Power equipment fault diagnosis method based on causal domain adaptation and acoustic vibration fusion

Abstract

The application belongs to the technical field of machine learning, and particularly relates to a power equipment fault diagnosis method based on causal domain adaptation and sound-vibration fusion. The power equipment fault diagnosis method provided by the application inputs sound real-time signal data of a target power equipment fault into a sound-vibration conversion model to obtain sound conversion signals; the sound conversion signals and vibration real-time signal data are input into a causal domain adaptation fault diagnosis model to obtain a fault diagnosis result. The sound-vibration conversion model solves the problem that a traditional deep learning model is difficult to train due to the scarcity of single-mode fault samples. In view of the problem of insufficient model generality, a fault diagnosis model based on domain adaptation fusion is constructed by using sound-vibration fusion data. After the model is trained, the model can be used to diagnose faults of sound-vibration data of different sources in actual tasks, thereby solving the problems of insufficient accuracy and poor generality of traditional deep learning diagnosis models when processing data of different sources.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of machine learning, and particularly relates to a power equipment fault diagnosis method based on causal domain adaptation and sound-vibration fusion. BACKGROUND

[0002] Deep learning is a research direction in the field of machine learning, aiming to enable machines to have analysis and learning capabilities like humans. Through learning the internal laws and representation levels of sample data, it can identify data such as text, images and sound. Deep learning is a complex machine learning algorithm, and the effect achieved in speech and image recognition far exceeds that of previous related technologies. It enables machines to imitate human activities such as vision, hearing and thinking, solves many complex pattern recognition problems, and makes great progress in artificial intelligence related technologies.

[0003] At present, with the deepening of research on deep learning algorithms, the powerful fitting ability of deep algorithms has been paid more and more attention. In the fault diagnosis task, due to the influence of factors such as limited fault samples and insufficient cross-scene flexibility, the fault diagnosis model based on deep learning still has some deficiencies. In terms of data, due to the scarcity of fault samples, it is difficult to effectively support the training of the fault diagnosis model, and the existing fault collection samples need to obtain complete labeled data through artificial means, which brings a huge cost problem; in terms of model, due to the large difference in working conditions of power equipment in different operating environments, the existing fault diagnosis model for a certain power equipment is difficult to be directly reused in the diagnosis task of the corresponding equipment in other working environments. SUMMARY

[0004] The purpose of the present application is to provide a power equipment fault diagnosis method based on causal domain adaptation and sound-vibration fusion, to solve the problems of scarcity of fault samples, poor model data resource reusability and the like in the prior art.

[0005] In order to achieve the above-mentioned purpose, the technical scheme adopted by the present application is as follows:

[0006] In the first aspect of the present application, a power equipment fault diagnosis method based on causal domain adaptation and sound-vibration fusion is provided, comprising the following steps:

[0007] Obtaining sound real-time signal data and vibration real-time signal data converted into soundprint representation signal during fault of a target power equipment;

[0008] Inputting the sound real-time signal data during fault of the target power equipment into a pre-trained sound-vibration conversion model, the sound-vibration conversion model outputs sound conversion signal; wherein a cycle consistent generative adversarial network is used as the basic network of the sound-vibration conversion model;

[0009] The sound conversion signal and the real-time vibration signal data are input together into a pre-trained causal domain adaptive fault diagnosis model, which outputs the fault diagnosis result of the target power equipment.

[0010] As a further improvement to this solution, before inputting the real-time sound signal data of the target power equipment during a fault into the pre-trained sound-vibration conversion model, the solution further includes: preprocessing the real-time sound signal data and the real-time vibration signal data of the target power equipment during a fault.

[0011] As a further improvement to this solution, the method for preprocessing the real-time sound signal data and real-time vibration signal data during the fault of the target power equipment includes:

[0012] Discretize and sample the real-time sound signal data and real-time vibration signal data of the target power equipment to obtain a data subset;

[0013] The data subset is then normalized.

[0014] The normalized subset of data is converted into Mel spectrogram form to complete the preprocessing.

[0015] As a further improvement to this scheme, the normalized subset of data is converted into Mel spectrogram form, including:

[0016] The normalized subset of data is represented as a spectrum using the Fast Fourier Transform;

[0017] Convert the spectrum into a Mel spectrum;

[0018] The Mel spectrum is passed through a Mel filter bank to obtain the Mel spectrum form.

[0019] As a further improvement to this solution, in the step of inputting the real-time sound signal data of the target power equipment during a fault into the pre-trained sound-vibration conversion model, the training method of the sound-vibration conversion model is as follows:

[0020] A periodically consistent adversarial network is used as the base network for the acoustic-vibration conversion model;

[0021] Determine the overall loss function of the acoustic-vibration conversion model;

[0022] Based on the basic network and overall loss function of the aforementioned acoustic-vibration conversion model, the model parameters are trained using the gradient descent method. After training, the trained acoustic-vibration conversion model is obtained.

[0023] As a further improvement to this scheme, the overall loss function of the acoustic-vibration conversion model is determined, and the overall loss function L... m (θ) is represented as follows:

[0024] L m (θ)=L adv1 (θ)+L cyc (θ)+L id (θ)+L adv2 (θ)

[0025] In the above formula, L adv1 (θ) represents a first adversarial loss term, L cyc (θ) represents a cycle-consistency loss term, L id (θ) represents an identity mapping loss term, L adv2 (θ) represents a second adversarial loss term.

[0026] The first adversarial loss term is measured by the Wasserstein distance to measure the difference between data, and is expressed as:

[0027]

[0028] The cycle-consistency loss term is expressed as:

[0029]

[0030] The identity mapping loss term is expressed as:

[0031]

[0032] The second adversarial loss term is expressed as:

[0033]

[0034] In the above formula, G X-Y represents a generator that converts data X to Y, D Y represents a domain discriminator, and E represents an expected value of corresponding data.

[0035] As a further improvement of the present scheme, the sound conversion signal is input into a pre-trained causal domain adaptive fault diagnosis model together with the vibration real-time signal data, and the training method of the causal domain adaptive fault diagnosis model is as follows:

[0036] Obtain sound signal data and vibration signal data when the power equipment fails, process the sound signal data using a sound-vibration conversion model to obtain sound conversion training data; the vibration signal data and the sound conversion training data are used for training of a causal domain adaptive fault diagnosis model;

[0037] Determine the overall loss function of the causal domain adaptive fault diagnosis model;

[0038] Train the causal domain adaptive fault diagnosis model based on the vibration signal data, the sound conversion training data and the overall loss function of the causal domain adaptive fault diagnosis model, and obtain the trained causal domain adaptive fault diagnosis model after the training is completed.

[0039] In the second aspect of the present application, a power equipment fault diagnosis device is provided, comprising:

[0040] A data acquisition module is configured to acquire real-time sound signal data and vibration real-time signal data converted into a soundprint representation signal when a target power equipment fails.

[0041] A data fusion module is configured to input the real-time sound signal data when the target power equipment fails into a pre-trained sound-vibration conversion model, and the sound-vibration conversion model outputs a sound conversion signal; wherein a periodically consistent adversarial network is used as a basic network of the sound-vibration conversion model.

[0042] A fault diagnosis module is configured to input the sound conversion signal and the vibration real-time signal data into a pre-trained causal domain adaptive fault diagnosis model, and the causal domain adaptive fault diagnosis model outputs a fault diagnosis result of the target power equipment.

[0043] In the third aspect of the present application, an electronic device is provided, comprising a processor and a memory, wherein the processor is configured to execute a computer program stored in the memory to realize the power equipment fault diagnosis method based on causal domain adaptation and sound-vibration fusion as described above.

[0044] In the fourth aspect of the present application, a computer readable storage medium is provided, wherein the computer readable storage medium stores at least one instruction, and the at least one instruction is executed by a processor to realize the power equipment fault diagnosis method based on causal domain adaptation and sound-vibration fusion as described above.

[0045] Compared with the prior art, the present application has the following advantages:

[0046] The present invention provides a power equipment fault diagnosis method based on causal domain adaptation and acoustic-vibration fusion. This method acquires real-time sound signal data and real-time vibration signal data converted into acoustic signatures when a target power equipment malfunctions. The real-time sound signal data is input into a pre-trained acoustic-vibration conversion model, which outputs a converted sound signal. A periodically consistent adversarial network is used as the base network of the acoustic-vibration conversion model. The converted sound signal and the real-time vibration signal data are then input into a pre-trained causal domain adaptation fault diagnosis model, which outputs the fault diagnosis result for the target power equipment. The acoustic-vibration conversion model addresses the problem of traditional deep learning models being difficult to train due to the scarcity of single-modal fault samples. To address the issue of insufficient model generality, a fault diagnosis model based on domain adaptation fusion is constructed using acoustic-vibration fusion data. After training, the model can be used to diagnose faults using acoustic-vibration data from different sources in real-world tasks, solving the problems of insufficient accuracy and poor generality of traditional deep learning diagnostic models when processing data from different sources. Compared to traditional deep learning-based fault diagnosis methods, this invention reduces the impact of scarce fault samples on model training through multi-source, multi-modal data fusion, and improves the model's general applicability and diagnostic accuracy. The apparatus, equipment, and media provided by this invention solve the same problem. Attached Figure Description

[0047] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0048] Figure 1 This is a flowchart illustrating a power equipment fault diagnosis method based on causal domain adaptation and acoustic-vibration fusion according to an embodiment of the present invention.

[0049] Figure 2 This is a schematic diagram illustrating the principle of a power equipment fault diagnosis method based on causal domain adaptation and acoustic-vibration fusion according to an embodiment of the present invention.

[0050] Figure 3 This is a schematic diagram of the sound-vibration conversion model in an embodiment of the present invention;

[0051] Figure 4 This is a structural block diagram of a power equipment fault diagnosis device according to an embodiment of the present invention;

[0052] Figure 5 This is a structural block diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0053] The application will be described in detail below with reference to the drawings and in conjunction with embodiments. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0054] The following detailed description is exemplary and is intended to provide further details in order to provide a further understanding of the present application. Unless otherwise defined, all technical terms used in the present application have the same meanings as generally understood by those skilled in the art to which the present application belongs. The terms used in the present application are only intended to describe specific embodiments and are not intended to limit the exemplary embodiments according to the present application.

[0055] Embodiment 1

[0056] The embodiment of the present application provides a power equipment fault diagnosis method based on causal domain adaptation and sound-vibration fusion. First, the actual collected sound and vibration signals are preprocessed, including signal denoising, normalization, signal clipping and alignment, etc. Second, the sound and vibration signals are converted into the data form of a mel spectrogram. Third, a sound-vibration conversion model based on domain adaptation migration is constructed to establish a sound conversion signal, solving the problem that a traditional deep learning model is difficult to train due to the scarcity of fault samples. Finally, a causal domain adaptation fault diagnosis model is constructed based on a fusion data set to support power equipment fault diagnosis and solve the problems of insufficient precision and poor generality of traditional deep learning diagnosis models.

[0057] In the embodiment, a power equipment fault diagnosis method based on causal domain adaptation and sound-vibration fusion includes the following specific steps:

[0058] S100, acquiring sound real-time signal data and vibration real-time signal data converted into soundprint representation signals when a target power equipment fails.

[0059] S200, inputting the sound real-time signal data when the target power equipment fails into a pre-trained sound-vibration conversion model, and the sound-vibration conversion model outputs a sound conversion signal; wherein a cycle-consistent generative adversarial network is used as a basic network of the sound-vibration conversion model.

[0060] S300, inputting the sound conversion signal and the vibration real-time signal data into a pre-trained causal domain adaptation fault diagnosis model, and the causal domain adaptation fault diagnosis model outputs a fault diagnosis result of the target power equipment.

[0061] Through the above scheme provided by the present application, a power equipment fault diagnosis model is constructed based on domain adaptation fusion, solving the problem that a traditional deep learning model is difficult to train due to the scarcity of single-modal fault samples; in view of the problem of insufficient generality of the model, a causal domain adaptation diagnosis network is constructed, solving the problems of insufficient precision and poor generality of traditional deep learning diagnosis models.

[0062] AsFigure 1 and Figure 2 As shown in the figure, one specific embodiment of the present scheme provides a power equipment fault diagnosis method based on causal domain adaptation and acoustic-vibration fusion, comprising the following steps:

[0063] S1, obtaining sound real-time signal data and vibration real-time signal data converted into soundprint representation signal when the target power equipment fails.

[0064] It should be noted that when the target power equipment fault diagnosis is performed, the collected data can be real-time data, and the current fault of the target power equipment can be judged according to the real-time data.

[0065] S2, inputting the sound real-time signal data when the target power equipment fails into a pre-trained sound-vibration conversion model, and the sound-vibration conversion model outputs a sound conversion signal; wherein a periodic consistent adversarial network is used as a basic network of the sound-vibration conversion model.

[0066] In one preferred embodiment, before inputting the sound real-time signal data when the target power equipment fails into the pre-trained sound-vibration conversion model, the method further comprises: pre-processing the sound real-time signal data and the vibration real-time signal data when the target power equipment fails.

[0067] In one optional embodiment, the method of pre-processing the sound real-time signal data and the vibration real-time signal data when the target power equipment fails comprises:

[0068] Discretely sampling the sound real-time signal data and the vibration real-time signal data of the target power equipment to obtain a data subset. Normalizing the data subset. Converting the normalized data subset into a mel spectrogram form to complete the preprocessing.

[0069] In one optional embodiment, after normalizing the data subset, it can be further uniformly cropped to realize signal scale alignment.

[0070] Specifically, the frequency of discretization sampling in the present scheme can be set to 16000Hz.

[0071] In one optional embodiment, converting the normalized data subset into a mel spectrogram form comprises:

[0072] Expressing the normalized data subset into a frequency spectrum through fast Fourier transform; converting the frequency spectrum into a mel frequency spectrum; and obtaining a mel spectrogram form through a mel filter bank. Specifically, the mel filter bank can perform logarithmic operation on all filter outputs, and then further perform discrete cosine transform (DCT) to obtain the mel spectrogram.

[0073] AsFigure 3 An embodiment of the application is shown in the schematic diagram of the sound-vibration conversion model principle. A generated network is used to process the sound real-time signal to obtain a sound conversion signal. The sound conversion signal and the vibration real-time signal are simultaneously input into a discriminant network for discrimination. Another generated network is used to process the sound conversion signal to obtain a reconstructed sound signal to meet the cycle consistency of the adversarial network.

[0074] In an optional embodiment, the sound-vibration conversion model is trained as follows:

[0075] The cycle-consistent adversarial network is used as the basic network (CycleGAN) of the sound-vibration conversion model, and the overall loss function of the sound-vibration conversion model is determined. Based on the basic network and the overall loss function of the sound-vibration conversion model, the gradient descent method is used to train the model parameters θ. After the training is completed, the trained sound-vibration conversion model is obtained.

[0076] Specifically, the overall loss function L m (θ) of the sound-vibration conversion model is as follows:

[0077] L m (θ) = L adv1 (θ) + L cyc (θ) + L id (θ) + L adv2 (θ)

[0078] The first adversarial loss term measures the difference between the data by Wasserstein distance, and is expressed as:

[0079]

[0080] The cycle-consistent loss term is expressed as:

[0081]

[0082] The identity mapping loss term is expressed as:

[0083]

[0084] The second adversarial loss term is expressed as:

[0085]

[0086] In the above formula, L adv1 (θ) represents the first adversarial loss term, L cyc (θ) represents the cycle-consistent loss term, L id (θ) represents the identity mapping loss term, and L adv2 (θ) represents the second adversarial loss term; G X-Y represents a generator that converts data X to Y, and D YE represents the expected value of the corresponding data.

[0087] The sound-vibration conversion model construction method provided by the scheme is based on the Wasserstein metric, improves the generation model based on the periodic consistent adversarial network by using the Wasserstein distance metric, constructs the sound-vibration conversion model, further reduces the difference between the converted data and the vibration data, solves the problem that the vibration data is insufficient to cause the diagnostic model to be difficult to train, and improves the diagnostic precision of the fault diagnosis model through data enhancement.

[0088] S3, input the sound conversion signal and vibration real-time signal data into the pre-trained causal domain adaptive fault diagnosis model, and the causal domain adaptive fault diagnosis model outputs the fault diagnosis result of the target power equipment.

[0089] In an optional embodiment, the training method of the causal domain adaptive fault diagnosis model is as follows:

[0090] S31, acquire sound signal data and vibration signal data when the power equipment fails, process the sound signal data by using the sound-vibration conversion model to obtain sound conversion training data; and the vibration signal data and the sound conversion training data are used for training of the causal domain adaptive fault diagnosis model.

[0091] Specifically, the sound signal data and the vibration signal data collected by the scheme are divided into training set data and test set data, the training set data and the test set data are discretized and sampled (sampling frequency is 16000Hz), and the sampled data is normalized according to the following formula:

[0092]

[0093] X = (X - X) / (X - X) norm X is the normalized training set data and test set data, X is the original data value of the training set data and the test set data, X max , and X min are the maximum value and the minimum value signal of the original data value, respectively.

[0094] The normalized data is uniformly cropped to realize signal scale alignment, and sound training set vibration training set sound test set vibration test set K is the signal length, N s , N v , N' s , and N' v are the data quantities of the corresponding data sets.

[0095] The sound training set vibration training set Sound test set Vibration test set Transformed into a mel-spectrogram form, specifically as follows:

[0096] The sound training set Vibration training set Sound test set Vibration test set Represented as a spectrogram, as follows:

[0097]

[0098] where X(n) is the transformed spectrum, x(k) represents a time-domain data sequence (corresponding to a single data in the sound training set Vibration training set and the sound test set Vibration test set K is the signal length.

[0099] For the obtained spectrum X(n), transform it into a mel-spectral representation through m = 2595log 10 (1 + f / 700), where f is the frequency expressed in hertz, and m is the corresponding mel frequency;

[0100] The obtained mel-spectrum is subjected to a logarithmic operation on all filter outputs through the following mel filter bank:

[0101]

[0102] Further discrete cosine transform (DCT) is performed, and a mel-spectrogram is obtained, as follows:

[0103]

[0104] S32, determine the overall loss function of the causal domain adaptive fault diagnosis model.

[0105] Specifically, the loss function is represented as follows:

[0106] L c (θ) = L domain (θ) + L cls (θ)

[0107] where L domain (θ) is a domain distribution matching loss, used to reduce the domain offset difference between different domain data features, and is represented as follows:

[0108]

[0109] where w dIt is to use the feature f(W) a ,X d Projected onto a 1-dimensional random fixed matrix, and L represents the number of features from different domains. cls (θ) represents the fault diagnosis loss of data in different domains.

[0110] L cls (θ) is represented as follows:

[0111]

[0112] S33. The causal domain adaptive fault diagnosis model is trained based on the vibration signal data, sound conversion training data, and the overall loss function of the causal domain adaptive fault diagnosis model. After training, the trained causal domain adaptive fault diagnosis model is obtained.

[0113] Specifically, vibration signal data and sound conversion training data are input into a feature extraction network with an attention mechanism. The network representation is as follows:

[0114] W a =Ψ(X1,X2,...,X) N )

[0115]

[0116] Where Ψ represents the network parameters of the attention mechanism, f(W) a X) is the output of the feature extraction network.

[0117] f(W) a X) Further, a hypothesis layer is used to diagnose faults in data from different domains, and the model parameters θ are trained using gradient descent.

[0118] In other embodiments, optimization methods such as stochastic gradient descent can be used to optimize the model training process.

[0119] The above-mentioned method for constructing a causal domain adaptive fault diagnosis model is based on causal domain adaptation theory. It combines attention mechanism and causal domain adaptation theory to construct a cross-domain acoustic and vibration fault diagnosis model. By introducing attention mechanism, the interference of noise data and outlier data on model training is solved. Based on causal domain adaptation theory, a cross-domain fault diagnosis model is constructed, which improves the stability and diagnostic accuracy of the fault diagnosis model and solves the problem of poor versatility of traditional diagnostic models.

[0120] As a specific application example of this solution, we will take mechanical fault diagnosis based on transformer acoustic and vibration data in the power sector as an example to explain this solution in detail:

[0121] Step 1: Obtain the sound and vibration real-time signal data of the target transformer fault, and pre-process:

[0122] Specifically, the sound and vibration real-time signal data of the equipment are collected by a sound and vibration collector, and are discretely sampled (sampling frequency 16000Hz). The normalized processing is performed on the sampled data:

[0123]

[0124] The normalized sound and vibration real-time signal data are converted into a mel-spectrogram form:

[0125] Further specifically, first, the fast Fourier transform is used to express it into a frequency spectrum graph:

[0126]

[0127] Second, the obtained frequency spectrum X(n) is converted into a mel-spectrum representation by m=2595log 10 (1+f / 700), and the obtained mel-spectrum is obtained by the following mel-filter bank to obtain a mel-spectrum representation:

[0128]

[0129] Step 2: Signal fusion based on sound-vibration conversion model:

[0130] First, a sound-vibration conversion model based on a cycle-consistent adversarial network (CycleGAN) is established, and the overall loss function is as follows:

[0131] L m (θ)=L adv1 (θ)+L cyc (θ)+L id (θ)+L adv2 (θ)

[0132] Wherein, L adv1 (θ) represents a first adversarial loss term, L cyc (θ) represents a cycle-consistent loss term, L id (θ) represents an identity mapping loss term, L adv2 (θ) represents a second adversarial loss term, and is specifically represented as:

[0133]

[0134]

[0135]

[0136]

[0137] The model is trained by a gradient descent method, and an optimization method such as a stochastic gradient descent method can be used to optimize the model training process. After the sound-vibration conversion model is trained, the sound real-time signal data is input into the sound-vibration conversion model to realize data fusion of the sound-vibration signal and obtain a sound conversion signal.

[0138] Step 3: Fault diagnosis based on the causal domain adaptive fault diagnosis model:

[0139] First, the vibration real-time signal data of different sources and the fusion data set converted by the sound-vibration conversion model are input into the feature extraction network with an attention mechanism, and the network representation form is as follows:

[0140] W a =Ψ(X1,X2,...,X N )

[0141]

[0142] wherein Ψ represents the network parameters of the attention mechanism, f(W a ,X) is the output of the feature extraction network, and f(W a ,X) is further input into the hypothesis layer to perform fault diagnosis on data of different domains, and the loss function is represented as follows:

[0143] L c (θ)=L domain (θ)+L cls (θ)

[0144] wherein L domain (θ) is the domain distribution matching loss, which is used to reduce the domain offset difference between the features of different domains, and is represented as follows:

[0145]

[0146] wherein w d is a random fixed matrix for projecting the feature f(W a ,X d ) to 1 dimension, and represent the number of features from different domains, and L cls (θ) is the fault diagnosis loss of the data of different domains, which is represented as follows:

[0147]

[0148] Like the sound-vibration conversion model, the model parameters θ of the causal domain adaptive fault diagnosis model are trained by a gradient descent method, and an optimization method such as a stochastic gradient descent method can be used to optimize the model training process.

[0149] In summary, after the sound-vibration conversion model and the causal domain adaptive fault diagnosis model are trained, the sound signal data is input into the sound-vibration conversion model for feature conversion, and then the converted data is input into the causal domain adaptive fault diagnosis model to realize fault diagnosis; the vibration data is directly input into the causal domain adaptive fault diagnosis model to realize fault diagnosis.

[0150] Embodiment 2

[0151] As shown in Figure 4 based on the same inventive concept as the above embodiment, the present application also provides a power equipment fault diagnosis device, comprising:

[0152] A data acquisition module is configured to acquire sound real-time signal data and vibration real-time signal data converted into soundprint representation signal when a target power equipment fails.

[0153] A data fusion module is configured to input the sound real-time signal data when the target power equipment fails into a pre-trained sound-vibration conversion model, and the sound-vibration conversion model outputs a sound conversion signal; wherein a periodic consistent adversarial network is used as a basic network of the sound-vibration conversion model.

[0154] A fault diagnosis module is configured to input the sound conversion signal and the vibration real-time signal data into a pre-trained causal domain adaptive fault diagnosis model, and the causal domain adaptive fault diagnosis model outputs a fault diagnosis result of the target power equipment.

[0155] Embodiment 3

[0156] As shown in Figure 5 the present application also provides an electronic device 100 for implementing the above-mentioned embodiment of a power equipment fault diagnosis method based on causal domain adaptation and sound-vibration fusion; the electronic device 100 comprises a memory 101, at least one processor 102, a computer program 103 stored in the memory 101 and executable on the at least one processor 102, and at least one communication bus 104. The memory 101 can be used to store the computer program 103, and the processor 102 can realize the steps of the embodiment 1 of the power equipment fault diagnosis method based on causal domain adaptation and sound-vibration fusion by running or executing the computer program stored in the memory 101 and calling the data stored in the memory 101.

[0157] The memory 101 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, application programs required by at least one function (such as a sound playing function, an image playing function, etc.), etc.; and the data storage area can store data (such as audio data) created according to the use of the electronic device 100, etc. In addition, the memory 101 can include a non-volatile memory, for example, a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device.

[0158] The at least one processor 102 can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The processor 102 can be a microprocessor or the processor 102 can also be any conventional processor, etc. The processor 102 is the control center of the electronic device 100, and connects all parts of the electronic device 100 through various interfaces and lines.

[0159] The memory 101 in the electronic device 100 stores a plurality of instructions to implement a power equipment fault diagnosis method based on causal domain adaptation and sound-vibration fusion, and the processor 102 can execute the plurality of instructions to implement:

[0160] Obtain sound real-time signal data at a target power equipment fault and vibration real-time signal data converted into a soundprint representation signal.

[0161] Input the sound real-time signal data at the target power equipment fault into a pre-trained sound-vibration conversion model, and the sound-vibration conversion model outputs a sound conversion signal; wherein a cycle-consistent adversarial network is used as a basic network of the sound-vibration conversion model.

[0162] Input the sound conversion signal together with the vibration real-time signal data into a pre-trained causal domain adaptation fault diagnosis model, and the causal domain adaptation fault diagnosis model outputs a fault diagnosis result of the target power equipment.

[0163] Embodiment 4

[0164] The modules / units integrated in the electronic device 100, if implemented in the form of software function units and sold or used as independent products, can be stored in a computer-readable storage medium. Based on such understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of each method embodiment described above can be implemented. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or some intermediate forms. The computer-readable medium can include any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, and read-only memory (ROM).

[0165] Those skilled in the art will appreciate that embodiments of the application can be supplied as a method, a system, or a computer program product. Therefore, the application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code.

[0166] The application is described with reference to flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing apparatus produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one or more flows and / or blocks Figure 1 The functions specified in one or more flows and / or blocks

[0167] These computer program instructions can also be stored in a computer-readable memory that can direct the computer or other programmable data processing apparatus to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including instruction means that implement the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one or more flows and / or blocks Figure 1 The functions specified in one or more flows and / or blocks

[0168] These computer program instructions can also be loaded into a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 Figure 1

[0169] In this description, references to "one embodiment", "an example", "certain examples" etc. mean that the particular feature, structure, material or characteristic being described is included in at least one embodiment or example of the application. The appearances of an expression like "in one embodiment", "in an example", "in certain examples" or the like in

[0170] Finally, it should be noted that the above-mentioned embodiments are merely given as an illustration of the technical solution of the present application, and are not intended to limit the present application. Although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that the specific embodiments described above are only used to explain the present application and are not intended to limit the present application. Any modification or equivalent replacement of the specific embodiments of the present application, which does not depart from the spirit and scope of the present application, should be covered by the claims of the present application.​​

Claims

1. A power equipment fault diagnosis method based on causal domain adaptation and acoustic vibration fusion, characterized in that, The method comprises the following steps: obtaining sound real-time signal data and vibration real-time signal data converted into a voiceprint representation signal when a target power equipment fails; inputting the sound real-time signal data when the target power equipment fails into a pre-trained sound-vibration conversion model, wherein the sound-vibration conversion model outputs a sound conversion signal; wherein a cycle-consistent generative adversarial network is used as a basic network of the sound-vibration conversion model; inputting the sound conversion signal and the vibration real-time signal data into a pre-trained causal domain adaptive fault diagnosis model, wherein the causal domain adaptive fault diagnosis model outputs a fault diagnosis result of the target power equipment; the training method of the sound-vibration conversion model comprises the following steps: using a cycle-consistent generative adversarial network as a basic network of the sound-vibration conversion model; determining an overall loss function of the sound-vibration conversion model; training model parameters based on the basic network of the sound-vibration conversion model and the overall loss function, and obtaining the trained sound-vibration conversion model after the training is completed; Overall loss function L m ( θ ) is represented as follows: In the above formula, represents a first adversarial loss term, represents a cycle consistency loss term, represents an identity mapping loss term, represents a second adversarial loss term; a first adversarial loss term is measured by a Wasserstein distance to measure the difference between data, and is expressed as: a cycle-consistent loss term is expressed as: an identity mapping loss term is expressed as: a second adversarial loss term is expressed as: In the above formula, a generator G that converts data X to Y, a domain discriminator D, E denotes an expected value of the corresponding data; the training method of the causal domain adaptive fault diagnosis model comprises the following steps: obtaining sound signal data and vibration signal data when a power equipment fails; processing the sound signal data by using the sound-vibration conversion model to obtain sound conversion training data; the vibration signal data and the sound conversion training data are used for training the causal domain adaptive fault diagnosis model; determining an overall loss function of the causal domain adaptive fault diagnosis model; training the causal domain adaptive fault diagnosis model based on the vibration signal data, the sound conversion training data, and the overall loss function of the causal domain adaptive fault diagnosis model, and obtaining the trained causal domain adaptive fault diagnosis model after the training is completed.

2. The power equipment fault diagnosis method based on causal domain adaptation and acoustic vibration fusion according to claim 1, characterized in that, Before the sound real-time signal data when the target power equipment fails is input into the pre-trained sound-vibration conversion model, the sound real-time signal data and the vibration real-time signal data when the target power equipment fails are preprocessed.

3. The power equipment fault diagnosis method based on causal domain adaptation and acoustic vibration fusion according to claim 2, characterized in that, The method for preprocessing the sound real-time signal data and the vibration real-time signal data when the target power equipment fails comprises the following steps: discretely sampling the sound real-time signal data and the vibration real-time signal data of the target power equipment to obtain a data subset; normalizing the data subset; converting the normalized data subset into a mel spectrogram form to complete the preprocessing.

4. The power equipment fault diagnosis method based on causal domain adaptation and acoustic vibration fusion according to claim 3, characterized in that, The method for converting the normalized data subset into a mel spectrogram form comprises the following steps: representing the normalized data subset into a frequency spectrum by using a fast Fourier transform; converting the frequency spectrum into a mel frequency spectrum; obtaining a mel spectrogram form by using a mel filter bank on the mel frequency spectrum.

5. A power equipment fault diagnosis apparatus for implementing the power equipment fault diagnosis method based on causal domain adaptation and acoustic vibration fusion according to claim 1, characterized in that, The method comprises the following steps: a data acquisition module is configured to obtain sound real-time signal data and vibration real-time signal data converted into a voiceprint representation signal when a target power equipment fails; The data fusion module is configured to input the sound real-time signal data of the target power equipment fault into a pre-trained sound-vibration conversion model, and the sound-vibration conversion model outputs a sound conversion signal; wherein a periodic consistent adversarial network is used as a basic network of the sound-vibration conversion model; The fault diagnosis module is configured to input the sound conversion signal and the vibration real-time signal data into a pre-trained causal domain adaptive fault diagnosis model, and the causal domain adaptive fault diagnosis model outputs a fault diagnosis result of the target power equipment.

6. An electronic device, comprising: The computer readable storage medium stores at least one instruction, and the at least one instruction is executed by the processor to implement the power equipment fault diagnosis method based on the causal domain adaptation and sound-vibration fusion.

7. A computer readable storage medium characterized by, The computer readable storage medium stores at least one instruction, and the at least one instruction is executed by the processor to implement the power equipment fault diagnosis method based on the causal domain adaptation and sound-vibration fusion.

Images