Early fault diagnosis method for power electronic equipment and computer equipment
Through the combination of multimodal feature extraction and Vision Transformer, the problem of inaccurate fault diagnosis in early power electronic equipment is solved, high-precision identification and diagnosis of faults is achieved, and the accuracy of fault prediction and equipment reliability are improved.
Patent Information
- Application Number
- CN202510121966.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-24
- Publication Date
- 2025-05-30
AI Technical Summary
The prior art is difficult to achieve accurate diagnosis of early failures of power electronic equipment, especially due to the weak and complex fault characteristics, it is difficult for traditional methods to effectively extract and identify them.
By collecting the current signal of power electronic equipment, using fast Fourier transform and variational modal decomposition to extract multimodal feature data, combining Euler's differential formula and thermal map mapping to generate feature images, and finally using Vision Transformer for feature fusion and classification to achieve accurate diagnosis of early faults.
It realizes high-precision diagnosis of early failures of power electronic equipment, improves the sensitivity and accuracy of fault identification, can detect and deal with faults earlier, and reduces equipment downtime and maintenance costs.
Smart Images
Figure CN120067854A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of equipment fault diagnosis, and particularly relates to an early fault diagnosis method for power electronic equipment and a computer device. Background Art
[0002] In recent years, with the development of the national high-tech industry, power electronics technology, as one of the main basic technologies, has been widely applied in various fields such as aerospace, military equipment, factory manufacturing, transportation, and power systems. Among them, as one of the main research directions in the power industry, the application of power electronic devices in the power system has penetrated into all aspects of power generation, storage, transmission, distribution, and power consumption. Therefore, issues related to the safe operation of power electronic devices have become increasingly prominent. However, in actual operation, the working environment of power electronic equipment is very complex, and external stress and internal electrothermal stress are very likely to cause various components in power electronic equipment to constantly withstand huge operating pressures, resulting in various faults in power electronic devices. Timely monitoring of the state of power electronic devices and implementing preventive maintenance on the basis of scheduled maintenance can reduce the possibility of equipment failure shutdown and maintenance costs, which is of great practical significance for economic production and safe production.
[0003] During the actual operation of power electronic equipment, due to the influence of the environment and internal stress, open-circuit faults, short-circuit faults, and aging failure of power devices will occur. Among them, open-circuit faults and short-circuit faults will cause obvious changes in the voltage parameters of power electronic devices in the equipment, and it is usually relatively easy to diagnose them. However, when early faults such as aging occur in power electronic devices, the parameters of the devices will drift, resulting in parametric faults in the equipment. However, the fault characteristics of different parametric faults have a small difference, making it difficult for general diagnostic methods to accurately diagnose parametric faults in power electronic equipment. Accurately judging the degree of parametric faults in the equipment is of great significance for formulating maintenance plans and ensuring equipment safety.
[0004] In the field of early fault diagnosis of existing electronic devices, most traditional diagnostic methods focus on extracting the single features of device operation data. For example, methods such as Fourier transform, EMD decomposition (Empirical Mode Decomposition), wavelet analysis, and Hilbert-Huang transform are used to perform simple digital feature analysis and calculation on the analog quantities of output voltage or current of the device. However, the feature data of a single modality often fails to fully characterize the fault features. In recent years, with the development of artificial intelligence and machine vision technologies, machine vision detection based on artificial intelligence algorithms has gradually gained attention in the field of fault diagnosis. However, currently in the field of machine vision fault diagnosis, it mostly targets some devices or component faults with appearance defects and abnormalities, and has a low ability to diagnose the early faults of power electronic devices with weak, complex or unclear fault phenomena, making it difficult to accurately judge the early faults of power electronic devices. Summary of the Invention
[0005] The purpose of the present invention is to provide a method for early fault diagnosis of power electronic devices, a computer device, a computer-readable storage medium, and a computer program product, which can achieve accurate diagnosis of the early faults of power electronic devices.
[0006] To achieve the above purpose, one aspect of the present invention provides a method for early fault diagnosis of power electronic devices, including:
[0007] Step S1, collect the current signal of the power electronic device, extract the frequency-domain features of the current signal through fast Fourier transform, and extract the time-frequency domain features of the current signal through variational mode decomposition. The original data of the current signal, the time-domain features of the original data, and the time-frequency domain features of the original data constitute multi-modal feature data;
[0008] Step S2, calculate the arccosine value of the eigenvalue of the multi-modal feature data: Wherein, φ represents the arccosine value corresponding to the eigenvalue;
[0009] Step S3, construct an Euler difference formula through Euler's formula: Substitute the arccosine value corresponding to the eigenvalue into the Euler difference formula to obtain the Euler difference eigenvalue of the multi-modal feature data;
[0010] Step S4, generate a feature image in multi-modal through heat map mapping of the Euler difference eigenvalue of the multi-modal feature data, and use the weighted fusion of the feature image in multi-modal as the input feature image of the Vision Transformer;
[0011] Step S5: Classify the input feature image using Vision Transformer to obtain the image category as the diagnosis result of different types of early faults of the power electronic device.
[0012] Preferably, in step S4, the weighting rule between the feature images in the multi-modal mode is as follows:
[0013]
[0014] where m 1 、m 2 、m 3 represent the mean values of the pixel points on the feature images in the three modes of the original data of the current signal, the time-domain features of the original data, and the time-frequency domain features of the original data respectively, and F 1 、F 2 、F 3 represent the feature images in the three modes respectively.
[0015] Preferably, in step S4, the heat map is mapped to the Viridis color map, and different numerical values are represented by color changes.
[0016] Preferably, step S1 further includes normalizing the multi-modal feature data through the following normalization formula:
[0017]
[0018] where X represents the set of multi-modal feature data, x i is the data point in the set, max(X) represents the maximum value in the set, min(X) represents the minimum value in the set, and represents the eigenvalue of the normalized multi-modal feature data.
[0019] Preferably, step S5 includes:
[0020] Evenly cut the input feature image into image patches, flatten each image patch into a one-dimensional vector, and the dimension of each vector is P 2 ×C, where H, W, and C represent the length, width, and height of the input feature image respectively, and p represents the size of the image patch;
[0021] Add positional encoding to each image patch to represent its position in the input feature image for calculating the attention between image patches;
[0022] Use the self-attention mechanism of the Transformer to calculate the attention between image patches to obtain the encoding of each image patch;
[0023] Encode the obtained image patches and input them into the fully connected layer of a multi-layer perceptron. After dimensional conversion, use the Softmax function to obtain the probability vector of the image belonging to each category, and take the maximum value as the image category, which is used as the diagnosis result of different types of early faults of power electronic devices.
[0024] Another aspect of the present invention provides a computer device, including a memory, a processor, and a computer program stored on the memory, and the processor executes the computer program to implement the steps of the above method.
[0025] Another aspect of the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the above method.
[0026] Another aspect of the present invention provides a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the steps of the above method.
[0027] According to the early fault diagnosis method, computer device, computer-readable storage medium, and computer program product of the above aspects of the present invention, it is possible to achieve accurate diagnosis of early faults of power electronic devices. BRIEF DESCRIPTION OF THE DRAWINGS
[0028] In order to more clearly illustrate the technical solutions of the present invention, the following will briefly introduce the drawings used in the description of the embodiments of the present invention. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings:
[0029] Figure 1 is a flowchart of the early fault diagnosis method for power electronic devices according to an embodiment of the present invention;
[0030] Figure 2 is a schematic diagram of the principle of the early fault diagnosis method for power electronic devices according to an embodiment of the present invention;
[0031] Figure 3 is a schematic diagram of a Vision Transformer according to an embodiment of the present invention;
[0032] Figure 4 is a schematic diagram of multi-head attention according to an embodiment of the present invention;
[0033] Figure 5 is a structural diagram of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0034] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Apparently, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the scope of protection of the present invention.
[0035] An embodiment of the present invention provides a method for early fault diagnosis of power electronic devices. By combining multi-modal data feature extraction, feature fusion, and the feature mining ability of Vision Transformer, high-precision fault diagnosis of early faults in power electronic devices is jointly achieved.
[0036] Figure 1 It is a flowchart of the method for early fault diagnosis of power electronic devices according to an embodiment of the present invention. As Figure 1 shown, the method for early fault diagnosis of power electronic devices according to the embodiment of the present invention includes steps S1 to S5. The following will be described in detail in conjunction with Figure 2 the schematic diagram to explain each step of the method according to the embodiment of the present invention in detail.
[0037] In step S1, the current signal of the power electronic device is collected, and the time-domain signal is converted into a frequency-domain signal through fast Fourier transform, so as to extract the frequency-domain features in the current signal; the current signal is decomposed by variational mode decomposition to construct multi-dimensional feature data, and the time-frequency domain in the feature data is captured, so as to extract the time-frequency domain features of the current signal and realize the extraction of multi-modal features. The original data of the collected current signal, the time-domain feature data of the original data, and the time-frequency domain feature data of the original data constitute multi-modal feature data. The multi-modal feature data is normalized through the normalization formula:
[0038]
[0039] where X represents the set of multi-modal feature data, x i is the data point in the set, max(X) represents the maximum value in the set, min(X) represents the minimum value in the set, represents the eigenvalue of the normalized multi-modal feature data.
[0040] In step S2, the arccosine value of the eigenvalue of the multi-modal feature data is obtained, so as to convert the feature data under different modes to the angular coordinate system to realize further feature extraction:
[0041] where, φ represents the arccosine value corresponding to the eigenvalue.
[0042] In step S3, the Euler difference formula is constructed through Euler's formula:
[0043]
[0044] Substitute the arccosine value corresponding to the eigenvalue into the Euler difference formula and then output the Euler difference eigenvalue. In this step, by substituting the arccosine value of the multimodal feature data into the Euler difference formula, the fault features in the feature signal are highlighted and the noise signal is suppressed.
[0045] In step S4, the Euler difference eigenvalue of the multimodal feature data is mapped through a heatmap to generate a feature image, which is used as the input feature image of the Vision Transformer after weighted fusion. In this step, the application of the heatmap mapping enables the feature image to retain both the fault features and the light weight during generation, thus generating a lightweight fault feature image and improving the diagnostic efficiency of the model.
[0046] In this embodiment, the multimodal feature data is three feature data modalities: the original data of the current signal, the time-domain feature data of the original data, and the time-frequency domain feature data of the original data. After generating the feature images in the three modalities, the feature images in the three feature data modalities are weighted and fused through feature fusion to generate the final fault data set, which is used as the input feature image of the Vision Transformer. When performing weighted fusion on the feature images, the weighting rules between the feature images in the three modalities are as follows:
[0047]
[0048] where m 1 , m 2 , m 3 respectively represent the means of the pixel points on the feature images generated in the three modalities, while F 1 , F 2 , F 3 respectively represent the feature images generated in the three modalities.
[0049] Heatmap mapping is a color mapping scheme for visualizing numerical data. It represents different numerical values through color changes, making the data easier to understand and interpret. The commonly used mapping is the Viridis color mapping. The characteristics of the Viridis color mapping are that it maps the starting numerical value with dark purple, transitions the intermediate numerical values with bright yellow, and finally represents the large numerical values with bright yellow.
[0050] The specific steps are as follows:
[0051] Step 1: Define the numerical range
[0052] Assume the numerical range is from 0 to 1, where 0 represents the minimum value and 1 represents the maximum value.
[0053] Step 2: Determine the color value
[0054] The Viridis color mapping maps each numerical value to a specific color value. In Viridis, colors are represented by three numbers, namely red (R), green (G), and blue (B). The range of each color is from 0 to 255.
[0055] For example: when the numerical value is 0, the color value is [68, 1, 84]; when the numerical value is 0.5, the color value is [253, 231, 37]; when the numerical value is 1, the color value is [254, 255, 0].
[0056] Step 3: Interpolation calculation
[0057] For numerical values between 0 and 1, interpolation calculation is required to obtain their corresponding color values. Interpolation is a method of estimating intermediate values through known data points.
[0058] For example, if there is a numerical value x = 0.25, the following formula can be used to calculate its corresponding RGB value:
[0059]
[0060]
[0061]
[0062] Where: (R 0 , G 0 , B 0 ) and (R 1 , G 1 , B 1 ) are the corresponding RGB values at the input values x 0 and x 1 .
[0063] In step S5, the Vision Transformer is used to classify the input feature image, and the image category is obtained as the diagnosis result of different types of early faults of power electronic devices.
[0064] Vision Transformer (ViT) is a diagnostic model that applies Transformer to image classification. Let the length, width, and height of the input image be H, W, and C respectively, and P represent the size of the image patch. The ViT model architecture is as Figure 3 shown, and the above step S5 specifically includes the following steps.
[0065] Step 1: Input feature image chunking. First, evenly cut the input feature image into image chunks. In the figure, they are image chunks a - i. Secondly, flatten each two-dimensional image chunk into a one-dimensional vector, and the dimension of each vector is P 2 ×C.
[0066] Step 2: Position encoding. Encode each image chunk, map each flattened image chunk to a D-dimensional vector, and add position encoding to each image chunk to represent its original position in the image, which is used to calculate the attention between image chunks.
[0067] Step 3: Self-attention encoding. Utilize the self-attention mechanism in Transformer, and use operations such as multi-head self-attention (Multi-Head Attention), multi-layer perceptron (Multi-Layer Perceptron, MLP), and layer normalization (Norm) to calculate the attention between image chunks and obtain the encoding of each image chunk.
[0068] Step 4: Calculate the image category. Input the image chunk encoding obtained by the Transformer encoder into the fully connected layer of the MLP. After dimension conversion, use the Softmax function to obtain the probability vector of the image belonging to each category, and take the maximum value to get the image category, which is used as the diagnosis result of different types of early faults of power electronic devices.
[0069] The multi-head attention mechanism is as Figure 4 shown. Linearly project the input into multiple feature subspaces and process it in parallel through several independent attention heads. Then the vectors are mapped in parallel to the final output. The process of the multi-head attention mechanism can be expressed as:
[0070]
[0071] Z i =Attention(Q i ,K i ,V i ), i = 1, 2, …, h
[0072] MultiHead(Q, K, V)=Concate(Z 1 ,Z 2 ,…,Z h )W 0
[0073] Among them, Q (Query) represents the query vector, K (Key) represents the key vector, V (Value) represents the value vector, h is the number of heads, is the output mapping matrix; Z i is the output vector of each head; are three different linear matrices, d mod el , d q , d k , d v respectively represent the vector dimensions of the model and Q, K, and V.
[0074] This process is expressed as a unified function as follows:
[0075]
[0076] In the formula, the self-attention weights are generated by the dot product operation between Q and K; the scaling factor and the Softmax function are used to normalize the self-attention weights, and the obtained weights are assigned to the corresponding elements of V, thereby generating the final output vector. Similar to the sparse connection of convolution, multi-head attention separates the input into h independent attention heads with -dimensional vectors, and merges the features of each head in parallel. Without increasing the additional computational cost, the diversity of the feature subspace is enriched.
[0077] In summary, the embodiments of the present invention address the problems in the prior art solutions, such as the low fault feature difference between different degrees of aging faults of power electronic devices in power electronic equipment, and the difficulty of effectively extracting degradation features by conventional methods, resulting in inaccurate early fault diagnosis and low recognition accuracy of power electronic equipment. Through multi-modal data feature extraction, feature fusion, and Vision Transformer, a PHM (Prognostic and Health Management) method for early faults of power electronic equipment is constructed. The current signal of the power electronic equipment is subjected to variational mode decomposition and fast Fourier transform to extract multi-modal features of the original data. After that, together with the original signal, feature extraction is performed through the Euler difference formula, and then it is converted into a lightweight fault feature image through heatmap mapping. Then, through feature fusion, the three feature images are weighted and fused to generate the final fault data set, which is used as the input feature image of Vision Transformer to achieve accurate diagnosis of early faults of power electronic equipment.
[0078] Compared with other fault diagnosis methods for power electronic equipment, the method of the embodiments of the present invention mainly has the following advantages:
[0079] 1) Extract multi-modal data of fault data through multiple feature extraction methods;
[0080] 2) Further highlight important features of multi-modal feature data through the Euler difference formula;
[0081] 3) The multimodal data with enhanced features through the Euler difference formula is converted into a feature image through heatmap mapping and used as the input feature image of the diagnostic model after weighting;
[0082] 4) Combine multimodal data feature extraction, feature fusion with ViSion Transformer to maximize the extraction of fault features in the feature image that can represent different types of early faults, and improve the sensitivity of the diagnostic model to early faults.
[0083] An embodiment of the present invention also provides a computer device, which can be a server, and its internal structure diagram can be as Figure 5 shown. The computer device includes a processor, a memory, and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the operation parameter data of each framework. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it implements the steps of the method of the embodiment of the present invention.
[0084] Those skilled in the art can understand that Figure 5 the structure shown in
[0085] is only a block diagram of some parts of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have a different component layout.
[0086] An embodiment of the present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the method of the embodiment of the present invention.
[0087] Only some exemplary embodiments of the present invention have been described by way of illustration above. Undoubtedly, for those of ordinary skill in the art, without departing from the spirit and scope of the present invention, the described embodiments can be modified in various different ways. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the protection scope of the claims of the present invention.
Claims
1. A method for early fault diagnosis of power electronic equipment, characterized in that: include: Step S1, collecting the current signal of the power electronic device, extracting the frequency domain characteristics of the current signal by fast Fourier transform, extracting the time-frequency domain characteristics of the current signal by variational mode decomposition, and the original data of the current signal, the time domain characteristics of the original data, and the time-frequency domain characteristics of the original data constitute multimodal feature data; Step S2, obtaining the arc cosine value of the eigenvalue of the multimodal feature data: in, φ represents the arccosine value corresponding to the eigenvalue; Step S3, constructing the Euler difference formula through the Euler formula: Substitute the arccosine value corresponding to the eigenvalue into the Euler difference formula to obtain the Euler difference eigenvalue of the multimodal feature data; Step S4, mapping the Euler differential eigenvalues of the multimodal feature data into a multimodal feature image through a heat map, and performing weighted fusion on the multimodal feature image as an input feature image of the Vision Transformer; Step S5: using Vision Transformer to classify the input feature image, and obtaining image categories as diagnosis results of different types of early faults of power electronic equipment.
2. The method according to claim 1, characterized in that In step S4, the weighting rule between feature images under multimodality is: Among them, m1, m2, and m3 represent the mean values of pixel points on the feature images of the three modes, namely, the original data of the current signal, the time domain characteristics of the original data, and the time-frequency domain characteristics of the original data, respectively. F1, F2, and F3 represent the feature images of the three modes, respectively.
3. The method according to claim 1 or 2, characterized in that In step S4, the heat map is mapped as a Viridis color map, and different values are represented by color changes.
4. The method according to claim 1 or 2, characterized in that: Step S1 also includes normalizing the multimodal feature data using the following normalization formula: Where X represents the set of multimodal feature data, x i is a data point in the set, max(X) represents the maximum value in the set, min(X) represents the minimum value in the set, Represents the eigenvalues of the normalized multimodal feature data.
5. The method according to claim 1 or 2, characterized in that: Step S5 includes: The input feature image is evenly cut into image blocks, flatten each image block into a one-dimensional vector, and the dimension of each vector is P 2 ×C, where H, W and C represent the length, width and height of the input feature image respectively, and P represents the size of the image block; Add a position code to each image block to indicate its position in the input feature image, which is used to calculate the attention between image blocks; Using the Transformer's self-attention mechanism, the attention between image blocks is calculated to obtain the encoding of each image block; The obtained image block encoding is input into the fully connected layer of the multi-layer perceptron. After dimension conversion, the Softmax function is used to obtain the probability vector of the image belonging to each category. The maximum value is taken as the image category, which is used as the diagnosis result of different types of early faults of power electronic equipment.
6. A computer device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 5.
7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.
8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.