Abnormality factor estimation system, method and program

Abstract

To provide an abnormality factor estimation system, method, and program capable of conveniently estimating abnormality factors of target facilities from abnormal sound data.SOLUTION: An abnormality factor estimation system according to the embodiment includes a first extraction unit, a reconstruction error calculation unit, a database, and an estimation unit. The first extraction unit extracts a first acoustic feature based on frequency, time, and signal intensity from sound data related to target facilities. The reconstruction error calculation unit calculates a first reconstruction error which is difference between a reconstructed acoustic feature obtained by reconstructing the first acoustic feature on the basis of a reconstruction model that encodes and decodes acoustic features, and the first acoustic feature. The database stores multiple abnormality factors and multiple sample vectors based on second reconstruction errors correspondingly. The estimation unit estimates the abnormality factor of the target facilities by performing a vector search of the database using a query vector based on the first reconstruction error.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] An embodiment of the present invention relates to an abnormality factor estimation system, method, and program. [Background technology]

[0002] Infrastructure facilities, including industrial equipment, undergo regular inspections to maintain safety. Abnormalities in the facilities are often manifested as sounds. For this reason, there is technology that can detect abnormalities in facilities by capturing the operating sounds of the facilities using microphones and analyzing the state of those sounds.

[0003] One technology for detecting anomalies from operating sounds is to estimate the cause of the anomaly based on frequency. However, when operating sounds contain a wide range of variations, it is difficult to design a correspondence between frequency and anomaly cause. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent No. 7056465 Summary of the Invention [Problem to be solved by the invention]

[0005] The problem to be solved by the present invention is to provide an abnormality factor estimation system, method, and program that can easily estimate the abnormality factor of target equipment from abnormal sound data. [Means for solving the problem]

[0006] An abnormality factor estimation system according to an embodiment includes a first extraction unit, a reconstruction error calculation unit, a database, and an estimation unit. The first extraction unit extracts a first acoustic feature based on frequency, time, and signal strength from sound data related to a target facility. The reconstruction error calculation unit calculates a first reconstruction error, which is the difference between a reconstructed feature obtained by reconstructing the first acoustic feature based on a reconstruction model that encodes and decodes the acoustic feature, and the first acoustic feature. The database stores a plurality of abnormality factors and a plurality of sample vectors based on a plurality of second reconstruction errors, in association with each other. The estimation unit estimates the abnormality factor of the target facility by performing a vector search in the database using a query vector based on the first reconstruction error. [Brief explanation of the drawings]

[0007] [Figure 1] FIG. 1 is a diagram showing an example of the configuration of an abnormality factor estimation system according to a first embodiment; [Figure 2] FIG. 2 is a diagram showing an example of the configuration of the information processing device shown in FIG. 1; [Figure 3] FIG. 10 is a diagram illustrating a processing procedure for estimating an abnormality factor according to the first embodiment; [Figure 4] FIG. 1 is a diagram schematically illustrating a processing procedure for estimating an abnormality factor according to the first embodiment; [Figure 5] FIG. 10 is a diagram illustrating a display screen. [Figure 6] FIG. 10 is a diagram showing an example of the configuration of an information processing device according to a second embodiment; [Figure 7] FIG. 10 is a diagram illustrating a processing procedure for estimating an abnormality factor according to the second embodiment; [Figure 8] FIG. 10 is a diagram schematically illustrating a processing procedure for estimating an abnormality factor according to a second embodiment. [Figure 9] Diagram showing examples of abnormality factors [Figure 10] FIG. 10 is a diagram illustrating a display screen. [Figure 11] FIG. 10 is a diagram showing an example of the configuration of an information processing device according to a third embodiment; [Figure 12] FIG. 11 is a diagram illustrating a processing procedure for estimating an abnormality factor according to the third embodiment; [Figure 13]FIG. 11 is a diagram schematically illustrating a processing procedure for estimating an abnormality factor according to the third embodiment. [Figure 14] FIG. 10 is a diagram showing an example of the configuration of an information processing device according to a fourth embodiment; [Figure 15] FIG. 13 is a diagram illustrating a processing procedure for estimating an abnormality factor according to the fourth embodiment; [Figure 16] FIG. 10 is a diagram schematically illustrating a processing procedure for estimating an abnormality factor according to a fourth embodiment. [Figure 17] FIG. 13 is a diagram showing an example of the configuration of an information processing device according to a fifth embodiment; [Figure 18] FIG. 13 is a diagram illustrating a processing procedure for generating a reconstructed model according to the fifth embodiment; [Figure 19] FIG. 13 is a diagram schematically illustrating a processing procedure for estimating an abnormality factor according to the fifth embodiment; DETAILED DESCRIPTION OF THE INVENTION

[0008] (First embodiment) Hereinafter, an abnormality factor estimation system, method, and program according to the present embodiment will be described with reference to the drawings.

[0009] FIG. 1 is a diagram illustrating an example of the hardware configuration of an abnormality factor estimation system 600 according to the first embodiment. As illustrated in FIG. 1, the abnormality factor estimation system 600 includes a target facility 300, a microphone 34, N (N is a natural number equal to or greater than 1) information processing devices 100k (k is a natural number between 1 and N), and a database 200. The target facility 300 is a facility for which an abnormality factor is to be estimated. Operational sounds are generated in the target facility 300 as the machine operates. Specific examples of the target facility 300 include machine tools, manufacturing equipment, elevators, and electrical equipment. The microphone 34 collects the operational sounds emitted by the target facility 300 and collects sound data, which are digital signals of the operational sounds over time. While sensing of the operational sounds using the microphone 34 is exemplified here as an example of a sensor, other sensors such as vibration sensors and acoustic emission sensors that output one-dimensional waveform signals other than sound may also be used. The information processing device 100k extracts features from the acquired sound data, generates a query vector based on the feature, and performs a vector search in the database 200 using the query vector to estimate the cause of the abnormality. Although not shown, each of the information processing devices 100k may be connected to another microphone to collect sound data from other target equipment. The database 200 is a computer that stores a plurality of abnormality causes and a plurality of sample vectors based on a plurality of second reconstruction errors in association with each other. Hereinafter, the term "information processing device 100" will be used to refer to the information processing devices 100a-100N without distinction. As an example, the abnormality cause estimation system 600 is a system in which the information processing device 100 is an edge device such as a personal computer and the database 200 is a server computer.

[0010] Fig. 2 is a diagram showing an example of the hardware configuration of the information processing device 100. As shown in Fig. 2, the information processing device 100 is a computer having a processor 1, a storage device 2, an input device 3, a display device 4, and a communication device 5. Transmission and reception of data and various signals between the processor 1, the storage device 2, the input device 3, the display device 4, and the communication device 5 is performed via a bus.

[0011] The processor 1 is an integrated circuit that controls the overall operation of the information processing device 100. For example, the processor 1 has a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), a DSP (Digital Signal Processor), and / or an FPU (Floating-Point Unit). The processor 1 may also have an internal memory and an I / O interface. The processor 1 executes various processes by interpreting and calculating programs stored in advance in a storage device 2 or the like. The processor 1 may be partially or entirely realized by hardware such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array).

[0012] The storage device 2 is a volatile memory and / or a non-volatile memory that stores various data. For example, the storage device 2 stores data and setting values ​​used when the processor 1 executes various processes, data generated by various processes in the processor 1, etc. The storage device 2 is configured with a ROM (Read Only Memory), a RAM (Random Access Memory), an HDD (Hard Disk Drive), an SSD (Solid State Drive), an integrated circuit storage device, etc. The storage device 2 may also include a non-transitory computer-readable storage medium that stores a program executed by the processor 1.

[0013] The input device 3 accepts various operation inputs from an operator. Examples of the input device 3 that can be used include a keyboard, a mouse, various switches, a touchpad, a touch panel display, and a microphone. An electrical signal corresponding to the accepted operation input (hereinafter referred to as an operation signal) is supplied to the processor 1. For example, sound data corresponding to input from the microphone 34 is supplied to the processor 1.

[0014] The display device 4 displays various data under the control of the processor 1. A CRT (Cathode-Ray Tube) display, a liquid crystal display, an organic EL (Electro Luminescence) display, an LED (Light-Emitting Diode) display, a plasma display, or any other display may be used as appropriate as the display device 4. The display device 4 may also be a projector.

[0015] The communication device 5 includes a communication interface such as a network interface card (NIC) for performing data communication with various devices connected to the information processing device 100 via a network. The information processing device 100 may acquire audio data collected by a microphone 34 external to the information processing device 100 via the communication device 5. Note that an operation signal may be supplied from a computer connected via the communication device 5 or an input device included in the computer, and various data may be displayed on a display device or the like included in the computer connected via the communication device 5. However, for the sake of simplicity of the following description, unless otherwise specified, it is assumed that the source of the operation signal is the input device 3 and the display destination of the various data is the display device 4. The input device 3 can be replaced by a computer connected via the communication device 5 or an input device included in the computer, and the display device 4 can be replaced by a display device or the like included in the computer connected via the communication device 5.

[0016] The information processing device 100 does not need to include all of the processor 1, storage device 2, input device 3, display device 4, and communication device 5. If necessary, some of the storage device 2, input device 3, display device 4, and communication device 5 may not be provided. The information processing device 100 may also be provided with any additional hardware device useful for executing the processing according to this embodiment. The information processing device 100 does not need to be physically configured as a single computer, but may be configured as a computer system having multiple computers communicably connected via wires, a network, or the like. The allocation of the series of processing according to this embodiment to the multiple processors 1 implemented in each of the multiple computers can be arbitrarily set. All processors 1 may execute all processing in parallel, or specific processing may be assigned to one or some of the processors 1, and the series of processing according to this embodiment may be executed by the entire computer system.

[0017] As shown in FIG. 2, the processor 1 has functional components such as a first extraction unit 11, a reconstruction error calculation unit 12, an estimation unit 13, and a display control unit 14.

[0018] The first extractor 11 extracts a first acoustic feature quantity defined by frequency, time, and signal intensity from the sound data relating to the target facility 300 collected by the microphone .

[0019] The reconstruction error calculation unit 12 calculates a first reconstruction error, which is a difference between a reconstructed feature obtained by reconstructing a first acoustic feature based on a reconstruction model that performs encoding and decoding of the acoustic feature, and the first acoustic feature used for reconstructing the reconstructed feature.

[0020] The estimation unit 13 estimates the cause of the abnormality in the target facility 300 by performing a vector search in the database 200 using a query vector based on the first reconstruction error calculated by the reconstruction error calculation unit 12.

[0021] The display control unit 14 displays various information on the display device 5. For example, the display control unit 14 displays the abnormality cause of the target equipment 300 estimated by the estimation unit 13 on the display device 5.

[0022] Fig. 3 is a diagram showing a processing procedure for estimating an abnormality factor of the target equipment 300 according to the first embodiment. Fig. 4 is a diagram showing a schematic diagram of the processing procedure according to the first embodiment. The processing for estimating an abnormality factor of the target equipment 300 according to the first embodiment will be described below with reference to the flow of Figs. 3 and 4.

[0023] As shown in FIG. 3, the first extraction unit 11 extracts acoustic features defined by frequency, time, and signal intensity from sound data related to the target facility 300 (step S11). The sound data is input, for example, via a microphone in the input device 3. However, the sound data may also be input from an external microphone 34 of the information processing device 100 via the communication device 5, or sound data acquired from the microphone 34 may be edited and input to the information processing device 100. The data type of the acoustic features is a vector or image. As an example, the first extraction unit 11 performs a Fourier transform on the sound data to extract acoustic features as a spectrogram or mel spectrogram image. The sound data may be data having a finite time period that allows a Fourier transform and allows abnormal components to be identified. Specifically, the time period is approximately one second. A spectrogram and a mel spectrogram are graphs that represent the distribution of signal intensity of sound data in a two-dimensional space of frequency and time. This converts the sound data into acoustic features of a data type that is easy for users to check.

[0024] After step S11 is performed, the reconstruction error calculation unit 12 calculates a first reconstruction error, which is the difference between the acoustic feature extracted in step S11 and a reconstructed feature obtained by reconstructing the acoustic feature extracted in step S11 based on the reconstruction model (step S12). In step S12, the reconstruction error calculation unit 12 first applies the acoustic feature extracted in step S11 to the reconstruction model, and calculates an acoustic feature reconstructed by encoding and decoding the acoustic feature (hereinafter, reconstructed feature). Next, the reconstruction error calculation unit 12 calculates the first reconstruction error, which is the difference between the reconstructed feature and the acoustic feature extracted in step S11.

[0025] The reconstruction model is a machine learning model that receives acoustic features, encodes and decodes the input acoustic features, and outputs acoustic features. For example, an autoencoder, which is a combination of a decoder and an encoder, may be used as the reconstruction model. The reconstruction model is generated by the information processing device 100 or another computer. Hereinafter, a device that generates a reconstruction model will be referred to as a learning device. The learning device trains the reconstruction model by performing optimization calculations based on training data, which are acoustic features (hereinafter, "correct acoustic features") extracted from normal sound data collected when the target equipment 300 is operating normally. More specifically, the learning device inputs the correct acoustic features into an untrained reconstruction model to calculate predicted acoustic features, and updates parameters such as weight parameters and biases of the untrained reconstruction model to minimize the error between the correct acoustic features and the predicted acoustic features. The parameter update is repeated until a predetermined stopping condition is satisfied. A trained reconstruction model is completed by assigning a set of parameters obtained when the predetermined stopping condition is satisfied to the untrained reconstruction model. The data type of the first reconstruction feature may be the same as that of the acoustic feature. By extracting acoustic features using the reconstruction model, it is possible to analyze abnormalities based on multiple components including frequency and time.

[0026] When step S12 is performed, the estimation unit 13 estimates the abnormality factor of the target equipment 300 by performing a vector search in a database using a query vector based on the first reconstruction error calculated in step S12 (step S13). Specifically, the estimation unit 13 generates a query vector based on the first reconstruction error. The estimation unit 13 performs a vector search in the database 200 using the generated query vector. The estimation unit 13 estimates the abnormality factor of the target equipment 300 from the database 200 by the vector search. The query vector is a vector that serves as a reference for a neighborhood search in the vector search. The database 200 stores a plurality of abnormality factors in association with a plurality of sample vectors based on a plurality of second reconstruction errors. The second reconstruction error is a sample equipment similar to the target equipment 300 and / or a reconstruction error of a past case calculated by the reconstruction error calculation unit 12 or the like in a simulation of the sample equipment. The abnormality factor associated with the second reconstruction error is an abnormality factor in the sample equipment and / or a simulation of the sample equipment. The second reconstruction error may include a reconstruction error calculated from the target equipment 300 itself.

[0027] More specifically, the estimation unit 13 estimates, as the abnormality factor of the target equipment 300, an abnormality factor corresponding to a specific sample vector among the multiple sample vectors, the sample vector having a similarity to the query vector that exceeds a first threshold. The sample vector is, for example, a vectorized (embedded) second reconstruction error. The query vector may be a vectorized first reconstruction error obtained by processing similar to that for the sample vector. Specifically, when the first reconstruction error is an image of a spectrogram or a melspectrogram, the estimation unit 13 vectorizes the first reconstruction error using a convolutional neural network (CNN). As the similarity, the Euclidean distance, cosine similarity, Manhattan distance, Hamming distance, Mahalanobis distance, or the like between the multiple sample vectors and the query vector in a vector space may be used. As a vector search method, for example, a nearest neighbor search algorithm (NN), a k-nearest neighbor search algorithm (kNN), or an approximate nearest neighbor search algorithm (aNN) is used. By using a vector search to estimate the abnormality factor, the effort required by the user to design the correspondence between various abnormality factors and reconstruction errors is reduced.

[0028] However, the estimation unit 13 may store the first reconstruction error in association with the query vector, separately from the query vector obtained by vectorizing the first reconstruction error, in the storage device 2. Hereinafter, a specific sample vector among the multiple sample vectors, whose similarity to the query vector exceeds a first threshold, will be referred to as a similar vector.

[0029] When step S13 is performed, the display control unit 14 displays the abnormality factor of the target equipment 300 estimated in step S13 (step S14). As an example, the display control unit 14 displays a display screen on the display device 4 to show the abnormality factor of the target equipment 300. The display screen may display the first acoustic feature, the reconstructed feature, the reconstruction error, the abnormality factor associated with the similar vector, and the second reconstruction error, which are input to the reconstruction error calculation unit.

[0030] FIG. 5 is a diagram illustrating a display screen I1 displaying an abnormality factor. The display screen I1 in FIG. 5 includes a display field I11. The display field I11 displays a character string representing the abnormality factor of the target equipment 300, such as "distortion of the guide rail." The display screen I1 also displays a spectrogram. The spectrograms displayed on the display screen I1 may be a spectrogram I12 relating to the first reconstruction error and a spectrogram I13 relating to the second reconstruction error associated with the similar vector. The vertical axes of the spectrograms I12 and I13 represent frequency. The horizontal axes represent time. The color tone represents signal intensity. The spectrogram I12 in FIG. 5 represents, from left to right, the first acoustic feature, the reconstruction feature, and the first reconstruction error. The date and time information displayed on the display screen I1 in FIG. 5 is information regarding the date and time of occurrence of the similar vector. In this case, the sample vector and the date and time of occurrence of the sample vector may be associated and stored in a database. By displaying the spectrogram I12 and the spectrogram I13 together, the user can easily check whether the estimation of the abnormality factor of the target equipment 300 is valid. By displaying the character string representing the abnormality factor and the information on the occurrence date and time of the similar vector, the user can check the occurrence interval of the abnormality factor.

[0031] This completes the processing procedure for estimating the cause of an abnormality in the target equipment 300.

[0032] The processing procedure for estimating the cause of an abnormality in the target equipment 300 shown in FIGS. 3 and 4 is an example, and various deletions, additions and / or modifications can be made without departing from the gist of the invention.

[0033] Here, the first embodiment will be compared with a comparative example in which an abnormality factor is estimated using a rule-based method. In the comparative example, frequency features are extracted from sound data and the abnormality factor is estimated by using the frequency features. However, when the sound data is not based on simple movements but contains a large amount of variation, using a rule base that associates abnormality factors with each frequency band makes it difficult to design an allocation rule for the abnormality factors as the variation in the abnormality factors increases. In contrast to the comparative example, this embodiment extracts frequency features and employs vector search to estimate the abnormality factor, making it possible to easily associate abnormality factors with feature values ​​even when the variation in the abnormality factors increases.

[0034] (Second embodiment) The abnormality factor estimation system in the first embodiment performs a vector search using a query vector based on the first reconstruction error. The abnormality factor estimation system in the second embodiment performs a vector search using a query vector based on the first frequency feature and the first time feature extracted from the first reconstruction error. The abnormality factor estimation system according to the second embodiment will be described below. However, components having the same functions as those in the first embodiment are denoted by the same reference numerals and will be described only when necessary.

[0035] Fig. 6 is a diagram illustrating an example of the hardware configuration of an information processing device 100 according to the second embodiment. As illustrated in Fig. 6, the processor 1 has functional components such as a first extraction unit 11, a reconstruction error calculation unit 12, an estimation unit 13, and a display control unit 14, as well as a second extraction unit 15 and a third extraction unit 16.

[0036] The second extraction unit 15 extracts a first frequency feature from the first reconstruction error calculated by the reconstruction error calculation unit 12. The second extraction unit 15 extracts the first frequency feature representing the frequency spatial distribution of the signal intensity of the sound data, for example, by compressing the first reconstruction error in the time direction.

[0037] The third extraction unit extracts a first temporal feature from the first reconstruction error. For example, the third extraction unit compresses the first reconstruction error in the frequency direction to extract the first temporal feature representing the spatiotemporal distribution of the signal strength of the sound data.

[0038] Fig. 7 is a diagram showing a processing procedure for estimating an abnormality factor of the target equipment 300 according to the second embodiment. Fig. 7 shows the processing procedure from step S12 shown in Fig. 3 to the end of the processing. Fig. 8 is a diagram showing a schematic diagram of the processing procedure according to the second embodiment. In the second embodiment, as in the first embodiment, the processor 1 extracts acoustic features from sound data in step S11 and calculates a reconstruction error in step S12.

[0039] As shown in FIG. 7, after step S12, the second extraction unit 15 extracts a first frequency feature from the first reconstruction error calculated in step S12 (step S23). For example, the second extraction unit 15 compresses the first reconstruction error in the time direction to obtain the first frequency feature. The compression process by the second extraction unit 15 may involve calculation of the sum of signal strengths for each frequency, the maximum value of signal strength, or the time average of signal strength. Specifically, the second extraction unit 15 obtains a frequency spectrum by calculating the sum of signal strengths of the first reconstruction error for each frequency. The frequency spectrum includes, for example, a distribution of frequency peaks of abnormal components with higher signal strengths than other frequencies. The first frequency feature may be calculated as the distribution of the frequency peaks of the abnormal components themselves.

[0040] After step S23, the third extraction unit 16 extracts a first temporal feature from the first reconstruction error calculated in step S12 (step S24). For example, the third extraction unit 16 compresses the first reconstruction error in the frequency direction to obtain the first temporal feature. The compression process by the third extraction unit 16 may involve calculation of the sum of signal intensities at each time, the maximum value of signal intensities, or the ensemble average of signal intensities. Specifically, the third extraction unit 16 calculates the sum of signal intensities of the reconstruction error for each time to obtain a waveform representing a temporal change in the sum of signal intensities as the first temporal feature. The waveform represents, for example, the occurrence timing and duration of an abnormal component with a higher signal intensity than at other times. The duration is the duration of signal intensity exceeding a second threshold. The second threshold may be set to a value that allows for separation of an abnormal component from a normal component. The first temporal feature may be calculated as the occurrence timing and duration of the abnormal component itself.

[0041] The second and third extraction units extract the first frequency feature and the first time feature from the first reconstruction error, which makes it easier to categorize the abnormality factors than the first reconstruction error. Note that the order of the processes in steps S23 and S24 may be reversed in time series or may be simultaneous.

[0042] After step S23 is performed, the estimation unit 13 performs a vector search in the database 200 using a query vector based on the first frequency feature and the first time feature extracted in step S23 (step S24). Specifically, the estimation unit 13 generates one query vector based on the first frequency feature and the first time feature, and searches the database using the query vector. The database in the second embodiment stores a plurality of abnormality factors in association with a plurality of sample vectors based on combinations of a plurality of second frequency feature and a second time feature. The plurality of abnormality factors are categorized according to combinations of frequency distributions related to the second frequency feature and durations of abnormal components in the second time feature. The second frequency feature and the second time feature are each feature extracted from the second reconstruction error. Reducing the dimensions of the standard vectors stored in the database 200 enables faster vector search. Furthermore, it is possible to conserve storage resources in the database 200. The estimation unit 13 may generate one query vector from one feature that combines the first frequency feature and the first time feature, or may generate one query vector by combining a frequency vector generated based on the first frequency feature and a time vector generated based on the first time feature.

[0043] FIG. 9 is a diagram showing a typical example of multiple categorized abnormality factors. As shown in FIG. 9, the abnormality factors may be categorized according to the frequency peak distribution of the abnormal component and the duration of the abnormal component. FIG. 9 uses an elevator as an example of the target equipment 300, but is not limited to this. Specifically, the abnormality factors are categorized into at least three types: brake abnormality, guide rail distortion, and car wheel abnormality. Brake abnormality is a type in which the frequency distribution has multiple peaks and lasts for a long time. Guide rail distortion is a type in which the frequency distribution has a signal strength above a predetermined threshold across almost the entire audible range and lasts for a short time. Car wheel abnormality is a type in which the frequency distribution has a signal strength above a predetermined threshold across almost the entire audible range and lasts for a long time. A long time indicates a duration longer than the predetermined threshold. A short time indicates a duration shorter than the predetermined threshold. More specifically, the predetermined threshold for the duration is approximately one second. The predetermined threshold value may be set to any value that allows abnormality factors to be categorized.

[0044] Furthermore, the estimation unit 13 may perform vector search using a value based on the duration of signal strength exceeding the second threshold and a deviation based on the duration for two or more sample vectors associated with the same abnormality factor among the multiple sample vectors. For example, the duration is calculated as a first temporal feature by the third extraction unit 16. The value based on the duration is a value corresponding to a vectorized duration. However, the value based on the duration may be the duration itself. The deviation of the value based on the duration may be calculated for each of two or more sample vectors associated with the same abnormality factor. More specifically, the estimation unit 13 uses the standard deviation of the durations for two or more sample vectors associated with the same abnormality factor and sets a predetermined threshold for the standard deviation to associate the duration up to the predetermined threshold with the abnormality factor. By performing a vector search of the database for the duration, it is possible to improve the accuracy of searching for abnormality factors with varying durations.

[0045] When step S24 is performed, the display control unit 14 displays the abnormality factor of the target equipment 300 estimated in step S24 (step S25). As one example, the display control unit 14 displays the spectrum of the first frequency feature amount and the spectrum of the second frequency feature amount in an overlapping manner. As another example, the display control unit 14 displays the waveform of the first time feature amount and the second time feature amount in an overlapping manner.

[0046] FIG. 10 is a diagram illustrating a display screen I2 displaying an abnormality factor. The display screen I2 of FIG. 10 includes a display field I21. The display field I21 displays a character string representing the abnormality factor of the target equipment, such as "distortion of the guide rail." The display screen I2 also displays a spectrum and / or a waveform. For example, the display screen I2 may display a spectrum I22 in which the spectrum of the first frequency feature and the spectrum of the second frequency feature associated with the similarity vector are superimposed. The display screen I2 may also display a waveform I23 in which the waveform of the first time feature and the waveform of the second time feature associated with the similarity vector are superimposed. The vertical axis of the spectrum I22 represents signal strength, and the horizontal axis represents frequency. The vertical axis of the waveform I23 represents signal strength, and the horizontal axis represents time. By displaying the spectrum I22 and the waveform I23 on the display screen I2, the user can easily compare the first frequency feature related to the query vector with the second frequency feature related to the similar vector, and / or the first temporal feature related to the query vector with the second temporal feature related to the similar vector. Note that the display screen I2 may further display other information such as the date and time of occurrence of the second frequency feature and the second temporal feature.

[0047] This completes the processing procedure for estimating the cause of an abnormality in the target equipment 300.

[0048] The processing procedure for estimating the cause of an abnormality in the target equipment 300 shown in FIGS. 7 and 8 is an example, and various deletions, additions and / or modifications can be made without departing from the gist of the invention.

[0049] Here, the second embodiment will be compared with a comparative example in which an abnormality factor is estimated without using temporal features of sound data. In the comparative example, frequency features are extracted from sound data and used to estimate an abnormality factor. However, when the sound data is not based on a simple action but contains many variations, estimation of the abnormality factor is insufficient without information about time. Compared to the comparative example, according to this embodiment, frequency features and temporal features are extracted and used to estimate an abnormality factor with many variations.

[0050] (Third embodiment) The anomaly factor estimation systems in the first and second embodiments performed a vector search using a query vector based on the first reconstruction error. In the third embodiment, the query vector based on the first reconstruction error is associated with the first reconstruction error, the first frequency feature, and / or the first time feature and input into the database in order to newly store the query vector based on the first reconstruction error as a sample vector. The anomaly factor estimation system according to the third embodiment will be described below. However, components having the same functions as those in the first or second embodiment are denoted by the same reference numerals and will be described only when necessary.

[0051] Fig. 11 is a diagram illustrating an example of the hardware configuration of an information processing device 100 according to the third embodiment. As illustrated in Fig. 11, the processor 1 has functional components such as a first extraction unit 11, a reconstruction error calculation unit 12, an estimation unit 13, a display control unit 14, a second extraction unit 15, a third extraction unit 16, a determination unit 17, an input unit 18, etc.

[0052] The determination unit 17 determines whether the sound data is normal or not based on the first reconstruction error calculated by the reconstruction error calculation unit 12.

[0053] The input unit 18, in accordance with a user's instruction, inputs other sample vectors and abnormality factors in association with each other into the database 200. The other sample vectors represent sample vectors different from the sample vectors previously stored in the database 200. In particular, the query vector used for estimation that is to be stored in the database 200 as a new sample vector is called the other sample vector.

[0054] Fig. 12 is a diagram showing a processing procedure for estimating an abnormality factor of the target equipment 300 according to the third embodiment. Fig. 12 shows the processing procedure from step S12 shown in Fig. 3 to step S25 in Fig. 7. In the third embodiment, similarly to the second embodiment, the processor 1 extracts acoustic features from sound data in step S11 and calculates a reconstruction error in step S12.

[0055] 12, when step S12 is performed, the determination unit 17 determines whether the sound data used to extract the first acoustic feature in step S11 is normal based on the first reconstruction error calculated in step S12 (step S33). If the degree of abnormality of the first reconstruction error exceeds a third threshold, the determination unit 17 determines that the input sound data is abnormal. The degree of abnormality is, for example, the sum or maximum value of the first reconstruction errors. If it is determined that the sound data used to extract the first acoustic feature in step S11 is normal (step S33: YES), the determination unit 17 performs step S11. By not extracting the first frequency feature and the first time feature from the sound data determined to be normal, it is possible to improve the efficiency of estimating the abnormality factor of the abnormality factor estimation system 600.

[0056] If it is determined in step S11 that the sound data used to extract the first acoustic feature is not normal (step S33: NO), the second extraction unit extracts the first frequency feature from the first reconstruction error calculated in step S12 (step S34).

[0057] After step S34 is performed, the third extraction unit extracts a first temporal feature amount from the first reconstruction error calculated in step S12 (step S35).

[0058] The processing order of step S34 and step S35 may be reversed in time series or may be simultaneous. If the processing order of step S35 and step S34 is reversed, step S35 may be executed when it is determined that the sound data used to extract the first acoustic feature in step S11 is abnormal. Furthermore, when it is determined that the sound data is abnormal, the display control unit 14 may display the reconstruction error, frequency feature, time feature and / or abnormality factor related to the sound data.

[0059] This completes the processing procedure for estimating the cause of an abnormality in the target equipment 300.

[0060] Next, a description will be given of a processing procedure for storing the estimated abnormality factor of the target equipment 300 in association with other sample vectors in the database 200. The following description starts after the abnormality factor of the target equipment 300 is estimated, in other words, from the point in time when the processing in FIG. 12 is completed.

[0061] FIG. 13 is a diagram schematically showing a processing procedure related to the input unit 18 according to the third embodiment.

[0062] The input unit 18 inputs other sample vectors and abnormality factors to the database 200 in accordance with a user instruction. For example, the input unit 18 inputs a query vector calculated by the estimation unit 13 as another sample vector to the database 200 in accordance with a user instruction. However, the input unit 18 may input a query vector calculated by the estimation unit 13 and then stored in the storage device 2 as another sample vector to the database 200. Furthermore, the input unit 18 may input the abnormality factors displayed by the display control unit 14 in association with the other sample vectors in accordance with a user instruction. However, the input unit 18 may input abnormality factors other than the abnormality factors displayed by the display control unit 14.

[0063] This completes the processing procedure for storing the estimated abnormality factors of the target equipment 300 and other sample vectors in association with each other in the database 200.

[0064] Note that the processing procedure for estimating the abnormality factor of the target facility 300 shown in FIG. 12 and the processing procedure for inputting other sample vectors and abnormality factors in association with each other into the database 200 shown in FIG. 13 are merely examples, and various deletions, additions, and / or modifications can be made without departing from the spirit of the invention.

[0065] According to the third embodiment, it is possible to extract frequency features and time frequencies according to the degree of abnormality. This makes it possible to efficiently extract frequency features and time features for abnormal sound data from acquired sound data, and ultimately to store a query vector based on the extracted frequency features and time features as another sample vector in the database 200.

[0066] (Fourth embodiment) The abnormality factor estimation systems in the first, second, and third embodiments judged abnormalities based on the length of the acquired sound data. In the fourth embodiment, when the operating sound of the target equipment 300 lasts for a long time, the abnormality factor estimation system divides the acquired sound data into shorter pieces and judges abnormalities in the sound data. Specifically, approximately 10 seconds of sound data is divided into sound data pieces each about one second long. Note that the length of the acquired sound data and the length of the divided sound data are not limited to the above lengths. The abnormality factor estimation system according to the fourth embodiment will be described below. However, components having the same functions as those in the first, second, or third embodiment are designated by the same reference numerals and will be described only when necessary.

[0067] FIG. 14 is a diagram illustrating an example of the hardware configuration of an information processing device 100 according to the fourth embodiment. As shown in FIG. 14, the processor 1 has functional components such as a first extraction unit 11, a reconstruction error calculation unit 12, an estimation unit 13, a display control unit 14, a second extraction unit 15, a third extraction unit 16, a determination unit 17, an input unit 18, as well as a division unit 19 and a combination unit 20.

[0068] The dividing unit 19 divides the sound data into a plurality of segments separated by a predetermined time.

[0069] When the judgment unit 18 judges that a plurality of consecutive segments divided by the division unit 19 in time series are not normal, the combination unit 20 combines a plurality of reconstruction errors corresponding to the plurality of segments to generate a first reconstruction error.

[0070] 15 and 16 are diagrams showing a processing procedure for estimating the cause of an abnormality in the target equipment 300 according to the fourth embodiment.

[0071] As shown in FIG. 15, the dividing unit 19 divides the sound data into multiple segments separated by a predetermined time interval (step S41). More specifically, the sound data is divided along a time series. The multiple segments may have the same data type as the sound data. In other words, each of the multiple segments may be treated as sound data with a shorter duration than the sound data before division. The dividing unit 19, for example, edits the digital signal of the portion near the start and end of each of the multiple segments in the time series so that it can be Fourier transformed. While an elevator will be used as an example of the target facility 300, this is not limiting. To estimate the cause of an abnormality, such as the elevator car's ascending and descending distance or a behavior different from the opening and closing of the elevator doors and the ascending and descending of the car, sound data of a length appropriate for each abnormality is required. By dividing the sound data into multiple segments, it is possible to categorize the abnormalities according to the target behavior.

[0072] The predetermined time may be set to any time shorter than the length of the sound data. The divided segments may be subjected to the following processing while retaining information about the time series of the sound data. For example, the information about the time series is information about the time of the sound data before division. For another example, if division is performed along the time series of the sound data, the information is the order in which the division unit 19 divided the multiple segments.

[0073] After step S41 is performed, the first extraction unit 11 extracts a first acoustic feature from each of the plurality of segments divided in step S41 (step S42). In the fourth embodiment, as in the first embodiment, the first extraction unit 11 extracts an acoustic feature from each of the plurality of segments divided.

[0074] When step S42 is performed, the reconstruction error calculation unit 12 calculates, based on the reconstruction model, a plurality of reconstruction errors corresponding to the plurality of first acoustic features extracted in step S42 (step S43). The reconstruction model is, for example, a machine learning model trained by optimization calculation based on training data that are ground-truth acoustic features extracted from each of a plurality of normal segments into which normal sound data collected when the target equipment 300 is operating normally is divided. By using the reconstruction model trained with the normal segments, it is possible to improve the efficiency of detecting abnormal components of a duration shorter than that of the sound data.

[0075] When step S43 is performed, the determination unit 17 determines whether each of the multiple segments divided in step S41 is normal or not based on the first reconstruction error calculated in step S43 (step S44). If the degree of abnormality of at least one of the multiple reconstruction errors exceeds a third threshold, the determination unit 17 may determine that the segment used to calculate the reconstruction error whose degree of abnormality exceeds the third threshold is abnormal. The degree of abnormality is, for example, the sum or maximum value of one of the multiple reconstruction errors. The third threshold may be set to a value different from that in the third embodiment. The determination unit 17 determines whether each of the multiple segments is normal or not.

[0076] If it is determined that all of the multiple segments divided in step S41, which were used to calculate the first reconstruction error in step S43, are normal (step S44: YES), the dividing unit 19 performs step S41.

[0077] If at least one of the segments divided in step S41 and used to calculate the first reconstruction error in step S43 is determined to be abnormal (step S44: NO), the determination unit 17 determines whether multiple segments consecutive in time series are normal (step S45). Multiple segments consecutive in time series are two or more segments that are divided from the same sound data and are consecutive in time series of the sound data. The determination unit 17 determines whether the segments determined to be abnormal in step S44 are consecutive in time series, for example, based on information about the time series of the segments. For example, the determination of whether the segments are consecutive in time series is made based on information about the time series.

[0078] If multiple consecutive segments in chronological order are determined to be normal (step S45: YES), the second extraction unit 15 performs the same processing as step S34 and subsequent steps in the third embodiment on each of the segments determined to be abnormal in step S44.

[0079] If it is determined that multiple segments consecutive in time series are not normal (step S45: NO), the combiner 20 combines the multiple reconstruction errors calculated in step S43 to generate a first reconstruction error (step S46). The combiner 20, for example, combines multiple segments consecutive in time series to generate a first reconstruction error. The combiner 20 may combine all of the multiple reconstruction errors to generate one first reconstruction error, or may combine some of the multiple reconstruction errors to generate multiple first reconstruction errors. By combining multiple reconstruction errors to generate a first reconstruction error, it is possible to perform a vector search for abnormal factors over a longer period of time than the divided period of time.

[0080] When step S46 is performed, the second extraction unit 15 performs the same processing as that from step S34 onward in the third embodiment on the first reconstruction error generated in step S46. If multiple first reconstruction errors are generated in step S46, the second extraction unit 15 performs the same processing as that from step S34 onward in the third embodiment on each of the multiple first reconstruction errors.

[0081] This completes the processing procedure for estimating the cause of an abnormality in the target equipment 300.

[0082] 15 and 16 are merely examples of the processing procedure for estimating the cause of an abnormality of the target equipment 300, and various deletions, additions, and / or modifications are possible without departing from the spirit and scope of the invention. In the fourth embodiment, the database may store a sample vector based on a first reconstruction error calculated based on divided segments and an abnormality cause in association with each other. As an example, the sample vector is based on a segment in which all of the divided segments are joined, a segment in which some of the segments are joined, and / or one segment. Furthermore, the hardware configuration including the splitting unit 17 can also be implemented in a hardware configuration that does not include the second extraction unit 15, the third extraction unit 16, and the determination unit 17. Specifically, the hardware configuration of the first embodiment can also be applied to a configuration in which the splitting unit 17 is added to the hardware configuration of the first embodiment.

[0083] According to the fourth embodiment, by dividing the collected sound data into short time segments, it is possible to estimate the cause of an abnormality over a short period of time. Furthermore, by combining the divided segments in chronological order, it is possible to estimate the cause of an abnormality over a longer period of time than the divided time.

[0084] (Fifth embodiment) The abnormality factor estimation systems in the first, second, third, and fourth embodiments use a trained reconstruction model to calculate the reconstruction error. The abnormality factor estimation system in the fifth embodiment generates a trained reconstruction model. The abnormality factor estimation system according to the fifth embodiment will be described below. However, components having the same functions as those in the first, second, third, or fourth embodiment are designated by the same reference numerals and will be described only when necessary.

[0085] Fig. 17 is a diagram illustrating an example of the hardware configuration of an information processing device 100 according to the fifth embodiment. As illustrated in Fig. 17, the processor 1 has functional components such as a first extraction unit 11, a reconstruction error calculation unit 12, an estimation unit 13, and a display control unit 14, as well as a training unit 21 and a model storage unit 22.

[0086] The training unit 21 receives the correct acoustic features based on the training data, which are the correct acoustic features extracted by the first extraction unit 11, and generates a trained reconstruction model that outputs the encoding and decoding data of the input correct acoustic features.

[0087] The model storage unit 22 stores the reconstruction model. As an example, the model storage unit 22 stores the reconstruction model in the storage device 2.

[0088] 18 and 19 are diagrams showing a processing procedure for generating a reconstructed model according to the fifth embodiment. The processing for generating a reconstructed model according to the fifth embodiment will be described below with reference to the flow of FIGS. 18 and 19. The processing for generating a reconstructed model should be performed before the processing for estimating an abnormality factor, which is performed in the first to fourth embodiments. More specifically, the reconstructed model in the first to fourth embodiments is generated in the fifth embodiment.

[0089] 18 , the first extraction unit 11 extracts correct acoustic features, which are training data, from sound data collected when the target equipment 300 is operating normally (step S51). The sound data may be collected from the target equipment 300 for which an abnormality factor is to be estimated. Specifically, the correct acoustic features are extracted from sound data of the target equipment 300 collected from the microphone 34. This makes it possible to acquire training data having features specific to the installation environment of the target equipment 300 and / or the mechanism of the target equipment 300.

[0090] The training data may be stored in the storage device 2 or the database 200. By using the stored training data to train a model to be used for other target facilities having a similar installation environment and / or mechanism, it is possible to shorten the time required to collect training data and enrich the training data.

[0091] After step S51 is performed, the training unit 21 inputs acoustic features based on the training data extracted in step S51 and generates a trained reconstruction model that outputs encoding and decoding data of the input acoustic features (step S52). More specifically, the training unit 21 inputs the training data to the model to be trained. Next, the training unit 21 evaluates the data output from the model to be trained using a loss function. Examples of the loss function that can be used include mean squared error (MSE) and cross entropy loss (CE Loss). Finally, the training unit 21 determines whether to end the training based on the evaluation. If the training is not to be ended, the parameters of the model to be trained are updated and training data is input to the model to be trained. If the training is to be ended, step S53 is performed. By using training data based on normal sound data collected from the target equipment 300 to train the model, it is possible to improve the accuracy of the generated reconstruction model in extracting abnormal components that are independent of the installation environment and / or mechanism of the target equipment 300.

[0092] The training unit 21 may generate a reconstructed model by performing additional learning on the reconstructed model using training data related to the installation environment and / or mechanism of the target facility 300, or may generate a reconstructed model by using training data on the installation environment and / or mechanism of the target facility 300 on an unlearned reconstructed model. In addition, it is preferable that an autoencoder in which a decoder and an encoder are connected is used as the model to be trained.

[0093] When step S52 is performed, the model storage unit 22 stores the reconstruction model generated in step S52 (step S53). As an example, the model storage unit 22 stores the generated reconstruction model in the storage device 2. The reconstruction error calculation unit 12 uses, for example, the stored reconstruction model. By using the reconstruction model trained by the training unit 21 in the reconstruction error calculation unit 12 in the third embodiment, a highly versatile first reconstruction error that is independent of the installation environment and / or mechanism of the target facility 300 can be input as a sample vector.

[0094] This completes the processing procedure for estimating the cause of an abnormality in the target equipment 300.

[0095] The processing procedure for estimating the cause of an abnormality in the target equipment 300 shown in FIGS. 18 and 19 is an example, and various deletions, additions and / or modifications can be made without departing from the gist of the invention.

[0096] According to the fifth embodiment, it is possible to generate a reconstructed model to be used in an anomaly cause estimation system. Furthermore, by generating a reconstructed model using correct acoustic features of the target facility 300 as training data, it is possible to improve the accuracy of detecting an anomaly in the target facility 300.

[0097] Although several embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These novel embodiments can be embodied in various other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their modifications are included within the scope and spirit of the invention, and are also included in the scope of the invention and its equivalents as defined in the claims. [Explanation of symbols]

[0098] 1...processor, 2...storage device, 3...input device, 4...communication device, 5...display device, 11...first extraction unit, 12...reconstruction error calculation unit, 13...estimation unit, 14...display control unit, 34...microphone, 100...information processing device, 200...database, 300...target equipment, 600...anomaly cause estimation system

Claims

1. a first extraction unit that extracts a first acoustic feature based on frequency, time, and signal intensity from sound data related to the target facility; a reconstruction error calculation unit that calculates a first reconstruction error, which is a difference between a reconstructed feature obtained by reconstructing the first acoustic feature based on a reconstruction model that performs encoding and decoding of the acoustic feature, and the first acoustic feature; a database that stores a plurality of abnormality factors and a plurality of sample vectors based on a plurality of second reconstruction errors in association with each other; an estimation unit that estimates an abnormality factor of the target equipment by performing a vector search in the database using a query vector based on the first reconstruction error; An abnormality cause estimation system comprising:

2. 2. The abnormality factor estimation system according to claim 1, wherein the estimation unit estimates, as the abnormality factor of the target equipment, an abnormality factor corresponding to a specific sample vector among the plurality of sample vectors whose similarity to the query vector exceeds a first threshold.

3. The abnormality factor estimation system according to claim 1 , further comprising a display unit that displays the abnormality factor of the target equipment.

4. 4. The abnormality factor estimation system according to claim 3, further comprising a display unit that displays a spectrogram based on the first reconstruction error and a spectrogram based on the second reconstruction error associated with the specific sample vector side by side.

5. The abnormality factor estimation system according to claim 4 , wherein the display unit further displays information on the date and time of occurrence of the specific sample vector.

6. The abnormality factor estimation system according to claim 1 , wherein the estimation unit performs a vector search in the database with the query vector based on a spectrogram related to the first reconstruction error.

7. a second extraction unit that extracts a first frequency feature from the first reconstruction error; a third extraction unit that extracts a first temporal feature from the first reconstruction error, the sample vector is a vector based on a second frequency feature and a second time feature extracted from the second reconstruction error, The abnormality factor estimation system according to claim 1 , wherein the estimation unit performs a vector search using the query vector based on the first frequency feature and the first time feature.

8. 8. The abnormality factor estimation system according to claim 7, wherein the plurality of abnormality factors are categorized according to a combination of a frequency distribution related to the second frequency feature and a duration of an abnormal component in the second time feature.

9. the plurality of abnormal factors include a first type, a second type, and a third type; the first type is a type in which the frequency distribution has a plurality of peaks and the duration is equal to or longer than a predetermined time, The second type is a type in which the frequency distribution has the signal strength equal to or greater than a predetermined threshold value in substantially the entire audible range and the duration is equal to or less than the predetermined time, 9. The abnormality factor estimation system according to claim 8, wherein the third type is a type in which the frequency distribution has the signal strength equal to or greater than the predetermined threshold value over substantially the entire audible range and the duration is equal to or greater than the predetermined time.

10. 8. The abnormality factor estimating system according to claim 7, further comprising a display unit that displays spectra of the first frequency feature amount and the second frequency feature amount side by side or in an overlapping manner, and / or displays waveforms of the first time feature amount and the second time feature amount side by side or in an overlapping manner.

11. a determination unit that determines whether the sound data is normal or not based on the first reconstruction error, the second extraction unit extracts the first frequency feature from the first reconstruction error when the sound data is determined to be abnormal; The abnormality factor estimation system according to claim 7 , wherein the third extraction unit extracts the first temporal feature amount from the first reconstruction error when the sound data is determined to be abnormal.

12. The abnormality factor estimation system according to claim 11 , wherein the determination unit determines that the input sound data is abnormal when the degree of abnormality of the first reconstruction error exceeds a third threshold value.

13. 2. The abnormality factor estimating system according to claim 1, further comprising an input unit for inputting other sample vectors and abnormality factors in association with each other into said database in accordance with a user's instruction.

14. 12. The abnormality factor estimation system according to claim 11, further comprising a display unit that, when the sound data is determined to be abnormal, displays the reconstruction error, the frequency feature amount, the time feature amount, and / or the abnormality factor related to the sound data.

15. a dividing unit that divides the sound data into a plurality of segments separated by a predetermined time interval, The abnormality factor estimation system according to claim 1 , wherein the first extractor extracts the first acoustic feature from each of the plurality of segments.

16. a dividing unit that divides the sound data into a plurality of segments separated by a predetermined time; 12. The abnormality factor estimation system according to claim 11, further comprising a combining unit that, when the plurality of segments consecutive in time series are determined to be abnormal, combines a plurality of reconstruction errors corresponding to the plurality of segments, respectively, to generate the first reconstruction error.

17. a training data storage unit that stores, as training data, acoustic features extracted from the normal sound data collected when the target equipment is operating normally; a training unit that receives the acoustic features based on the training data and generates the reconstruction model trained to output encoding and decoding data of the input acoustic features; The abnormality factor estimation system according to claim 1 , further comprising: a model storage unit that stores the reconstructed model.

18. 2. The abnormality factor estimation system according to claim 1, wherein the estimation unit performs vector search for two or more sample vectors associated with the same abnormality factor among the plurality of sample vectors, using a value based on a duration of signal strength exceeding a second threshold and a deviation of the value based on the duration.

19. The computer extracting a first acoustic feature based on frequency, time, and signal strength from sound data related to the target facility; calculating a first reconstruction error that is a difference between a reconstructed feature obtained by reconstructing the first acoustic feature based on a reconstruction model that performs encoding and decoding of the acoustic feature and the first acoustic feature; storing the plurality of sample vectors based on the plurality of second reconstruction errors in association with the plurality of abnormality factors in a database; performing a vector search in the database using a query vector based on the first reconstruction error to estimate an abnormality factor of the target equipment; The abnormality factor estimation method comprises:

20. On the computer, a function of extracting a first acoustic feature based on frequency, time, and signal strength from sound data related to the target facility; a function of calculating a first reconstruction error, which is a difference between a reconstructed feature obtained by reconstructing the first acoustic feature based on a reconstruction model that performs encoding and decoding of the acoustic feature, and the first acoustic feature; a function of previously associating a plurality of abnormality factors with a plurality of sample vectors based on a plurality of second reconstruction errors and storing the associations in a database; a function of estimating an abnormality factor of the target equipment by performing a vector search in the database using a query vector based on the first reconstruction error; An abnormality cause estimation program that achieves this.

Images