Equipment anomaly detection method and related device and system

Abstract

The invention discloses an equipment anomaly detection method and a related device and system, and the method comprises the steps: obtaining a plurality of to-be-detected sounds collected at a target detection point of to-be-detected equipment; obtaining an abnormal score of each to-be-detected sound; wherein the abnormal score of the to-be-detected sound represents the abnormal degree of the to-be-detected sound compared with the target measurement point when the to-be-detected equipment is normal; based on the feature distance between the sound features of the to-be-detected sound, fusing the abnormal scores of at least part of the to-be-detected sound to obtain a fused score; and determining whether the target measuring point runs abnormally based on the fusion score. According to the scheme, the accuracy of equipment anomaly detection can be improved.

Description

Technical Field

[0001] The present application relates to the technical field of fault monitoring, and in particular to a method for detecting anomalies in equipment and related devices and systems. Background Art

[0002] The normal operation of industrial equipment is an important prerequisite for ensuring operational efficiency. Timely understanding of equipment operation status can reduce the probability of equipment failure and can also take timely remedial measures when equipment failure occurs.

[0003] At present, the existing technology usually directly uses sound signals to determine whether an abnormality occurs in the device. However, since the sound signal is prone to fluctuations, the abnormality detection result of the existing technology is not accurate.

[0004] Therefore, how to improve the accuracy of equipment anomaly detection has become an urgent problem to be solved. Summary of the invention

[0005] The main technical problem solved by the present application is to provide a device anomaly detection method and related devices and systems, which can improve the accuracy of device anomaly detection.

[0006] In order to solve the above technical problems, the first aspect of the present application provides a device abnormality detection method, which includes: obtaining a number of sounds to be tested collected at a target measuring point of the device to be tested; obtaining an abnormality score of each sound to be tested; wherein the abnormality score of the sound to be tested represents: the degree of abnormality of the sound to be tested compared to the sound of the target measuring point when the device to be tested is operating normally; based on the feature distance between the sound features of the sound to be tested, the abnormality scores of at least part of the sounds to be tested are fused to obtain a fused score; based on the fused score, it is determined whether the target measuring point is operating abnormally.

[0007] In order to solve the above technical problems, the second aspect of the present application provides an equipment abnormality detection device, which includes a first acquisition module, a second acquisition module, a fusion module and a determination module; the first acquisition module is used to acquire a number of sounds to be tested collected at a target measuring point of the equipment to be tested; the second acquisition module is used to acquire the abnormality score of each sound to be tested; wherein the abnormality score of the sound to be tested represents: the degree of abnormality of the sound to be tested compared to the sound of the target measuring point when the equipment to be tested is operating normally; the fusion module is used to fuse the abnormality scores of at least part of the sounds to be tested based on the feature distance between the sound features of the sounds to be tested to obtain a fusion score; the determination module is used to determine whether the target measuring point is operating abnormally based on the fusion score.

[0008] In order to solve the above technical problems, the third aspect of the present application provides an electronic device, including a memory and a processor coupled to each other, and the processor is used to execute program instructions stored in the memory to implement the device abnormality detection method in the above first aspect.

[0009] In order to solve the above-mentioned technical problems, the fourth aspect of the present application provides an equipment abnormality detection system, which includes at least an optical fiber, a demodulation device and a detection device. The demodulation device is communicatively connected between the optical fiber and the detection device. The optical fiber is distributed with a number of sensing points, which are respectively laid at each point to be tested of the equipment to be tested. The demodulation device is used to demodulate the operating sound of the equipment to be tested at the point to be tested based on the optical signal of the optical fiber at the sensing point, and the detection device includes the electronic device of the third aspect mentioned above.

[0010] In order to solve the above technical problems, the fifth aspect of the present application provides a computer-readable storage medium on which program instructions are stored. When the program instructions are executed by a processor, the device abnormality detection method in the above first aspect is implemented.

[0011] The above scheme obtains several sounds to be tested collected at the target measuring point of the device to be tested, and obtains the abnormality score of each sound to be tested, and then fuses the abnormality scores of at least part of the sounds to be tested based on the characteristic distance between the sound features of the sounds to be tested, obtains the fusion score, and then determines whether the target measuring point is operating abnormally based on the fusion score; wherein the abnormality score of the sound to be tested represents: the abnormality degree of the sound to be tested compared with the sound of the target measuring point when the device to be tested is operating normally. Therefore, in the case of obtaining the abnormality score of each sound to be tested, the abnormality score of at least part of the sounds to be tested is further fused by the characteristic distance between the sound features of the sounds to be tested. On the one hand, using multiple sounds to be tested at the same measuring point as the data source of the device abnormality detection judgment standard can reduce the interference of invalid data collected under the influence of sound signal fluctuations, and thus can improve the accuracy of device abnormality detection from the data validity level; on the other hand, by combining the abnormality score of the sound to be tested and the characteristic distance between the sound features, compared with directly using the abnormality score of a single dimension as the judgment criterion for device abnormality detection, the accuracy of device abnormality detection can be improved from a fine-grained level. Therefore, the accuracy of device anomaly detection can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 This is a flow chart of an embodiment of a method for detecting abnormality of a device according to the present application;

[0013] Figure 2 A process diagram of an embodiment of a method for training a device feature extraction model;

[0014] Figure 3 This is a schematic diagram of the framework of an embodiment of the device abnormality detection device of the present application;

[0015] Figure 4 A schematic diagram of a framework of an embodiment of an electronic device of the present application;

[0016] Figure 5 This is a schematic diagram of the framework of an embodiment of the device abnormality detection system of the present application;

[0017] Figure 6 for Figure 5 A schematic diagram of the structure of an embodiment of a device abnormality detection system;

[0018] Figure 7 A schematic diagram of a framework of an embodiment of a computer-readable storage medium of the present application. DETAILED DESCRIPTION

[0019] The scheme of the embodiment of the present application is described in detail below in conjunction with the drawings of the specification.

[0020] In the following description, for the purpose of explanation rather than limitation, specific details such as specific system structures, interfaces, and technologies are provided to facilitate a thorough understanding of the present application.

[0021] The term "multiple" herein means two or more than two. In addition, the term "at least one" herein means any combination of at least two of any one or more of a plurality of, for example, including at least one of A, B, and C, and may mean including any one or more elements selected from the set consisting of A, B, and C.

[0022] See also Figure 1 , Figure 1 This is a flow chart of an embodiment of a method for detecting abnormality of a device of the present application, specifically, comprising the following steps:

[0023] Step S11: Acquire a number of sounds to be tested collected at target test points of the device to be tested.

[0024] In the embodiments of the present disclosure, the device to be tested may include equipment for transporting materials such as belt conveyors and screw conveyors, and may also include equipment for daily life such as escalators and baggage carousels. The types and application scenarios of the device to be tested are not specifically limited. In addition, the device to be tested may include one or more measurement points for sound collection, and when the device to be tested includes multiple measurement points, one of the measurement points may be selected as the target measurement point; or, some of the measurement points may be selected as the target measurement points; or, each measurement point may be selected as the target measurement point. The specific setting may be based on the detection requirements of the device to be tested, and is not limited here.

[0025] In one implementation scenario, the target sensing point of the device to be tested is provided with a target sensing point of an optical fiber 51. In order to obtain a number of sounds to be tested at the target sensing point, the operating sound of the target measuring point obtained by demodulating the optical signal collected from the target sensing point by the demodulation device 52 can be received as the sound to be tested. Therefore, through the communication between the optical fiber 51 and the demodulation device 52, the demodulation device 52 can collect the optical signal reflected by the optical fiber 51 from the target sensing point of the optical fiber 51, and then can obtain the sound to be tested obtained by demodulating the optical signal from the demodulation device 52, thereby realizing the automation process of sound collection for the device to be tested. The sound collection process realized by the optical fiber 51 and the demodulation device 52 can refer to the relevant description in the subsequent embodiments, which will not be repeated here.

[0026] In a specific implementation scenario, several sounds to be tested can be obtained by demodulating the optical signal collected once several times by the demodulation device 52. Specifically, the optical signal collected once is divided into several segments by the demodulation device 52, and the demodulation device 52 demodulates the several segments respectively to obtain several sounds to be tested.

[0027] In another specific implementation scenario, unlike the aforementioned method in which the demodulation device 52 demodulates the optical signal collected once for several times to obtain the plurality of sounds to be tested, the demodulation device 52 may also demodulate the optical signals collected several times to obtain the plurality of sounds to be tested respectively.

[0028] In an implementation scenario, after obtaining the sound to be tested at the target test point, the sound to be tested can be subjected to sound denoising, and then the sound to be tested after the sound denoising is used as the sound to be tested in subsequent steps 12, 13, and 14. Among them, the sound denoising method may include but is not limited to: wavelet denoising, median filter, Kalman filter, etc. Therefore, by performing sound denoising on the sound to be tested, filtering the noise signal of the sound to be tested, and retaining the effective signal of the sound to be tested, the robustness of the device abnormality detection can be improved.

[0029] In a specific implementation scenario, the wavelet denoising method is used to perform sound denoising on the sound to be tested. Specifically, the signal f(i) representing the sound to be tested is divided into two parts: an effective signal and noise:

[0030] f(i)=x(i)+w(i)

[0031] Where i=1,2,…,n, n is the signal length, x(i) is the effective signal component to be retained, and w(i) is the noise component. After selecting a suitable wavelet basis, the signal f(i) is decomposed by wavelet to obtain a series of wavelet coefficients, and the wavelet coefficients are threshold processed, that is, the coefficients smaller than the threshold are removed as interference noise, and then the inverse wavelet transform is performed to obtain the signal after noise removal. Wavelet coefficient threshold processing methods include hard threshold, soft threshold and other methods. As a possible example, in order to reduce the risk of discontinuous jumps caused by threshold processing, an improved threshold processing method is adopted:

[0032]

[0033] Among them, the threshold λ is obtained by the VisuShrink calculation method:

[0034]

[0035] For other threshold processing methods, please refer to the relevant technology, and no further examples are given here.

[0036] Step S12: obtaining an abnormality score of each sound to be tested; wherein the abnormality score of the sound to be tested represents: the abnormality degree of the sound to be tested compared with the sound of the target measuring point when the device to be tested is operating normally.

[0037] In one implementation scenario, the calibration value of the device under test in normal operation can be obtained so that the device under test can be operated based on the calibration value, and then the operating sound of the target measuring point when the device under test is normal can be obtained.

[0038] In one implementation scenario, the abnormality score of each sound to be tested is obtained through the speech features of each sound to be tested. Among them, the speech features of the sound to be tested may include but are not limited to: FBank features (Filter Bank, filter bank coefficients), MFCC features (Mel-Frequency Cepstral Coefficients, Mel-Frequency Cepstral Coefficients), PLP features (Perceptual Linear Predictive, perceptual linear prediction), etc. Taking the extraction of MFCC features based on the sound to be tested as an example, the MFCC feature acquisition process includes pre-emphasis, framing, windowing, Mel filter bank, discrete cosine transform and other speech feature extraction processing processes, specifically including: pre-emphasis of the sound to be tested to obtain the pre-emphasized sound signal:

[0039] y(n)=x(n)-a*x(n-1)

[0040] Among them, the value of a is between 0.9 and 1.

[0041] The pre-emphasized sound signal is divided into 20-30ms frames, and different frames overlap to a certain extent, for example, different frames can overlap 10ms. By dividing the signal into short-time frames, and performing Fourier transform in the short-time frames, adjacent frames can be connected to obtain a good approximation of the frequency contour of the signal. Then, the spectrum leakage is suppressed by windowing, and each frame is multiplied by a window function (such as a Hamming window) to increase the continuity of the left and right ends of the frame. The specific calculation is as follows:

[0042]

[0043]

[0044] By performing fast Fourier transform on each frame signal after frame division and windowing, each frame sound signal can be converted from the time domain to the frequency domain and the amplitude of each point in the frequency domain is calculated to obtain the amplitude spectrum, and then the triangular bandpass filter group (Mel filter group) is used to convert the sound spectrum. The calculation formula of the triangular bandpass filter is as follows:

[0045]

[0046] Where m = 1, 2, ..., M, M is the number of filters, k is the weighting coefficient of the amplitude spectrum after filtering, and f(m) is the center frequency of the Mel bandpass filter. The conversion relationship between Mel frequency and physical frequency is as follows:

[0047]

[0048] The center frequency calculation formula is as follows:

[0049]

[0050]

[0051] Where, f is the physical frequency, where f l and f h are the lower and upper limits of the filter frequency, N is the number of sampling points, and f s is the sampling frequency. By multiplying the amplitude spectrum and the triangular filter bank, a smooth sound amplitude spectrum is obtained. Then, the MFCC feature C(n) of the sound to be tested is obtained by discrete cosine transform:

[0052]

[0053] Among them, E kis the logarithmic amplitude of the output of the kth Mel filter bank, and M is the number of Mel filters. It should be noted that the acquisition process of Fbank features and PLP features also includes pre-emphasis, framing, windowing, Fourier transform and other processes. The specific acquisition process of Fbank features and PLP features can refer to the relevant technology and will not be repeated here.

[0054] In an implementation scenario, the operating sound of the target measuring point when the device under test is normal and each sound to be tested can be directly compared manually or with the assistance of a machine, and then each sound to be tested can be scored according to the comparison result to obtain an abnormality score.

[0055] In another implementation scenario, in order to improve the objectivity of obtaining the abnormality score of the sound to be tested, the abnormality score of the sound to be tested is obtained with the assistance of an abnormality detection model, and the abnormality detection model is obtained by adjusting the parameters of at least the operating sound of the target measuring point when the device to be tested is operating normally. Specifically, the sound to be tested is processed based on the abnormality detection model to obtain the predicted sound corresponding to the sound to be tested under the parameter conditions of the abnormality detection model. Therefore, the predicted sound corresponding to the sound to be tested under the parameter conditions of the abnormality detection model is obtained through the abnormality detection model, so that the degree of abnormality of the sound to be tested compared to the operating sound of the target measuring point when the device to be tested is normal can be obtained through the difference between the sound to be tested and the predicted sound, and then the abnormality score of the sound to be tested is obtained with the assistance of the abnormality detection model, which can improve the objectivity of device abnormality detection compared to the possible objectivity caused by manual scoring.

[0056] In a specific implementation scenario, the anomaly detection model is obtained by adjusting parameters based on the GMM model (Gaussian mixture model). The GMM model is defined as follows:

[0057]

[0058] Among them, the GMM model includes K Gaussian distributions, φ i represents the weight of each Gaussian distribution, μ i and Σ i Respectively represent the mean vector and covariance matrix of the i-th Gaussian distribution. At least the operating sound of the target measuring point when the device under test is normal is used as the modeling input of the GMM. That is to say, when the device under test includes multiple measuring points, the operating sounds of multiple measuring points including the target measuring point when the device under test is normal can also be used as the modeling input of the GMM, and the EM algorithm is used to calculate the parameters of the GMM model based on the modeling input data of the GMM to obtain the anomaly detection model. The specific process of adjusting the parameters of the GMM model using the EM algorithm can refer to the relevant technology and will not be repeated here. Then, the sound to be tested is processed by the anomaly detection model obtained after the parameter adjustment, and the corresponding sound to be tested under the parameter conditions of the anomaly detection model can be obtained:

[0059]

[0060] in, is the posterior probability of the sound to be tested on the i-th Gaussian distribution under the condition of the anomaly detection model parameters, and is calculated by combining the posterior probability of the sound to be tested on the i-th Gaussian distribution with the mean vector μ i Multiply them to get the prediction vector of the sound to be tested on the i-th Gaussian distribution, then sum the prediction vectors of the sound to be tested on each Gaussian distribution, and use the summed vector as the predicted sound corresponding to the sound to be tested under the parameter conditions of the GMM model. The predicted sound represents the sum of the proportions of the data points of the sound to be tested as a data point allocated to each Gaussian distribution of the GMM model. As a possible implementation method, the sound to be tested C and the predicted sound can be directly The difference R is used as the abnormality score of the sound to be tested:

[0061]

[0062] As another possible implementation, R may be further normalized, and the normalized result is used as the abnormality score F of the sound to be tested:

[0063]

[0064] Wherein, D is the vector dimension of the feature vector (such as MFCC feature vector) representing the sound to be tested. Of course, the anomaly detection model can also be obtained by adjusting parameters based on HMM (Hidden Markov Model) or K-Means model, etc. The specific process can refer to the parameter adjustment related technology of the corresponding model, and no examples are given here one by one.

[0065] Step S13: Based on the feature distances between the sound features of the sound to be tested, the abnormal scores of at least part of the sound to be tested are fused to obtain a fused score.

[0066] In one implementation scenario, the sound features of the sound to be tested are obtained by extracting features from the sound to be tested by a feature extraction model, the feature extraction model is trained by sample device sounds, the sample device sounds are annotated with device states that represent whether the operation is abnormal, and the feature extraction model is specifically trained based on sound triples, the sound triples include: any sample device sound as a sample target sound, a sample device sound with the same device state as the sample target sound as a sample positive sound, and a sample device sound with a different device state from the sample target sound as a sample negative sound. Therefore, by using sample device sounds in different device states that represent whether the operation is abnormal as the training input of the feature extraction model, the feature extraction model can be trained to have the ability to make sample device sounds in the same device state have a high degree of intra-class aggregation, and sample device sounds in different device states have a high degree of differentiation, and then the sound features of several sounds to be tested are obtained by the trained feature extraction model, which can improve the classification accuracy of the degree of aggregation and differentiation between sound features.

[0067] In a specific implementation scenario, the training steps of the feature extraction model include: extracting the sound features of the sample target sound as the target sound features through the feature extraction model, extracting the sound features of the sample positive sound as the positive sound features through the feature extraction model, and extracting the sound features of the sample negative sound as the negative sound features through the feature extraction model, and then obtaining the first distance between the positive sound features and the target sound features, and obtaining the second distance between the negative sound features and the target sound features, and then adjusting the network parameters of the feature extraction model based on the difference between the first distance and the second distance. Therefore, through the difference between the first distance and the second distance, the feature extraction model can be trained to have the ability to have a higher intra-class aggregation between the positive sound features and the target sound features, and a higher degree of differentiation between the negative sound features and the target sound features.

[0068] As a possible implementation, Triplet Loss (triplet loss function) is used to adjust the network parameters of the feature extraction model. For example, f is used to represent the embedding function of the feature extraction model to map the input sample sound to the sound feature, A is used to represent the sample target sound, P is used to represent the sample positive sound, and N is used to represent the sample negative sound. The sound triplet consisting of A, P, and N is input into the feature extraction model. The feature extraction model extracts the sample target sound A, the sample positive sound P, and the sample negative sound N, respectively, to obtain the target sound feature f(A), the positive sound feature f(P), and the negative sound feature f(N). The first distance between the positive sound feature f(A) and the target sound feature f(P) is ‖f(A)-f(P)‖2, and the second distance between the negative sound feature f(N) and the target sound feature f(A) is ‖f(A)-f(N)‖2. Triplet Loss is used to adjust the network parameters of the feature extraction model based on the difference between the first distance and the second distance. The calculation formula of Triplet Loss is as follows:

[0069] L(A,P,N)=max(‖f(A)-f(P)‖2-‖f(A)-f(N)‖2+α,0)

[0070] Here, α is a constant greater than 0.

[0071] When ‖f(A)-f(P)‖2-‖f(A)-f(N)‖2+α is less than 0, L(A,P,N)=0.

[0072] When ‖f(A)-f(P)‖2-‖f(A)-f(N)‖2+α is greater than 0, L(A,P,N)=‖f(A)-f(P)‖2-‖f(A)-f(N)‖2+α.

[0073] Since the formula L(A,P,N)=max(‖f(A)-f(P)‖2-‖f(A)-f(N)2,0) is directly used, it is only necessary to satisfy that the distances between A and P, and A and N are the same. It is difficult for the model to correctly distinguish between positive samples and negative samples. Therefore, by setting the constant α, the feature extraction model can be forced to learn hard to make the distance between A and N larger and the distance between A and P smaller. This allows the feature extraction model to be trained to have a higher intra-class aggregation between the positive sound features and the target sound features, and a higher degree of differentiation between the negative sound features and the target sound features.

[0074] In a specific implementation scenario, the sound features of the sound to be tested are obtained by extracting features from the time-frequency graph of the sound to be tested. Among them, the time-frequency graph is in the form of image features, and the feature extraction model may include but is not limited to: ResNet50, ViT and other models for image feature extraction. In addition, it should be noted that in the process of training the feature extraction model, the training is also based on the time-frequency graph of the sample device sound, that is, whether before the sound to be tested is input into the feature extraction model for feature extraction, or before the sample device sound is input into the feature extraction model for training, it is necessary to perform a short-time Fourier transform on the sound to be tested and the sample device sound to obtain the feature form of the time-frequency graph. The specific process can be referred to the relevant technology of speech feature extraction, which will not be repeated here. Therefore, by extracting features from the time-frequency graph of the sound to be tested, the feature information of the sound features can be extracted from the image feature dimension of the sound to be tested, thereby combining the sound features of the sound to be tested at the image feature level and the abnormal score of the sound to be tested, and enriching the data source of the fusion score.

[0075] As a possible example, see Figure 2 , Figure 2 The present invention is a process diagram of an embodiment of a training method for a feature extraction model, in which a plurality of sample device sounds representing a normal operation of the device and a plurality of sample device sounds representing an abnormal operation of the device are used as a sound training set, and a short-time Fourier transform is performed on each sample device sound in the sound training set to obtain a time-frequency graph. The feature extraction model extracts features from the time-frequency graph to obtain sound features, and uses the sound features as the model output. The network parameters of the feature extraction model are then adjusted by using a loss function such as gradient descent until the feature extraction model converges.

[0076] In another specific implementation scenario, unlike the aforementioned sound features of the sound to be tested that are obtained by extracting features from the time-frequency diagram of the sound to be tested, the sound features of the sound to be tested can also be obtained by extracting features from speech features other than the time-frequency diagram of the sound to be tested, such as Fbank features, MFCC features, PLP features, etc. These speech features can be found in the relevant descriptions in the aforementioned embodiments and will not be repeated here.

[0077] In one implementation scenario, in order to obtain a fusion score, the sound to be tested can be selected as the first sound according to the abnormality score of the sound to be tested, and the sounds to be tested other than the first sound can be selected as candidate sounds, and the candidate sounds can be selected as the second sound by the characteristic distances between the sound features of the first sound and the sound features of each candidate sound, and then the abnormality score of the first sound and the abnormality score of the second sound are fused to obtain a fusion score. Therefore, by selecting the first sound according to the abnormality scores of each sound to be tested, selecting the second sound according to the characteristic distance between the sound features of each sound to be tested, and fusing the abnormality scores of the specifically selected first sound and the abnormality scores of the second sound, the abnormality scores of the sounds to be tested can be selected for fusion in a targeted manner, which helps to improve the efficiency of obtaining the fusion score.

[0078] In a specific implementation scenario, in order to obtain the first sound, by comparing the size relationship between the abnormal scores of several sounds to be tested, the sound to be tested with the maximum abnormal score is selected as the first sound. Therefore, by selecting the sound to be tested with the maximum abnormal score as the first sound, the sound to be tested with the highest abnormal risk can be used as the detection basis for device abnormality detection, which helps to improve the sensitivity of device abnormality detection. In addition, the average value of the abnormal scores of several sounds to be tested can be calculated, and the sound to be tested with the abnormal score closest to the average value can be selected as the first sound. The method of selecting the first sound is not specifically limited here.

[0079] In a specific implementation scenario, in order to obtain the second sound, the characteristic distance between the sound feature of the first sound and the sound feature of each candidate sound can be compared, and after each candidate sound is sorted from small to large according to the characteristic distance, the candidate sound before the preset sequence is selected as the second sound. Among them, the number of selected second sounds can be 1, 2, 3, 4, etc., and the number of second sounds is not specifically limited here. Therefore, by sorting each candidate sound from small to large according to the characteristic distance, at least one second sound closest to the first sound can be selected from the candidate sounds, that is, by utilizing the correlation of the sound to be tested in two dimensions (abnormal score, characteristic distance), the organic combination between the abnormal score and the characteristic distance is achieved, which can improve the efficiency of device abnormality detection. In addition, by setting the distance threshold of the characteristic distance between the sound feature of the first sound and the sound feature of each candidate sound, the candidate sound within the distance threshold range is selected as the second sound, and the selection method of the second sound is not specifically limited here.

[0080] In a specific implementation scenario, the fusion score is obtained by averaging the abnormal score of the first sound and the abnormal score of the second sound, which can achieve neighborhood smoothing of the abnormal score of the first sound, reduce the interference of sound signal fluctuations, and improve the robustness of device abnormality detection. As a possible example, the target measurement point collects several sounds to be tested, including N, and F i Indicates the abnormal score of the sound to be tested, V i represents the sound characteristics of the sound to be tested, i = 1, 2, ..., N, and the abnormal score of the first sound is F j , the subscript set of k second sounds is K, and the abnormal scores of k second sounds are represented by F i (i∈K), based on the anomaly score of the first sound and the anomaly score of the second sound, the fusion score F is obtained. smooth for:

[0081]

[0082] In another implementation scenario, in order to obtain a fusion score, unlike the aforementioned method of selecting a candidate sound as the second sound based on the characteristic distance between the sound feature of the first sound and the sound features of each candidate sound, after selecting the sound to be tested as the first sound according to the abnormality score of the sound to be tested, the weighting factor of the corresponding sound to be tested can be obtained based on the characteristic distance between the sound feature of the first sound and the sound features of each sound to be tested, and then the weighting factor and abnormality score under the same sound to be tested can be weighted to obtain a fusion score. Among them, the weighting factor is negatively correlated with the characteristic distance. The method of selecting the sound to be tested as the first sound can refer to the relevant description in the aforementioned embodiment, which will not be repeated here. Exemplarily, the target measurement point includes three sounds to be measured, the abnormal score of the first sound is F1, and the weighting factor is 1. The abnormal scores of the sounds to be measured other than the first sound are F2 and F3, and the weighting factors are 0.7 and 0.3, respectively. In this case, weighting is performed based on the weighting factors and abnormal scores of each of the same sounds to be measured to obtain a fusion score: F0 = (F1 + 0.7F2 + 0.3F3) / 2. Therefore, according to the characteristic distance between the sound feature of the first sound and the sound features of each sound to be measured, the weighting factors corresponding to all sounds to be measured that are negatively correlated with the characteristic distance are obtained, and the abnormal scores of all sounds to be measured can be fused to consider the impact of all sounds to be measured at the target measurement point on the abnormal detection of the device as comprehensively as possible, thereby helping to improve the data validity of the fusion score.

[0083] Step S14: Based on the fusion score, determine whether the target measuring point is operating abnormally.

[0084] In the embodiment of the present disclosure, by setting an abnormal threshold of the device under test at the target measuring point, and judging whether the fusion score exceeds the abnormal threshold, it is determined whether the target measuring point is operating abnormally. The abnormal threshold can be obtained by statistically analyzing the abnormal score of the operating sound of the target measuring point when the device under test is operating normally.

[0085] In one implementation scenario, after determining that the target measuring point is operating abnormally, an alarm prompt of the target measuring point operating abnormally may be generated, wherein the form of the alarm prompt may include but is not limited to: a pop-up window prompt, a sound prompt, and the like.

[0086] The above scheme obtains several sounds to be tested collected at the target measuring point of the device to be tested, and obtains the abnormality score of each sound to be tested, and then fuses the abnormality scores of at least part of the sounds to be tested based on the characteristic distance between the sound features of the sounds to be tested, obtains the fusion score, and then determines whether the target measuring point is operating abnormally based on the fusion score; wherein the abnormality score of the sound to be tested represents: the abnormality degree of the sound to be tested compared with the sound of the target measuring point when the device to be tested is operating normally. Therefore, in the case of obtaining the abnormality score of each sound to be tested, the abnormality score of at least part of the sounds to be tested is further fused by the characteristic distance between the sound features of the sounds to be tested. On the one hand, using multiple sounds to be tested at the same measuring point as the data source of the device abnormality detection judgment standard can reduce the interference of invalid data collected under the influence of sound signal fluctuations, and thus can improve the accuracy of device abnormality detection from the data validity level; on the other hand, by combining the abnormality score of the sound to be tested and the characteristic distance between the sound features, compared with directly using the abnormality score of a single dimension as the judgment criterion for device abnormality detection, the accuracy of device abnormality detection can be improved from a fine-grained level. Therefore, the accuracy of device anomaly detection can be improved.

[0087] See also Figure 3 , Figure 3 This is a framework diagram of an embodiment of an equipment anomaly detection device of the present application, wherein the equipment anomaly detection device 30 includes a first acquisition module 31, a second acquisition module 32, a fusion module 33 and a determination module 34; the first acquisition module 31 is used to acquire a number of sounds to be tested collected at a target measuring point of the equipment to be tested; the second acquisition module 32 is used to acquire the anomaly score of each sound to be tested; wherein the anomaly score of the sound to be tested represents: the degree of anomaly of the sound to be tested compared to the sound of the target measuring point when the equipment to be tested is operating normally; the fusion module 33 is used to fuse the anomaly scores of at least part of the sounds to be tested based on the feature distance between the sound features of the sounds to be tested to obtain a fusion score; the determination module 34 is used to determine whether the target measuring point is operating abnormally based on the fusion score.

[0088] The above scheme obtains several sounds to be tested collected at the target measuring point of the device to be tested, and obtains the abnormality score of each sound to be tested, and then fuses the abnormality scores of at least part of the sounds to be tested based on the characteristic distance between the sound features of the sounds to be tested, obtains the fusion score, and then determines whether the target measuring point is operating abnormally based on the fusion score; wherein the abnormality score of the sound to be tested represents: the abnormality degree of the sound to be tested compared with the sound of the target measuring point when the device to be tested is operating normally. Therefore, in the case of obtaining the abnormality score of each sound to be tested, the abnormality score of at least part of the sounds to be tested is further fused by the characteristic distance between the sound features of the sounds to be tested. On the one hand, using multiple sounds to be tested at the same measuring point as the data source of the device abnormality detection judgment standard can reduce the interference of invalid data collected under the influence of sound signal fluctuations, and thus can improve the accuracy of device abnormality detection from the data validity level; on the other hand, by combining the abnormality score of the sound to be tested and the characteristic distance between the sound features, compared with directly using the abnormality score of a single dimension as the judgment criterion for device abnormality detection, the accuracy of device abnormality detection can be improved from a fine-grained level. Therefore, the accuracy of device anomaly detection can be improved.

[0089] In some disclosed embodiments, the fusion module 33 includes a first selection submodule, a second selection submodule and a fusion submodule. The first selection submodule is used to select the sound to be tested as the first sound based on the abnormality score of the sound to be tested; the second selection submodule is used to select the candidate sound as the second sound based on the feature distance between the sound features of the first sound and the sound features of each candidate sound; wherein the candidate sound is the sound to be tested other than the first sound; and the fusion submodule is used to fuse the abnormality score of the first sound and the abnormality score of the second sound to obtain a fusion score.

[0090] Therefore, by selecting the first sound according to the abnormal scores of each sound to be tested, selecting the second sound according to the feature distance between the sound features of each sound to be tested, and fusing the abnormal scores of the specifically selected first sound and the abnormal scores of the second sound, the abnormal scores of the sounds to be tested can be selected for fusion in a targeted manner, which helps to improve the efficiency of obtaining the fusion score.

[0091] In some disclosed embodiments, the first selection submodule is specifically configured to select the sound to be tested having the maximum anomaly score as the first sound.

[0092] Therefore, by selecting the sound to be tested with the maximum anomaly score as the first sound, the sound to be tested with the highest anomaly risk can be used as the detection basis for device anomaly detection, which helps to improve the sensitivity of device anomaly detection.

[0093] In some disclosed embodiments, the second selection submodule is specifically used to select the candidate sound located before a preset order as the second sound after sorting the candidate sounds from small to large according to the characteristic distance.

[0094] Therefore, by sorting the candidate sounds from small to large according to the feature distance, at least one second sound that is closest to the first sound can be selected from the candidate sounds. That is, by utilizing the correlation between the sound to be tested in two dimensions (abnormal score, feature distance), an organic combination between the abnormal score and the feature distance is achieved, which can improve the efficiency of device abnormality detection.

[0095] In some disclosed embodiments, the fusion submodule is specifically configured to obtain a fusion score based on an average of the anomaly score of the first sound and the anomaly score of the second sound.

[0096] Therefore, by averaging the anomaly score of the first sound and the anomaly score of the second sound, neighborhood smoothing of the anomaly score of the first sound can be achieved, interference from sound signal fluctuations can be reduced, and the robustness of device anomaly detection can be improved.

[0097] In some disclosed embodiments, the fusion module 33 is also specifically used to select the sound to be tested as the first sound based on the abnormality score of the sound to be tested; obtain the weighting factor of the corresponding sound to be tested based on the characteristic distance between the sound characteristics of the first sound and the sound characteristics of each sound to be tested; wherein the weighting factor is negatively correlated with the characteristic distance; and obtain the fusion score based on the weighting factor and abnormality score of the same sound to be tested.

[0098] Therefore, according to the characteristic distances between the sound features of the first sound and the sound features of each sound to be tested, a weighting factor negatively correlated with the characteristic distance corresponding to all the sounds to be tested is obtained, and the abnormal scores of all the sounds to be tested can be fused to consider the impact of all the sounds to be tested at the target measuring point on the device abnormality detection as comprehensively as possible, thereby helping to improve the data validity of the fused score.

[0099] In some disclosed embodiments, sound features of a sound to be tested are obtained by extracting features of the sound to be tested by a feature extraction model, the feature extraction model is trained by sample device sounds, the sample device sounds are annotated with device states representing whether the operation is abnormal, and the feature extraction model is specifically trained based on sound triples, the sound triples include: any sample device sound as a sample target sound, a sample device sound with the same device state as the sample target sound as a sample positive sound, and a sample device sound with a different device state from the sample target sound as a sample negative sound.

[0100] Therefore, by using sample device sounds in different device states that characterize whether they are operating abnormally as the training input of the feature extraction model, the feature extraction model can be trained to have the ability to have a high intra-class aggregation between sample device sounds in the same device state, and a high degree of differentiation between sample device sounds in different device states. Then, by obtaining the sound features of several sounds to be tested through the trained feature extraction model, the classification accuracy of the degree of aggregation and differentiation between the sound features can be improved.

[0101] In some disclosed embodiments, the equipment abnormality detection device 30 also includes a training module, which is used to train a feature extraction model, specifically to extract sound features of sample target sounds as target sound features based on the feature extraction model, and to extract sound features of sample positive sound as positive sound features based on the feature extraction model, and to extract sound features of sample negative sound as negative sound features based on the feature extraction model; to obtain a first distance between the positive sound feature and the target sound feature, and to obtain a second distance between the negative sound feature and the target sound feature; and to adjust the network parameters of the feature extraction model based on the difference between the first distance and the second distance.

[0102] Therefore, through the difference between the first distance and the second distance, the feature extraction model can be trained to have the ability to have a higher intra-class clustering between the positive example sound features and the target sound features, and a higher degree of differentiation between the negative example sound features and the target sound features.

[0103] In some disclosed embodiments, the sound features of the sound to be measured are obtained by extracting features from a time-frequency diagram of the sound to be measured.

[0104] Therefore, by performing feature extraction on the time-frequency graph of the sound to be tested, feature information of the sound features can be extracted from the image feature dimension of the sound to be tested, thereby combining the sound features of the sound to be tested at the image feature level and the abnormality score of the sound to be tested to enrich the data source of the fusion score.

[0105] In some disclosed embodiments, the second acquisition module 32 is specifically used to process the sound to be tested based on the anomaly detection model to obtain the predicted sound corresponding to the sound to be tested under the parameter conditions of the anomaly detection model; wherein the anomaly detection model is obtained by adjusting the parameters of at least the operating sound of the target measuring point when the device to be tested is operating normally; based on the difference between the predicted sound and the sound to be tested, the anomaly score of the sound to be tested is obtained.

[0106] Therefore, the predicted sound corresponding to the sound to be tested under the parameter conditions of the anomaly detection model is obtained through the anomaly detection model. Through the difference between the sound to be tested and the predicted sound, the degree of abnormality of the sound to be tested compared with the sound of the target measuring point when the device to be tested is operating normally can be obtained. Then, the anomaly detection model is used to assist in obtaining the anomaly score of the sound to be tested. Compared with the possible objectivity caused by manual scoring, the objectivity of equipment abnormality detection can be improved.

[0107] In some disclosed embodiments, a target measuring point of the device to be tested is provided with a target sensing point of the optical fiber 51, and the first acquisition module is specifically used to receive the operating sound of the target measuring point obtained by demodulating the optical signal collected from the target sensing point by the demodulation device 52 as the sound to be tested.

[0108] Therefore, through the communication between the optical fiber 51 and the demodulation device 52, the demodulation device 52 can collect the optical signal reflected by the optical fiber 51 from the target sensing point of the optical fiber 51, and then can obtain the sound to be tested obtained by demodulating the optical signal from the demodulation device 52, thereby realizing the automated process of sound collection for the device to be tested.

[0109] See also Figure 4 , Figure 4 The schematic diagram of the framework of an embodiment of an electronic device 40 of the present application. The electronic device 40 includes a memory 41 and a processor 42 coupled to each other, and the processor 42 is used to execute program instructions stored in the memory 41 to implement the steps in any of the above-mentioned device abnormality detection method embodiments. Specifically, the electronic device 40 may include but is not limited to: a microcomputer, a server, a laptop computer, a tablet computer, a smart phone, etc., which are not limited here.

[0110] Specifically, the processor 42 is used to control itself and the memory 41 to implement the steps of any of the above-mentioned device abnormality detection method embodiments. The processor 42 can also be called a CPU (Central Processing Unit). The processor 42 may be an integrated circuit chip with signal processing capabilities. The processor 42 can also be a general-purpose processor, a digital signal processor (Digital Signal Processor, DSP), an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), a field-programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. In addition, the processor 42 can be implemented by an integrated circuit chip.

[0111] In the above scheme, the electronic device 40 obtains a number of sounds to be tested collected at the target measuring point of the device to be tested, and obtains the abnormality score of each sound to be tested, and then fuses the abnormality scores of at least part of the sounds to be tested based on the characteristic distance between the sound features of the sounds to be tested to obtain a fusion score, and then determines whether the target measuring point is operating abnormally based on the fusion score; wherein the abnormality score of the sound to be tested represents: the degree of abnormality of the sound to be tested compared to the sound of the target measuring point when the device to be tested is operating normally. Therefore, when the abnormal scores of each sound to be tested are obtained, the abnormal scores of at least part of the sounds to be tested are further fused through the characteristic distance between the sound features of the sounds to be tested. On the one hand, using multiple sounds to be tested at the same measuring point as the data source for the equipment abnormality detection judgment standard can reduce the interference of invalid data collected under the influence of sound signal fluctuations, and thus can improve the accuracy of equipment abnormality detection from the data validity level; on the other hand, by combining the two dimensions of the abnormal score of the sound to be tested and the characteristic distance between the sound features, compared with directly using the abnormal score of a single dimension as the criterion for equipment abnormality detection, the accuracy of equipment abnormality detection can be improved from a fine-grained level. Therefore, the accuracy of equipment abnormality detection can be improved.

[0112] See also Figure 5 , Figure 5 This is a framework diagram of an embodiment of a device abnormality detection system 50 of the present application. The device abnormality detection system 50 at least includes an optical fiber 51, a demodulation device 52 and a detection device 53. The demodulation device 52 is communicatively connected between the optical fiber 51 and the detection device 53. The optical fiber 51 is distributed with a plurality of sensing points, which are respectively laid at each of the test points of the device to be tested. The demodulation device 52 is used to demodulate the operating sound of the device to be tested at the test point based on the optical signal of the optical fiber 51 at the sensing point, and the detection device 53 includes the electronic device 40 in the aforementioned embodiment.

[0113] In one implementation scenario, the equipment abnormality detection system 50 uses distributed fiber acoustic sensing (DAS) technology to collect the sound to be tested. Specifically, the demodulation device 52 sends a laser pulse to the optical fiber 51 laid on the equipment to be tested. The laser will undergo Rayleigh scattering (backscattered light signal, opposite to the laser emission direction) in the optical fiber 51, and the backscattered light signal will be transmitted back to the demodulation device 52. When the optical fiber 51 is affected by the nearby sound field and produces strain, the phase of the scattered light will also change. The position of each sensing point of the optical fiber 51 from the demodulation device 52 is known. The demodulation device 52 can sense the sound information of the equipment to be tested near the sensing point by detecting the change in the phase of the scattered light transmitted back from the sensing point of the optical fiber 51 at a certain position. After the demodulation device 52 collects the data transmitted back from the optical fiber 51 once, the sound at any position on the optical fiber 51 can be demodulated from this data, so that the demodulation device 52 can demodulate the operating sound of each test point at different positions on the equipment to be tested.

[0114] In one implementation scenario, see Figure 6 , Figure 6 for Figure 5 A schematic diagram of the structure of an embodiment of a device abnormality detection system 50, wherein the device to be tested includes a plurality of points to be tested, an optical fiber 51 is distributed with a plurality of sensing points, which are respectively laid at each point to be tested of the device to be tested, and a measuring point n is used as a target measuring point. A demodulation device 52 is communicatively connected between the optical fiber 51 and a detection device 53, and the detection device 53 includes a first conversion unit, an abnormality detection model, a calculation unit, a score determination unit, a second conversion unit, a feature extraction model, a fusion unit and an alarm unit. The first conversion unit is used to convert the sound to be tested into an MFCC feature as an input sound of the abnormality detection model, the abnormality detection model is used to predict the predicted sound corresponding to the input sound under the model's own parameter conditions, the first comparison unit is used to compare the difference between the predicted sound and the sound to be tested, the score determination unit is used to determine the abnormal score of the sound to be tested, the second conversion unit is used to convert the sound data into a time-frequency diagram, the feature extraction model is used to extract the sound features of the sound to be tested, the second comparison unit is used to compare the feature distances between the sound features, the fusion unit is used to obtain the fusion score, and the alarm unit is used to issue an alarm prompt.

[0115] In the above scheme, the equipment abnormality detection system 50 obtains a number of test sounds collected at the target measuring point of the equipment to be tested, and obtains the abnormality score of each test sound, and then fuses the abnormality scores of at least part of the test sounds based on the characteristic distance between the sound features of the test sounds to obtain a fusion score, and then determines whether the target measuring point is operating abnormally based on the fusion score; wherein the abnormality score of the test sound represents: the degree of abnormality of the test sound compared to the sound of the target measuring point when the equipment to be tested is operating normally. Therefore, when the abnormal scores of each sound to be tested are obtained, the abnormal scores of at least part of the sounds to be tested are further fused through the characteristic distance between the sound features of the sounds to be tested. On the one hand, using multiple sounds to be tested at the same measuring point as the data source for the equipment abnormality detection judgment standard can reduce the interference of invalid data collected under the influence of sound signal fluctuations, and thus can improve the accuracy of equipment abnormality detection from the data validity level; on the other hand, by combining the two dimensions of the abnormal score of the sound to be tested and the characteristic distance between the sound features, compared with directly using the abnormal score of a single dimension as the criterion for equipment abnormality detection, the accuracy of equipment abnormality detection can be improved from a fine-grained level. Therefore, the accuracy of equipment abnormality detection can be improved.

[0116] See also Figure 7 , Figure 7 The schematic diagram of the framework of an embodiment of a computer-readable storage medium 70 of the present application. The computer-readable storage medium 70 stores program instructions 71 that can be executed by a processor, and the program instructions 71 are used to implement the steps of the embodiment of the above-mentioned device abnormality detection method.

[0117] In the above scheme, the computer-readable storage medium 70 obtains a number of sounds to be tested collected at a target measuring point of the device to be tested, and obtains an abnormality score of each sound to be tested, and then fuses the abnormality scores of at least part of the sounds to be tested based on the characteristic distance between the sound features of the sounds to be tested to obtain a fusion score, and then determines whether the target measuring point is operating abnormally based on the fusion score; wherein the abnormality score of the sound to be tested represents: the degree of abnormality of the sound to be tested compared to the sound of the target measuring point when the device to be tested is operating normally. Therefore, when the abnormal scores of each sound to be tested are obtained, the abnormal scores of at least part of the sounds to be tested are further fused through the characteristic distance between the sound features of the sounds to be tested. On the one hand, using multiple sounds to be tested at the same measuring point as the data source for the equipment abnormality detection judgment standard can reduce the interference of invalid data collected under the influence of sound signal fluctuations, and thus can improve the accuracy of equipment abnormality detection from the data validity level; on the other hand, by combining the two dimensions of the abnormal score of the sound to be tested and the characteristic distance between the sound features, compared with directly using the abnormal score of a single dimension as the criterion for equipment abnormality detection, the accuracy of equipment abnormality detection can be improved from a fine-grained level. Therefore, the accuracy of equipment abnormality detection can be improved.

[0118] In some embodiments, the functions or modules included in the device provided by the embodiments of the present disclosure can be used to execute the method described in the above method embodiments. The specific implementation can refer to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0119] The above description of various embodiments tends to emphasize the differences between the various embodiments. The same or similar aspects can be referenced to each other, and for the sake of brevity, they will not be repeated herein.

[0120] In the several embodiments provided in the present application, it should be understood that the disclosed methods and devices can be implemented in other ways. For example, the device implementation described above is only schematic. For example, the division of modules or units is only a logical function division. There may be other division methods in actual implementation, such as units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0121] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0122] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) or a processor (processor) to perform all or part of the steps of each implementation method of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program code.

[0123] If the technical solution of this application involves personal information, the product using the technical solution of this application has clearly informed the personal information processing rules and obtained the individual's voluntary consent before processing the personal information. If the technical solution of this application involves sensitive personal information, the product using the technical solution of this application has obtained the individual's separate consent before processing the sensitive personal information, and at the same time meets the "explicit consent" requirement. For example, on personal information collection devices such as cameras, clear and prominent signs are set to inform that the personal information collection scope has been entered and personal information will be collected. If the individual voluntarily enters the collection scope, it is deemed that he or she agrees to the collection of his or her personal information; or on the device that processes personal information, the personal information processing rules are notified by obvious signs / information, and the individual's authorization is obtained through pop-up information or by asking the individual to upload his or her personal information; among them, the personal information processing rules may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the type of personal information processed.

Claims

1. A device anomaly detection method, characterized in that: include: Acquire a number of sounds to be tested collected at a target test point of the device to be tested; Obtaining an abnormality score of each of the sounds to be tested; wherein the abnormality score of the sound to be tested represents: the degree of abnormality of the sound to be tested compared to the sound of the target measuring point when the device to be tested is operating normally; Based on the characteristic distances between the sound features of the sound to be tested, fusing at least part of the abnormality scores of the sound to be tested to obtain a fusion score; Based on the fusion score, it is determined whether the target measuring point operates abnormally.

2. The method according to claim 1, characterized in that The step of fusing at least part of the abnormality scores of the sounds to be tested based on the characteristic distances between the sound features of the sounds to be tested to obtain a fusion score includes: Based on the abnormality score of the sound to be tested, selecting the sound to be tested as the first sound; Based on the characteristic distances between the sound characteristics of the first sound and the sound characteristics of each candidate sound, selecting the candidate sound as the second sound; wherein the candidate sound is a sound to be tested other than the first sound; The anomaly score of the first sound and the anomaly score of the second sound are fused to obtain the fusion score.

3. The method according to claim 2, characterized in that: The step of selecting the sound to be tested as the first sound based on the abnormality score of the sound to be tested includes: The sound to be tested with the largest anomaly score is selected as the first sound.

4. The method according to claim 2, characterized in that: The selecting the candidate sound as the second sound based on the feature distances between the sound feature of the first sound and the sound features of each candidate sound respectively includes: After the candidate sounds are sorted from small to large according to the characteristic distance, the candidate sound located before the preset order is selected as the second sound.

5. The method according to claim 2, characterized in that: The fusing the anomaly score based on the first sound and the anomaly score based on the second sound to obtain the fused score includes: The fusion score is obtained by averaging the anomaly score of the first sound and the anomaly score of the second sound.

6. The method according to claim 1, characterized in that The step of fusing at least part of the abnormality scores of the sounds to be tested based on the characteristic distances between the sound features of the sounds to be tested to obtain a fusion score includes: Based on the abnormality score of the sound to be tested, selecting the sound to be tested as the first sound; Based on the characteristic distances between the sound feature of the first sound and the sound features of each of the sounds to be measured, a weighting factor corresponding to the sound to be measured is obtained; wherein the weighting factor is negatively correlated with the characteristic distance; The fusion score is obtained by weighting based on the weighting factor and the abnormal score of the same sound to be tested.

7. The method according to claim 1, characterized in that The sound features of the sound to be tested are obtained by extracting features from the sound to be tested by a feature extraction model, the feature extraction model is trained by sample device sounds, the sample device sounds are annotated with device states representing whether the operation is abnormal, and the feature extraction model is specifically trained based on sound triples, the sound triples include: any one of the sample device sounds as a sample target sound, a sample device sound having the same device state as the sample target sound as a sample positive sound, and a sample device sound having a different device state from the sample target sound as a sample negative sound.

8. The method according to claim 7, characterized in that The training steps of the feature extraction model include: Extracting the sound features of the sample target sound as the target sound features based on the feature extraction model, extracting the sound features of the sample positive sound as the positive sound features based on the feature extraction model, and extracting the sound features of the sample negative sound as the negative sound features based on the feature extraction model; Acquire a first distance between the positive example sound feature and the target sound feature, and acquire a second distance between the negative example sound feature and the target sound feature; Based on the difference between the first distance and the second distance, a network parameter of the feature extraction model is adjusted.

9. The method according to claim 1 or 7, characterized in that: The sound feature of the sound to be measured is specifically obtained by extracting features from the time-frequency diagram of the sound to be measured.

10. The method according to claim 1, characterized in that The obtaining of the abnormality score of each of the sounds to be tested comprises: Process the sound to be tested based on the anomaly detection model to obtain a predicted sound corresponding to the sound to be tested under the parameter conditions of the anomaly detection model; wherein the anomaly detection model is obtained by adjusting the parameters of at least the operating sound of the target measuring point when the device to be tested is operating normally; Based on the difference between the predicted sound and the sound to be tested, an abnormality score of the sound to be tested is obtained.

11. The method according to claim 1, characterized in that: The target measuring point of the device under test is provided with a target sensing point of an optical fiber, and the acquiring of a plurality of sounds to be tested collected at the target measuring point of the device under test comprises: The operating sound of the target measuring point obtained by demodulating the optical signal collected from the target sensing point by a demodulation device is received as the sound to be measured.

12. A device for detecting abnormality of equipment, characterized in that: include: A first acquisition module is used to acquire a number of sounds to be tested collected at a target measurement point of the device to be tested; The second acquisition module is used to obtain the abnormality score of each of the sounds to be tested; wherein the abnormality score of the sound to be tested represents: the abnormality degree of the sound to be tested compared with the sound of the target measurement point when the device to be tested is operating normally; A fusion module, configured to fuse at least part of the abnormality scores of the sounds to be tested based on the feature distances between the sound features of the sounds to be tested to obtain a fusion score; A determination module is used to determine whether the target measuring point operates abnormally based on the fusion score.

13. An electronic device, characterized in that: It comprises a memory and a processor coupled to each other, wherein the processor is used to execute program instructions stored in the memory to implement the device abnormality detection method according to any one of claims 1 to 11.

14. A device anomaly detection system, characterized in that: The method comprises at least an optical fiber, a demodulation device and a detection device, wherein the demodulation device is communicatively connected between the optical fiber and the detection device, the optical fiber is provided with a plurality of sensing points which are respectively laid at each of the test points of the device to be tested, the demodulation device is used for demodulating the operating sound of the device to be tested at the test point based on the optical signal of the optical fiber at the sensing point, and the detection device comprises the electronic device according to claim 13.

15. A computer-readable storage medium having program instructions stored thereon, characterized in that: When the program instructions are executed by a processor, the device abnormality detection method according to any one of claims 1 to 11 is implemented.

Images