150 stores the difference calculation results for each physical parameter in the parameter difference DB 193 .

[0032] (During diagnosis) The parameter difference calculation unit 150 calculates, for each physical parameter, the difference between the estimation result using the normal / abnormal diagnosis model and the estimation result using each parameter estimation model as the difference calculation result. Furthermore, for each physical parameter, the parameter difference calculation unit 150 acquires the difference in the normal data from the parameter difference DB 193 as a reference value, and determines whether the current difference calculation result exceeds the acquired reference value. The reference value corresponding to the target physical parameter corresponds to the difference threshold corresponding to the target physical parameter. The difference related to the i-th physical parameter is calculated as shown in [Equation 3] as a specific example.

[0033]

[0034] The integrating unit 160 integrates the determination result by the normal / abnormality diagnosis unit 130 and the determination result by the parameter difference calculation unit 150. Specifically, the integrating unit 160 outputs information corresponding to the difference between the estimated value of the target physical parameter according to the normal / abnormality diagnosis model and the estimated value of the target physical parameter according to the parameter estimation model corresponding to the target physical parameter, and the result of the determination as to whether an abnormality has occurred in the diagnosis target device. When the difference between the estimated value of the target physical parameter according to the normal / abnormality diagnosis model and the estimated value of the target physical parameter according to the parameter estimation model corresponding to the target physical parameter exceeds a difference threshold corresponding to the target physical parameter, the integrating unit 160 may output information indicating that, for the physical quantity corresponding to the target physical parameter, the physical characteristics at the time of learning the normal / abnormality diagnosis model differ from the physical characteristics at the time the target input data was acquired. For any target physical parameter, when the difference between the estimated value of the target physical parameter based on the normal / abnormal diagnosis model and the estimated value of the target physical parameter based on the parameter estimation model corresponding to the target physical parameter is equal to or less than the difference threshold corresponding to the target physical parameter, the integrating unit 160 may output information indicating the result of determining whether or not an abnormality has occurred in the diagnosis target device, without outputting information indicating that the physical characteristics of the physical quantity corresponding to the target physical parameter at the time of learning the normal / abnormal diagnosis model differ from the physical characteristics at the time of acquiring the target input data. As a specific example, the integrating unit 160 appropriately outputs the result of the normal / abnormal determination and each physical parameter that may be causing the domain shift. FIG. 5 shows specific examples of each output of the integrating unit 160 and the conditions corresponding to each output. This example will be described below. When the normal / abnormal diagnosis unit 130 determines that the diagnosis target device is normal, if all difference calculation results related to the physical parameters are equal to or less than the reference value, the integrating unit 160 outputs information indicating that the diagnosis target device is normal.When the normality / abnormality diagnosis unit 130 determines a normal state and any difference calculation results related to the physical parameters exceed a reference value, the integrating unit 160 outputs information indicating that a domain shift has occurred and that each physical parameter corresponding to the difference calculation results that exceed the reference value may be the cause of the domain shift. When the normality / abnormality diagnosis unit 130 determines a normal state and all difference calculation results related to the physical parameters are equal to or less than the reference value, the integrating unit 160 outputs information indicating that the diagnosis target device is abnormal. When the normality / abnormality diagnosis unit 130 determines a normal state and any difference calculation results related to the physical parameters exceed a reference value, the integrating unit 160 outputs information indicating that the degree of abnormality is increasing due to the occurrence of a domain shift or an abnormality in the diagnosis target device. Furthermore, the integrating unit 160 outputs information indicating that each physical parameter corresponding to the difference calculation results that exceed the reference value may be the cause of the domain shift.

[0035] The output unit 170 outputs the output of the integration unit 160 to the outside of the information processing device 100 .

[0036] 6 shows an example of the hardware configuration of an information processing device 100 according to this embodiment. The information processing device 100 is composed of a computer. The information processing device 100 may be composed of multiple computers. Sensors, input devices, and output devices are connected to the information processing device 100.

[0037] As shown in the figure, the information processing device 100 is a computer that includes a processor 11, a main storage device 12, and an auxiliary storage device 13. These pieces of hardware are connected as appropriate via signal lines.

[0038] The processor 11 is an integrated circuit (IC) that performs arithmetic processing and controls the hardware of the computer. Specific examples of the processor 11 include a central processing unit (CPU), a digital signal processor (DSP), or a graphics processing unit (GPU). The information processing device 100 may include multiple processors that replace the processor 11. The multiple processors share the role of the processor 11.

[0039] The main storage device 12 is typically a volatile storage device, specifically a RAM (Random Access Memory). The main storage device 12 is also called a main storage device or a main memory. Data stored in the main storage device 12 is saved in the auxiliary storage device 13 as needed.

[0040] The auxiliary storage device 13 is typically a non-volatile storage device, and specific examples thereof include a ROM (Read Only Memory), an HDD (Hard Disk Drive), or a flash memory. Data stored in the auxiliary storage device 13 is loaded into the main storage device 12 as needed. The main storage device 12 and the auxiliary storage device 13 may be configured integrally.

[0041] The auxiliary storage device 13 stores an information processing program. The information processing program is a program that causes a computer to realize the functions of each unit included in the information processing device 100. The information processing program is loaded into the main storage device 12 and executed by the processor 11. The functions of each unit included in the information processing device 100 are realized by software.

[0042] Data used when executing an information processing program and data obtained by executing the information processing program are stored in a storage device as appropriate. Each part of the information processing device 100 uses a storage device as appropriate. Specific examples of the storage device include at least one of the main storage device 12, the auxiliary storage device 13, a register in the processor 11, and a cache memory in the processor 11. Note that the terms "data" and "information" may have the same meaning. A storage device may be independent of a computer. A database is realized by a storage device. The functions of the main storage device 12 and the auxiliary storage device 13 may be realized by other storage devices.

[0043] The information processing program may be recorded on a computer-readable non-volatile recording medium. Specific examples of the non-volatile recording medium include an optical disk and a flash memory. The information processing program may be provided as a program product.

[0044] ***Description of Operation*** The operational procedure of the information processing device 100 corresponds to an information processing method. Also, the program that realizes the operation of the information processing device 100 corresponds to an information processing program.

[0045] 7 is a flowchart showing an example of the operation of the information processing device 100 during learning. This operation will be described with reference to FIG.

[0046] (Step S101) The input unit 110 acquires sound data through a microphone and acquires various physical parameters through various sensors.

[0047] (Step S102) The input unit 110 stores the acquired data as normal data.

[0048] (Step S103) The feature amount calculation unit 120 calculates feature amounts from the sound data of the stored normal data.

[0049] (Step S104) The normal / abnormality diagnosis unit 130 uses the stored normal data and the feature amounts calculated by the feature amount calculation unit 120 to learn a normal / abnormality diagnosis model.

[0050] (Step S105 ) The normality / abnormality diagnosis unit 130 sets a reference value for determining whether the diagnosis target device is normal or abnormal, and stores the set reference value in the normal data reference value DB 191 .

[0051] (Step S106) The parameter estimation unit 140 learns a parameter estimation model for each physical parameter using the stored normal data and the feature amount calculated by the feature amount calculation unit 120. The parameter estimation unit 140 stores each learned parameter estimation model in the parameter estimation model DB 192.

[0052] (Step S107) The parameter difference calculation unit 150 calculates the difference between the estimation result by the normal / abnormal diagnosis model and the estimation result by each parameter estimation model for each physical parameter, and sets a reference value based on the calculated difference. The parameter difference calculation unit 150 stores the set reference value corresponding to each physical parameter in the parameter difference DB 193.

[0053] 8 is a flowchart showing an example of the operation of the information processing device 100 during diagnosis. This operation will be described with reference to FIG.

[0054] (Step S121) The input unit 110 acquires sound data through the microphone.

[0055] (Step S122 ) The feature amount calculation unit 120 calculates feature amounts from the sound data acquired by the input unit 110 .

[0056] (Step S123) The normal / abnormality diagnosis unit 130 calculates the degree of abnormality using the feature amount calculated by the feature amount calculation unit 120 and the normal / abnormality diagnosis model. Thereafter, the normal / abnormality diagnosis unit 130 refers to the normal data reference value DB 191 to determine whether the calculated degree of abnormality corresponds to normality or abnormality.

[0057] (Step S124) The normal / abnormality diagnosis unit 130 estimates each physical parameter using the normal / abnormality diagnosis model.

[0058] (Step S125) The parameter estimation unit 140 estimates each physical parameter using each parameter estimation model.

[0059] (Step S126) The parameter difference calculation unit 150 calculates the difference between the estimation result by the normal / abnormal diagnosis model and the estimation result by each parameter estimation model for each physical parameter.

[0060] (Step S127) The parameter difference calculation unit 150 refers to the parameter difference DB 193, and if any of the calculated differences for each physical parameter exceeds a reference value, it transitions to step S128, and otherwise transitions to step S131.

[0061] (Step S128) If the diagnosis result by the normality / abnormality diagnosis unit 130 is normal, the integrating unit 160 proceeds to step S129. Otherwise, the integrating unit 160 proceeds to step S130.

[0062] (Step S129) The integrating unit 160 outputs information indicating that a domain shift may have occurred and each physical parameter corresponding to the domain shift.

[0063] (Step S130) The integrating unit 160 outputs information indicating that the diagnosis target device is abnormal, and also outputs information indicating that a domain shift may have occurred and each physical parameter corresponding to the domain shift.

[0064] (Step S131) ​​The integrating unit 160 outputs a diagnosis result based on the normal / abnormal diagnosis model.

[0065] ***Description of Effects of First Embodiment*** As described above, according to this embodiment, by utilizing the normal / abnormality diagnosis model and each parameter estimation model, it is possible to detect whether the value of each physical parameter has changed since learning. Furthermore, according to this embodiment, the output of the integrating unit 160 makes it possible to grasp data in which a domain shift that may have occurred and that would reduce the accuracy of normal / abnormal diagnosis of the equipment to be diagnosed, as well as physical factors that may be the cause of the domain shift. Therefore, according to this embodiment, it is possible to reduce the time required for data analysis work in normal / abnormal diagnosis of the equipment to be diagnosed.

[0066] ***Other Configurations*** <Modification 1> Fig. 9 shows an example of the hardware configuration of an information processing device 100 according to this modification. The information processing device 100 includes a processing circuit 18 instead of the processor 11, the processor 11 and main storage device 12, the processor 11 and auxiliary storage device 13, or the processor 11, main storage device 12, and auxiliary storage device 13. The processing circuit 18 is hardware that realizes at least a portion of the components included in the information processing device 100. The processing circuit 18 may be dedicated hardware, or may be a processor that executes a program stored in the main storage device 12.

[0067] When the processing circuitry 18 is dedicated hardware, the processing circuitry 18 may be, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or a combination thereof. The information processing device 100 may include multiple processing circuits that replace the processing circuitry 18. The multiple processing circuits share the role of the processing circuitry 18.

[0068] In the information processing device 100, some of the functions may be realized by dedicated hardware, and the remaining functions may be realized by software or firmware.

[0069] The processing circuitry 18 is realized by, for example, hardware, software, firmware, or a combination of these. The processor 11, the main memory device 12, the auxiliary memory device 13, and the processing circuitry 18 are collectively referred to as the "processing circuitry." In other words, the functions of the functional components of the information processing device 100 are realized by the processing circuitry. Information processing devices 100 according to other embodiments may also have a configuration similar to that of this modified example.

[0070] Second Embodiment The following mainly describes the differences from the above-described embodiment with reference to the drawings.

[0071] *** Description of Configuration *** Fig. 10 shows an example configuration of an information processing device 100 at the time of collecting preliminary data according to embodiment 2. The information processing device 100 includes an input unit 110, a feature calculation unit 120, and a parameter estimation unit 140. The information processing device 100 also stores a parameter estimation model DB 192. The feature calculation unit 120 and the parameter estimation unit 140 are collectively referred to as a preliminary data collection unit.

[0072] In this embodiment, before the learning process for each model is performed, data on each physical parameter is collected in advance for each device equipped with various sensors. Specifically, each device is a device of the same type as the device to be diagnosed, or a device similar to the device to be diagnosed. Each device corresponds to another device corresponding to the device to be diagnosed. After the data collection, as in the first embodiment, the parameter estimation unit 140 creates a parameter estimation model for each physical parameter based on the data collected in advance, and stores the created parameter estimation models in the parameter estimation model DB 192.

[0073] 11 shows an example of the configuration of the information processing device 100 during learning according to embodiment 2. The information processing device 100 does not include a parameter estimation unit 140, but instead includes a label assignment unit 210. Furthermore, the input unit 110 does not include various sensors. Note that the feature calculation unit 120, normal / abnormal diagnosis unit 130, parameter difference calculation unit 150, and label assignment unit 210 are collectively referred to as a learning unit.

[0074] The input device inputs data indicating each physical parameter whose value changed when the learning data was collected. At this time, the user may operate the input device, or the input device may automatically select each physical parameter to be input.

[0075] The labeling unit 210 calls up each parameter estimation model corresponding to the data input by the input device from the parameter estimation model DB 192. Then, for each normal data, the labeling unit 210 estimates each physical parameter using the normal data and each called parameter estimation model, and assigns the estimated physical parameters to the normal data. The normal data is sound data or vibration data.

[0076] The normalcy / abnormality diagnosis unit 130 according to this embodiment uses, as the physical quantity acquired by a sensor that acquires a physical quantity corresponding to the target physical parameter, an estimated value acquired by inputting input data into a parameter estimation model corresponding to the target physical parameter. As a specific example, the normalcy / abnormality diagnosis unit 130 performs training of the normalcy / abnormality diagnosis model using each normal data and each physical parameter assigned to each normal data. In this embodiment, the sensor that acquires the physical quantity corresponding to the target physical parameter is a sensor included in another device that corresponds to the device to be diagnosed.

[0077] The information processing device 100 at the time of diagnosis is as described above. In this embodiment, the objective function when VAE is used as the normal / abnormal diagnosis model is as shown in [Equation 4]. In [Equation 4], p １ '' to p ｎ Each of p '' indicates a physical parameter output by each parameter estimation model when normal data is input. １ '' to p ｎ Each of p'' is treated as the true value of each physical parameter when learning the normal / abnormal diagnosis model. １ '', p ２ '', ..., p ｎ '').

[0078]

[0079] ***Explanation of Operation*** Fig. 12 is a flowchart showing an example of the operation of the information processing device 100 during advance data collection. This operation will be described with reference to Fig. 12.

[0080] (Step S201) The input unit 110 acquires sound data through a microphone and acquires various physical parameters through various sensors.

[0081] (Step S202 ) The feature calculation unit 120 calculates feature amounts from the sound data acquired by the input unit 110 .

[0082] (Step S203) The parameter estimation unit 140 learns a parameter estimation model for each physical parameter using the physical parameters acquired by the various sensors and the feature amounts calculated by the feature amount calculation unit 120.

[0083] (Step S204) The parameter estimation unit 140 stores each of the learned parameter estimation models in the parameter estimation model DB 192 as a reference parameter estimation model.

[0084] 13 is a flowchart showing an example of the operation of the information processing device 100 during learning. This operation will be described with reference to FIG.

[0085] (Step S221) The input unit 110 acquires sound data through the microphone.

[0086] (Step S222) The labeling unit 210 refers to the parameter estimation model DB 192 and reads each parameter estimation model that corresponds to the device to be diagnosed and is a reference parameter estimation model.

[0087] (Step S223) The labeling unit 210 estimates each physical parameter using the feature amount calculated by the feature amount calculation unit 120 and each parameter estimation model that has been read.

[0088] (Step S224) The normal / abnormality diagnosis unit 130 learns a normal / abnormality diagnosis model using the feature amounts calculated by the feature amount calculation unit 120 and the physical parameters estimated by the label assignment unit 210. At this time, the normal / abnormality diagnosis unit 130 treats the physical parameters estimated by the label assignment unit 210 as the true values ​​of the physical parameters.

[0089] ***Explanation of Effects of Embodiment 2*** As described above, according to this embodiment, true values ​​of each physical parameter in the device to be diagnosed are generated using each parameter estimation model created based on measurement results of devices other than the device to be diagnosed. Therefore, according to this embodiment, even if the device to be diagnosed does not have sensors for acquiring each physical parameter, it is possible to diagnose the device to be diagnosed and estimate the domain shift.

[0090] ***Other Embodiments*** The above-described embodiments can be freely combined, or any of the components of each embodiment can be modified, or any of the components can be omitted from each embodiment. Furthermore, the embodiments are not limited to those shown in embodiments 1 and 2, and various modifications are possible as needed. The procedures described using flowcharts, etc., can be modified as appropriate.

[0091] 11 Processor, 12 Main memory device, 13 Auxiliary memory device, 18 Processing circuit, 100 Information processing device, 110 Input unit, 120 Feature calculation unit, 130 Normality / abnormality diagnosis unit, 140 Parameter estimation unit, 150 Parameter difference calculation unit, 160 Integration unit, 170 Output unit, 191 Normal data reference value DB, 192 Parameter estimation model DB, 193 Parameter difference DB, 210 Label assignment unit.

Claims

1. An information processing device comprising: a normal / abnormality diagnosis unit that, when each physical parameter of a physical parameter group consisting of a plurality of physical parameters each corresponding to a physical quantity in a device to be diagnosed is defined as a target physical parameter, acquires target output data used to diagnose the device to be diagnosed and an estimated value of the target physical parameter by inputting feature values ​​corresponding to target input data, which is data acquired in the device to be diagnosed and corresponds to time-series data indicating frequency, into a normal / abnormality diagnosis model, and determines whether or not an abnormality has occurred in the device to be diagnosed based on the target output data; a parameter estimation unit that acquires an estimated value of the target physical parameter by inputting feature values ​​corresponding to the target input data into a parameter estimation model corresponding to the target physical parameter; and an integration unit that outputs information corresponding to the difference between the estimated value of the target physical parameter obtained by the normal / abnormality diagnosis model and the estimated value of the target physical parameter obtained by the parameter estimation model corresponding to the target physical parameter, and the result of determining whether or not an abnormality has occurred in the device to be diagnosed.

2. The information processing device described in claim 1, wherein the target output data is data corresponding to the target input data, and the normal / abnormal diagnosis unit determines whether an abnormality has occurred in the equipment to be diagnosed based on the difference between the target input data and the target output data.

3. An information processing device as described in claim 1 or 2, wherein the parameter estimation unit learns a parameter estimation model corresponding to the target physical parameter based on input data, which is data acquired in the equipment to be diagnosed and corresponds to time series data indicating frequency, and a physical quantity acquired by a sensor that acquires a physical quantity corresponding to the target physical parameter.

4. The information processing device described in claim 3, wherein the normal / abnormal diagnosis unit learns the normal / abnormal diagnosis model based on input data, which is data acquired in the equipment to be diagnosed and corresponds to time series data indicating frequency, and physical quantities acquired by a sensor that acquires physical quantities corresponding to the target physical parameters.

5. The information processing device described in claim 4, wherein the sensor that acquires the physical quantity corresponding to the target physical parameter is a sensor provided in another device that corresponds to the device to be diagnosed, and the normal / abnormal diagnosis unit uses an estimated value acquired by inputting the input data into a parameter estimation model corresponding to the target physical parameter as the physical quantity acquired by the sensor that acquires the physical quantity corresponding to the target physical parameter.

6. The information processing device according to claim 4 or 5, wherein the normal / abnormal diagnosis unit learns the normal / abnormal diagnosis model by unsupervised learning.

7. An information processing device described in any one of claims 1 to 6, wherein the integration unit outputs information indicating that, for a physical quantity corresponding to the target physical parameter, the physical characteristics at the time of learning the normal / abnormal diagnosis model differ from the physical characteristics at the time the target input data was acquired when the difference between the estimated value of the target physical parameter by the normal / abnormal diagnosis model and the estimated value of the target physical parameter by a parameter estimation model corresponding to the target physical parameter exceeds a difference threshold corresponding to the target physical parameter.

8. An information processing device as described in any one of claims 1 to 7, wherein, when the difference between the estimated value of the target physical parameter by the normal / abnormal diagnosis model and the estimated value of the target physical parameter by the parameter estimation model corresponding to the target physical parameter is equal to or less than the difference threshold corresponding to the target physical parameter, the integration unit does not output information indicating that the physical characteristics of the physical quantity corresponding to the target physical parameter at the time of learning the normal / abnormal diagnosis model are different from the physical characteristics at the time the target input data was acquired, but outputs information indicating the result of determining whether or not an abnormality has occurred in the equipment to be diagnosed.

9. The information processing device according to any one of claims 1 to 8, wherein the target input data is either data representing sound or data representing vibration.

10. An information processing method in which, when a computer sets each physical parameter of a physical parameter group consisting of multiple physical parameters each corresponding to a physical quantity in a device to be diagnosed as a target physical parameter, the computer inputs feature quantities corresponding to target input data, which is data acquired in the device to be diagnosed and corresponds to time-series data indicating frequency, into a normal / abnormal diagnosis model to obtain target output data used to diagnose the device to be diagnosed and an estimated value of the target physical parameter, and determines whether or not an abnormality has occurred in the device to be diagnosed based on the target output data; the computer inputs feature quantities corresponding to the target input data into a parameter estimation model corresponding to the target physical parameter to obtain an estimated value of the target physical parameter; and the computer outputs information corresponding to the difference between the estimated value of the target physical parameter obtained by the normal / abnormal diagnosis model and the estimated value of the target physical parameter obtained by the parameter estimation model corresponding to the target physical parameter, and the result of determining whether or not an abnormality has occurred in the device to be diagnosed.

11. An information processing program that causes an information processing device that is a computer to execute the following steps: a normal / abnormal diagnosis process that, when each physical parameter of a physical parameter group consisting of multiple physical parameters each corresponding to a physical quantity in a device to be diagnosed is a target physical parameter, acquires target output data used to diagnose the device to be diagnosed and an estimated value of the target physical parameter by inputting feature values ​​corresponding to target input data, which is data acquired in the device to be diagnosed and corresponds to time-series data indicating frequency, into a normal / abnormal diagnosis model, and determines whether or not an abnormality has occurred in the device to be diagnosed based on the target output data; a parameter estimation process that acquires an estimated value of the target physical parameter by inputting feature values ​​corresponding to the target input data into a parameter estimation model corresponding to the target physical parameter; and an integration process that outputs information corresponding to the difference between the estimated value of the target physical parameter obtained by the normal / abnormal diagnosis model and the estimated value of the target physical parameter obtained by the parameter estimation model corresponding to the target physical parameter, and the result of determining whether or not an abnormality has occurred in the device to be diagnosed.

Images