Equipment fault sound source type diagnosis method and system, equipment and storage medium

Abstract

The invention provides an equipment fault sound source type diagnosis method and system, equipment and a storage medium. The method comprises the following steps: acquiring an original sound signal of a power equipment fault sound source; time domain data amplification processing is carried out on the original sound signal to obtain time domain amplification data, and the time domain data amplification processing method comprises the steps of signal time amplitude offset overturning, Gaussian noise jitter, amplitude scaling and data time axis random replacement; inputting the time domain augmentation and expansion data into a feature encoder, extracting key features, and mapping the key features to a high-dimensional space by the feature encoder to obtain high-dimensional features; mapping the obtained high-dimensional features back to a low-dimensional space through a projection head; and performing classification prediction on the output of the projection head through the prediction head, and finally obtaining a sound source type diagnosis result. According to the method, data enhancement and self-supervised comparative learning are combined, so that the problem of defect category imbalance in defect type identification of the power detection data can be solved, and the robustness and accuracy of a feature extraction network are improved.

Description

Technical Field

[0001] The present application relates to the technical field of power equipment status detection, and in particular to a method, system, device and storage medium for diagnosing equipment fault sound source types based on unbalanced defect categories. Background Art

[0002] Acoustic imaging locators are used to locate and diagnose fault sound sources of power equipment. As a non-contact, non-destructive testing technology, acoustic imaging locators have the advantages of high detection sensitivity, long detection distance, convenient and safe operation, and the ability to accurately locate fault points. In recent years, they have gradually been widely used in power partial discharge detection.

[0003] However, most acoustic imaging locators currently on the market do not have the function of diagnosing the sound source of power equipment faults or the diagnostic effect is not ideal. As a core component of the power system, the operating status of power equipment is directly related to the safety and stability of the entire power grid. However, the status detection of power equipment faces many challenges, including but not limited to: High voltage and strong electromagnetic environment: Power equipment usually operates in a high voltage and strong electromagnetic environment, which places extremely high demands on the anti-interference ability and safety of the detection equipment. Diverse equipment types: The power system contains various types of equipment such as transformers, switchgear, cables, busbars, etc. Each device has its own unique failure mode and detection difficulty. Complex working environment: Power equipment is often installed in closed or inaccessible spaces, such as substations, underground cable corridors, etc., which increases the difficulty and cost of detection.

[0004] Therefore, the existing technologies for diagnosing the sound sources of power equipment faults have the following main limitations: (1) Difficulty in data acquisition: Due to the particularity and safety requirements of power equipment, it is difficult to obtain sufficient fault sample data without affecting the normal operation of the equipment. (2) Insufficient algorithm adaptability: Traditional diagnostic algorithms are often based on ideal assumptions and sufficient data support, and are difficult to adapt to the actual situation where power equipment fault samples are scarce and the categories are unbalanced. (3) Lack of intelligent diagnosis: Most existing diagnostic methods rely on manual analysis and judgment, lack intelligent automatic diagnosis capabilities, and are difficult to meet the power system's needs for efficient and accurate diagnosis.

[0005] In view of this, the present application proposes a method, system, device and storage medium for diagnosing the type of fault sound source of power equipment under the condition of unbalanced defect categories, which can effectively improve the diagnostic accuracy of the fault sound source of power equipment. Summary of the invention

[0006] In order to solve the problem that existing defect samples have category imbalance and the ability of neural networks to extract features is greatly affected by data distribution bias, the present application provides a device fault sound source type diagnosis method, system, device and storage medium to solve the above-mentioned technical defects.

[0007] According to the first aspect of the present application, a method for diagnosing the type of sound source of equipment failure is proposed, the method comprising the following steps:

[0008] S1. Collecting the original sound signal of the fault sound source of the power equipment;

[0009] S2, performing time domain data augmentation processing on the original sound signal to obtain time domain augmented data;

[0010] S3, inputting the time domain augmented data into a feature encoder to extract key features, and the feature encoder maps the key features to a high-dimensional space to obtain high-dimensional features;

[0011] S4, mapping the obtained high-dimensional features back to the low-dimensional space through a projection head to obtain low-dimensional features;

[0012] S5. Classify and predict the low-dimensional features through a prediction head to finally obtain a sound source type diagnosis result.

[0013] Through the above technical solution, the present application combines data enhancement with self-supervised comparative learning, which can solve the problem of imbalanced defect categories in defect type identification of power detection data, and can improve the robustness and accuracy of the feature extraction network.

[0014] Preferably, the method for time domain data augmentation processing includes: a time amplitude offset flipping method, a Gaussian noise jitter method, an amplitude scaling method, and a data time axis random replacement method.

[0015] Through the above technical solution, combined with a variety of time domain data augmentation methods, the problem of data category imbalance in defect type identification can be effectively solved.

[0016] Preferably, in step S2, the offset flipping of the signal time amplitude specifically includes the following contents:

[0017] By moving the original sound signal forward or backward by a preset amount on the time axis, and looping to fill or cut off the missing part;

[0018] Adding a DC component of random size to the original sound signal so that the original sound signal moves vertically as a whole;

[0019] The signal image corresponding to the original sound signal is flipped horizontally and / or flipped upside down to obtain time domain augmented data.

[0020] Further preferably, in step S2,

[0021] The Gaussian noise jitter method comprises: obtaining the time domain augmentation data by adding random noise obeying Gaussian distribution to the original sound signal;

[0022] The amplitude scaling method comprises: obtaining the time domain augmented data by multiplying the original sound signal by a random factor that obeys a normal distribution;

[0023] The data time axis random replacement method includes: randomly replacing the time axis of the original sound signal to disrupt the original time sequence and obtain the time domain augmented data.

[0024] Preferably, in step S1, collecting the original sound signal of the fault sound source of the electric power equipment specifically includes: collecting any discharge acoustic data such as surface discharge, suspended discharge or low-frequency discharge.

[0025] Preferably, in step S3, the feature encoder includes a first feature encoder and a second feature encoder, and the two time domain augmented data obtained by two different time domain data augmentation processing methods are input into the first feature encoder and the second feature encoder respectively, and the key features are obtained by weight sharing optimization between the first feature encoder and the second feature encoder.

[0026] Preferably, the projection head comprises a first projection head and a second projection head, and the first projection head and the second projection head optimize output by weight sharing to obtain low-dimensional features.

[0027] Preferably, the prediction head also includes a classifier, which is composed of a fully connected layer, and the parameters of the classifier are optimized using a cross entropy loss function; the classifier is used to output the predicted probability or confidence of each category in the sound source type diagnosis result.

[0028] Through the above technical solution, in the classification learning stage, the pre-trained feature encoder is used to fine-tune the labeled data, and combined with the classifier composed of the fully connected layer, by optimizing the classification loss, it is possible to accurately classify the data type of power equipment (for example: high-voltage switchgear discharge type).

[0029] In a second aspect, the present application provides a device fault sound source type diagnosis system, the system comprising:

[0030] A signal acquisition module, configured to collect original sound signals of fault sound sources of power equipment;

[0031] A signal acquisition module is used to collect the original sound signal of the fault sound source of the power equipment;

[0032] A time domain data augmentation module, used to perform time domain data augmentation processing on the original sound signal to obtain time domain augmented data;

[0033] A feature extraction module is used to extract key features from the time domain augmented data; and the feature extraction module maps the key features to a high-dimensional space to obtain high-dimensional features;

[0034] A feature mapping module, used for mapping the high-dimensional features back to a low-dimensional space;

[0035] In a third aspect, the present application provides a terminal device, including a processor, a memory, and a computer program stored in the memory, wherein the computer program is executed by the processor to implement a method for diagnosing the type of sound source of a device fault as described in any one of the above.

[0036] In a fourth aspect, the present application provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, a method for diagnosing the type of sound source of a device fault as described above is implemented.

[0037] Compared with the prior art, the beneficial results of the present invention are:

[0038] Improve the robustness of feature extraction: Through unsupervised self-supervised contrastive learning in the representation learning stage, combined with a variety of time domain data augmentation methods, the problem of data category imbalance in defect type identification is effectively solved, allowing the feature extraction network to learn more robust and generalized feature representations.

[0039] Improve classification accuracy: In the classification learning stage, the pre-trained feature encoder is used to fine-tune the labeled data, and combined with the classifier composed of the fully connected layer, the classification loss is optimized to achieve accurate classification of power equipment data types. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Other features, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting embodiments made with reference to the following drawings:

[0041] Figure 1 is a flow chart of a method for diagnosing equipment fault sound source types according to the present application;

[0042] Figure 2 is a schematic diagram of two-stage classifier learning according to the present application;

[0043] Figure 3 is a structural diagram of a device fault sound source type diagnosis system according to the present application;

[0044] Figure 4It is a structural diagram of a computer system suitable for implementing an electronic device of an embodiment of the present application. DETAILED DESCRIPTION

[0045] The present application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the relevant invention, rather than to limit the invention. It should also be noted that, for ease of description, only the parts related to the relevant invention are shown in the accompanying drawings.

[0046] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0047] Figure 1 The flowchart of the device fault sound source type diagnosis method of the present application is shown as follows: Figure 1 As shown, the method comprises the following steps:

[0048] S1. Collect the original sound signal of the fault sound source of the power equipment.

[0049] In this embodiment, since the acoustic imaging locator has an array of sound collecting elements, it can collect local discharge acoustic data (original sound signals) caused by different defects such as surface discharge, suspended discharge, and low-frequency discharge to support the training of the neural network method model. The acoustic imaging sample spectra of different discharge types obtained have their own characteristics, reflecting the acoustic signal characteristics of different defect types. By deeply exploring the characteristics of these spectra, the discharge type can be effectively and accurately judged, making it possible to more accurately identify the discharge type in local discharge detection.

[0050] S2. Perform time domain data augmentation processing on the original sound signal to obtain time domain augmented data, wherein the time domain data augmentation processing method includes: an offset flip method of the signal time amplitude, a Gaussian noise jitter method, an amplitude scaling method, and a data time axis random replacement method.

[0051] In this embodiment, compared with supervised learning, due to the lack of real labels, the training process of self-supervised contrastive learning is more dependent on the reasonable combination of data augmentation. Although many data augmentation schemes have been widely used in the field of images, for one-dimensional mechanical monitoring signals, effective time domain augmentation methods that do not destroy the physical meaning of the original signal remain to be explored.

[0052] According to the periodicity and time domain characteristics of the original sound signal of the equipment fault sound source, the following time domain data augmentation methods suitable for the sound signal can be used:

[0053] (1) Offset flip of signal time amplitude

[0054] The idea of ​​signal time shifting of the original sound signal is to randomly move the monitoring signal forward or backward on the time axis. After the movement, the missing elements at the corresponding position are filled with overflow fragments. The idea of ​​signal amplitude shifting is to add a DC component to the monitoring signal so that the overall signal is slightly shifted upward or downward. The image corresponding to the original sound signal is flipped horizontally and upside down.

[0055] (2) Gaussian noise jitter

[0056] Gaussian noise dithering is performed on the original sound signal. This application generates random noise from a normal distribution and then adds the noise to the input data.

[0057] (3) Amplitude scaling

[0058] The basic principle of amplitude scaling is to scale the input data by multiplying the original sound signal by a random factor that obeys a normal distribution to obtain time domain augmented data.

[0059] (4) Random replacement of data time axis

[0060] The time axis of the original sound signal is randomly permuted. Its purpose is to introduce temporal uncertainty so that the model can better adapt to different time sequences.

[0061] A neural network method is used to identify defect types. In view of the imbalance of defect categories in training samples, a combination of data enhancement and self-supervised contrastive learning is used for model training, and a feature encoder is used as the feature extraction network.

[0062] Figure 2 A schematic diagram of a two-stage classifier learning according to the present application is shown, Figure 2 As shown, the application also includes the following steps:

[0063] S3. Input the time domain augmented data into the feature encoder to extract key features, and the feature encoder maps the key features to a high-dimensional space to obtain high-dimensional features.

[0064] Specifically, a convolutional neural network (CNN) is used as a feature encoder. In image processing, a convolutional neural network (CNN) extracts key features of an image through a convolutional layer and gradually maps the features to a higher-dimensional space through a pooling layer.

[0065] S4. Map the obtained high-dimensional features back to the low-dimensional space through the projection head to obtain low-dimensional features.

[0066] Specifically, a point in a high-dimensional space is mapped to a low-dimensional space through a linear transformation. In linear projection, the original high-dimensional features are transformed by a linear matrix to obtain a representation in a low-dimensional space.

[0067] S5. The output of the projection head is classified and predicted by the prediction head to finally obtain the sound source type diagnosis result.

[0068] To solve the problem of imbalanced defect categories in defect type recognition, how to obtain a robust feature extraction network is the key to solving the problem. Figure 2 As shown, this application first uses unlabeled data for representation learning. In addition, the defect type recognition task can essentially be regarded as a binary classification problem in machine learning. Therefore, for the pre-trained feature encoder E, the unbalanced data with labels is used for fine-tuning, and the classifier C is trained at the same time. The classifier in this task is composed of a fully connected layer, and its parameter update is optimized by classification loss. The cross entropy (CE) loss L used in classification CE .

[0069] The representation learning stage mainly includes five parts: time domain data augmentation module T, feature encoder E, projection head G, prediction head P and symmetric loss. The classification learning stage is to fine-tune the labeled data on the encoder trained in the representation learning stage, and then classify and judge whether the high-voltage switchgear is discharged.

[0070] Specifically, in the representation learning phase, this phase includes:

[0071] The data input step is to input the original sound signal data x into the time domain data augmentation module T for two amplification processes to obtain the time domain augmented data x j and the time domain amplification data x i , and then respectively transform the time domain augmented data x j and the time domain amplification data x i Input into different encoders (first encoder and second encoder);

[0072] A weight sharing processing step, processing the data output by the encoder E through multiple projection heads G, wherein the multiple projection heads G share the same weight;

[0073] Gradient interruption step: a gradient interruption mechanism is implemented between the encoder E and the projection head G to improve the stability and efficiency of model training;

[0074] Feature extraction step, generating and outputting feature representation L ij , this feature representation is used for subsequent classification or prediction tasks;

[0075] The prediction head generation step uses the feature representation Lij Generate a prediction head P for prediction or classification tasks.

[0076] In this classification learning stage, this stage includes:

[0077] The fine-tuning step uses the labeled dataset to fine-tune the encoder E while maintaining the weight sharing property of the projection head G;

[0078] Classifier training step, train classifier C, which accepts feature representation L ij As input, and output classification results;

[0079] The category determination step determines, based on the output result of classifier C, which category the data to be identified belongs to, where the categories include but are not limited to category 1, category 2, and category 3.

[0080] The above-mentioned representation learning stage and classification learning stage together constitute the training process of the deep learning model, through which the accuracy and robustness of the model in defect recognition are improved.

[0081] In order to meet the computing power requirements of the proposed deep learning algorithm, this application can also focus on the acceleration and optimization of the neural network algorithm. Explore and adopt advanced hardware acceleration technology to increase the training and inference speed of the algorithm. In addition, consider model pruning, quantization and other methods at the algorithm level to reduce the model size, reduce computational complexity, and further improve the algorithm's operating efficiency.

[0082] Due to objective conditions, the number of samples collected may be small. In order to enable the model to learn and generalize in the face of a small number of labeled samples, the model can use transfer learning to help the model better utilize limited training data. Transfer learning is to improve the learning effect on another related task by using the knowledge and features learned on one task. The core idea is to use the source domain to train a pre-trained model, and then fine-tune the model based on the training data of the target domain. In this way, the learning process of the target task can be accelerated, especially when the training data of the target task is limited or insufficient. In order to solve the problem of imbalanced defect categories in defect type recognition, a variety of augmentation methods suitable for time domain data are introduced. Several widely used time domain data augmentation methods include time amplitude offset, time amplitude flipping, Gaussian noise jitter, etc. Through data augmentation, the number of defect samples of different types is balanced, and the model's feature learning ability for samples of a small number of categories is improved.

[0083] Further references Figure 3 As an implementation of the above method, in a second aspect, the present application provides an embodiment of a structure diagram 300 of a device fault sound source type diagnosis system, which can be specifically applied to various electronic devices. The system 300 includes the following modules:

[0084] The signal collection module 310 is configured to collect the original sound signal of the fault sound source of the power equipment;

[0085] The time domain data augmentation module 320 is configured to perform time domain data augmentation processing on the original sound signal to obtain time domain augmented data, wherein the time domain data augmentation processing method includes: offset flipping of signal time amplitude, Gaussian noise jitter, amplitude scaling and random replacement of data time axis;

[0086] The feature extraction module 330 is configured to input the time domain augmented data into a feature encoder to extract key features, and the feature encoder maps the key features to a high-dimensional space to obtain high-dimensional features;

[0087] A feature mapping module 340 is configured to map the extracted high-dimensional features back to a low-dimensional space through a projection head;

[0088] The result output module 350 is configured to classify and predict the output of the projection head through the prediction head, and finally obtain the sound source type diagnosis result.

[0089] In a third aspect, the present application provides a terminal device, including a processor, a memory, and a computer program stored in the memory, wherein the computer program is executed by the processor to implement a method for diagnosing the type of sound source of a device fault as described in any one of the above.

[0090] In a fourth aspect, the present application provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, a method for diagnosing the type of sound source of a device fault as described above is implemented.

[0091] Reference below Figure 4 , which shows a schematic diagram of the structure of a computer system 400 of a terminal device or server suitable for implementing an embodiment of the present application. Figure 4 The terminal device or server shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0092] like Figure 4 As shown, the computer system 400 includes a central processing unit (CPU) 401, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 402 or a program loaded from a storage part 408 into a random access memory (RAM) 403. In the RAM 403, various programs and data required for the operation of the computer system 400 are also stored. The CPU 401, the ROM 402, and the RAM 403 are connected to each other via a bus 404. An input / output (I / O) interface 405 is also connected to the bus 404.

[0093] The following components are connected to the I / O interface 405: an input section 406 including a keyboard, a mouse, etc.; an output section 407 including a liquid crystal display (LCD), etc. and a speaker, etc.; a storage section 408 including a hard disk, etc.; and a communication section 409 including a network interface card such as a LAN card, a modem, etc. The communication section 409 performs communication processing via a network such as the Internet. A drive 410 is also connected to the I / O interface 405 as needed. A removable medium 411, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 410 as needed, so that a computer program read therefrom is installed into the storage section 408 as needed.

[0094] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 409, and / or installed from the removable medium 411. When the computer program is executed by the central processing unit (CPU) 401, the above functions defined in the method of the present application are executed. It should be noted that the computer-readable medium described in the present application can be a computer-readable signal medium or a computer-readable medium or any combination of the above two. The computer-readable medium can be, for example, - but not limited to - a system, device or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples of computer-readable media may include, but are not limited to, an electrical connection with one or more conductors, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, an apparatus or a device. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable medium, which may send, propagate, or transmit a program for use by or in combination with an instruction execution system, an apparatus or a device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0095] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as C or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0096] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0097] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above invention concept. For example, the above features are replaced with the technical features with similar functions disclosed in this application (but not limited to) by each other to form a technical solution.

Claims

1. A method for diagnosing the type of equipment fault sound source, characterized in that: The method comprises the following steps: S1. Collecting the original sound signal of the fault sound source of the power equipment; S2. Perform time domain data augmentation processing on the original sound signal to obtain time domain augmented data; the time domain data augmentation processing method includes: a time amplitude offset flip method, a Gaussian noise jitter method, an amplitude scaling method, and a data time axis random replacement method. S3, inputting the time domain augmented data into a feature encoder to extract key features, and the feature encoder maps the key features to a high-dimensional space to obtain high-dimensional features; S4, mapping the obtained high-dimensional features back to a low-dimensional space through a linear projection head to obtain low-dimensional features; S5. Classify and predict the low-dimensional features through a prediction head to finally obtain a sound source type diagnosis result.

2. The device fault sound source type diagnosis method according to claim 1, characterized in that: In step S2, the offset flipping of the signal time amplitude specifically includes the following contents: By moving the original sound signal forward or backward by a preset amount on the time axis, and looping to fill or cut off the missing part; Adding a DC component of random size to the original sound signal so that the original sound signal moves vertically as a whole; The signal image corresponding to the original sound signal is flipped horizontally and / or flipped upside down to obtain time domain augmented data.

3. The method for diagnosing the type of equipment fault sound source according to claim 1, characterized in that: In step S2, the Gaussian noise jitter method includes: obtaining the time domain augmentation data by adding random noise obeying Gaussian distribution to the original sound signal; The amplitude scaling method comprises: obtaining the time domain augmented data by multiplying the original sound signal by a random factor that obeys a normal distribution; The data time axis random replacement method includes: randomly replacing the time axis of the original sound signal to disrupt the original time sequence and obtain the time domain augmented data.

4. The device fault sound source type diagnosis method according to claim 1, characterized in that: The original sound signal includes: collecting any discharge acoustic data such as surface discharge, suspension discharge or low-frequency discharge.

5. The device fault sound source type diagnosis method according to claim 1, characterized in that: In step S3, the feature encoder includes a first feature encoder and a second feature encoder, and the two time domain augmented data obtained by two different time domain data augmentation processing methods are respectively input into the first feature encoder and the second feature encoder, and the key features are obtained by weight sharing optimization between the first feature encoder and the second feature encoder.

6. The method for diagnosing the type of equipment fault sound source according to claim 1, characterized in that: In step S4, the projection head includes a first projection head and a second projection head, and the first projection head and the second projection head optimize output by weight sharing to obtain low-dimensional features.

7. The method for diagnosing the type of equipment fault sound source according to claim 1, characterized in that: The prediction head also includes a classifier, which is composed of a fully connected layer. The parameters of the classifier are optimized using a cross entropy loss function; the classifier is used to output the predicted probability or confidence of each category in the sound source type diagnosis result.

8. A system for diagnosing equipment fault sound source types, characterized in that: The system comprises: A signal acquisition module is used to collect the original sound signal of the fault sound source of the power equipment; A time domain data augmentation module, used to perform time domain data augmentation processing on the original sound signal to obtain time domain augmented data; A feature extraction module is used to extract key features from the time domain augmented data; and the feature extraction module maps the key features to a high-dimensional space to obtain high-dimensional features; A feature mapping module, used for mapping the high-dimensional features back to a low-dimensional space; The result output module is used to classify and predict the output of the projection head to finally obtain a sound source type diagnosis result.

9. A terminal device, characterized in that: The method comprises a processor, a memory and a computer program stored in the memory, wherein the computer program is executed by the processor to implement the device fault sound source type diagnosis method according to any one of claims 1 to 8.

10. A computer-readable storage medium, wherein a computer program is stored in the medium, and when the computer program is executed by a processor, the method for diagnosing the type of sound source of equipment fault according to any one of claims 1 to 8 is implemented.

Images