A Lightweight Neural Network Training Data Construction Method and System Based on Wavelet Adaptive Edge Feature Extraction

Abstract

This invention relates to the field of mechanical equipment condition monitoring and intelligent diagnosis, and discloses a lightweight neural network training data construction method and system based on wavelet adaptive edge feature extraction. The method includes: collecting vibration signals for different fault types, different fault degrees under the same fault type, and different operating conditions; dividing the collected long-sequence vibration signals into vibration sample sequences of length N by slicing them through a sliding window; performing continuous wavelet transform on each vibration sample sequence, calculating the modulus matrix of the wavelet coefficients, and performing image preprocessing to obtain a grayscale time-frequency image; performing adaptive edge fusion based on four-level statistical characteristics on the grayscale time-frequency image to obtain a fused edge matrix; converting the fused edge matrix into a binary image and performing image standardization to obtain a black-and-white binary image, which serves as the training data for the lightweight neural network. This invention enables accurate, rapid, and low-computing-power condition assessment of mechanical equipment faults.

Description

Technical Field

[0001] This invention relates to the field of mechanical equipment condition monitoring and intelligent diagnosis, and in particular to a lightweight neural network training data construction method and system based on wavelet adaptive edge feature extraction. Background Technology

[0002] In the field of mechanical equipment condition monitoring and intelligent diagnosis, vibration signals are the core carrier reflecting the health status of equipment. However, due to their nonlinear and non-stationary characteristics, direct analysis makes it difficult to accurately extract fault features. Traditional methods rely on manually designing time-domain and frequency-domain features, which requires a high level of experience and has weak generalization ability. While existing deep learning diagnostic solutions can automatically extract features, they suffer from complex network structures, large parameter counts, and high computational requirements, making them difficult to deploy on embedded devices or edge computing nodes in industrial settings. Time-frequency analysis (such as wavelet transform) can map vibration signals into time-frequency images, achieving the fusion of time-domain and frequency-domain information. However, existing time-frequency image feature extraction often uses a single edge detection algorithm, which suffers from edge information loss and insufficient noise resistance. Furthermore, there is a lack of efficient feature input schemes adapted to lightweight neural networks, making it difficult to balance diagnostic accuracy and deployment flexibility. Therefore, there is an urgent need for a fault diagnosis technology that can accurately extract time-frequency image features and is adapted to lightweight model input. Summary of the Invention

[0003] To address the aforementioned problems, the purpose of this invention is to provide a lightweight neural network training data construction method and system based on wavelet adaptive edge feature extraction, which can achieve accurate, fast, and low-computing-power condition assessment of mechanical equipment faults.

[0004] To achieve the above objectives, in a first aspect, the technical solution adopted by the present invention is as follows: a lightweight neural network training data construction method based on wavelet adaptive edge feature extraction, comprising: collecting vibration signals for different fault types of equipment, different fault degrees under the same fault type, and different operating conditions; dividing the collected long sequence vibration signals into vibration sample sequences of length N by slicing them through a sliding window; performing continuous wavelet transform on each vibration sample sequence, calculating the modulus matrix of wavelet coefficients, and performing image preprocessing to obtain a grayscale time-frequency image; performing adaptive edge fusion based on four-level statistical characteristics on the grayscale time-frequency image to obtain a fused edge matrix; converting the fused edge matrix into a binary image and performing image standardization processing to obtain a black-and-white binary image, which serves as the training data for the lightweight neural network.

[0005] Furthermore, the specific implementation process of image preprocessing includes:

[0006] The modulus matrix is ​​normalized to the [0,255] interval, and a three-channel color time-frequency image is generated by applying pseudo-color mapping. The pseudo-color image is then converted into a single-channel grayscale image. The grayscale image is processed by median filtering using a window with a set number of pixels, and the image is then scaled to a uniform size.

[0007] Furthermore, adaptive edge fusion based on four-level statistical characteristics is performed on the grayscale time-frequency image to obtain a fused edge matrix, including: The Sobel and Canny operators are applied to the grayscale time-frequency image respectively to obtain the Sobel edge map. And Canny edge map ; The gradient magnitude matrix is ​​obtained by calculating the grayscale time-frequency image using the Scharr operator. Statistical gradient magnitude matrix Gradient magnitude of all non-zero pixels Calculate their mean μ and standard deviation σ to determine the high threshold, medium threshold and low threshold; Constructing the Canny weight matrix and Sobel weight matrix Based on the current pixel gradient magnitude The interval in question is assigned a Canny weight matrix based on a four-level strategy with high, medium, and low thresholds. and Sobel weight matrix Assignment; Based on the assigned Canny weight matrix and Sobel weight matrix The edge intensity after fusion is calculated pixel by pixel, and the edge intensity is used as the element of the fused edge matrix to obtain the fused edge matrix.

[0008] Furthermore, the high threshold, medium threshold, and low threshold are as follows: High threshold ; Middle threshold ; low threshold .

[0009] Furthermore, construct the Canny weight matrix. and Sobel weight matrix Specifically: .

[0010] Furthermore, based on the current pixel gradient magnitude The interval in question is assigned a Canny weight matrix based on a four-level strategy with high, medium, and low thresholds. and Sobel weight matrix Assignment, specifically: The first level is the fine-edge region: when At that time, take , ; The second level is a finer edge region: when At that time, take , ; The third level is a coarser edge region: when At that time, take , ; The fourth level is the coarse edge area: when At that time, take , .

[0011] Furthermore, based on the assigned Canny weight matrix and Sobel weight matrix The edge intensity after fusion is calculated pixel by pixel, specifically as follows:

[0012] In the formula, Let (i,j) be the pixel value of the fused edge intensity image at coordinates (i,j). Canny edge map The pixel value at coordinates (i,j); Sobel edge graph The pixel value at coordinates (i,j).

[0013] Secondly, the technical solution adopted by this invention is as follows: a lightweight neural network training data construction system based on wavelet adaptive edge feature extraction, comprising: a data acquisition module, which collects vibration signals for different fault types of equipment, different fault degrees under the same fault type, and different operating conditions, and divides the collected long sequence vibration signals into vibration sample sequences of length N by slicing through a sliding window; a preprocessing module, which performs continuous wavelet transform on each vibration sample sequence, calculates the modulus matrix of wavelet coefficients, and performs image preprocessing to obtain a grayscale time-frequency image; a four-level adaptive edge fusion module, which performs adaptive edge fusion based on four-level statistical characteristics on the grayscale time-frequency image to obtain a fused edge matrix; and an image conversion and output module, which converts the fused edge matrix into a binary image and performs image standardization processing to obtain a black and white binary image, which serves as training data for the lightweight neural network.

[0014] Thirdly, the technical solution adopted by the present invention is: a computer-readable storage medium for storing one or more programs, wherein the one or more programs include instructions, which, when executed by a computing device, cause the computing device to perform any of the methods described above.

[0015] Fourthly, the technical solution adopted by the present invention is: a computing device comprising: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include instructions for performing any of the methods described above.

[0016] The present invention has the following advantages due to the adoption of the above technical solutions: 1. This invention first utilizes Morlet wavelet to perform continuous wavelet transform (CWT) on the acquired vibration signal, generating a grayscale image containing non-stationary features in the time-frequency domain. Then, based on the mean and variance of the image gradient amplitude, a four-level adaptive threshold interval is constructed, and dynamic weighted fusion of Sobel coarse edge detection and Canny fine edge detection is performed on the time-frequency image. In the high-gradient region, the Canny operator is mainly used to capture fine fault impact textures, while in the low-gradient transition region, the Sobel operator is mainly used to smooth background noise, ultimately generating a binarized edge feature map with redundant background removed.

[0017] 2. The four-level adaptive edge fusion strategy adopted in this invention can significantly suppress "pseudo-edge" interference caused by noise in industrial settings, compared with single edge detection or direct input of the original grayscale image, while fully preserving the high-frequency details of weak faults. By inputting the sparse binarized edge image into a lightweight neural network containing depthwise separable convolutions, the dimensionality and redundancy of the input data are greatly reduced. While ensuring diagnostic accuracy, the number of model parameters and computational complexity are significantly reduced, solving the problem that complex models are difficult to run in real time on industrial embedded devices.

[0018] 3. This invention dynamically generates four threshold intervals by utilizing the statistical distribution (mean and standard deviation) of gradient amplitude. Compared to traditional edge detection methods that rely on fixed parameters, this invention has extremely strong adaptability and generalization ability. Regardless of the equipment's speed or load conditions, even if the environmental noise floor fluctuates, the algorithm can automatically adjust the sensitivity of edge extraction based on the current signal characteristics, overcoming the defect of fixed thresholds being prone to failure under varying conditions. It can maintain stable diagnostic performance without the need for repeated manual parameter tuning for specific scenarios. Attached Figure Description

[0019] Figure 1 This is a flowchart of a lightweight neural network training data construction method based on wavelet adaptive edge feature extraction in an embodiment of the present invention. Detailed Implementation

[0020] To address the problems in existing mechanical fault diagnosis methods, such as the large number of parameters in deep learning models making them difficult to deploy at the edge, strong background noise interference in the original time-frequency images, and the difficulty of a single edge detection algorithm in balancing positioning accuracy and noise resistance, this invention provides a lightweight neural network training data construction method and system based on wavelet adaptive edge feature extraction. By adaptively fusing multi-scale edge features according to gradient statistical characteristics, the complex time-frequency image is converted into a high signal-to-noise ratio binary edge texture, achieving accurate, fast, and low-computing-power condition assessment of mechanical equipment faults.

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention are within the scope of protection of the present invention.

[0022] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0023] In one embodiment of the present invention, a method for constructing training data for a lightweight neural network based on wavelet adaptive edge feature extraction is provided. Vibration signals are converted into time-frequency images, and edge detection algorithms extract features from these images, which are then used as input training data for a lightweight neural network for lightweight neural network diagnostics in equipment health assessment. In this embodiment, as... Figure 1 As shown, the method includes the following steps: 1) Vibration signals are collected for different fault types, different fault degrees under the same fault type, and different operating conditions of the equipment. The collected long sequence vibration signals are then divided into vibration sample sequences of length N by sliding window slicing.

[0024] 2) Perform continuous wavelet transform (CWT) on each vibration sample sequence, calculate the modulus matrix of the wavelet coefficients, and then perform image preprocessing to obtain a grayscale time-frequency image.

[0025] 3) Perform adaptive edge fusion based on four-level statistical characteristics on the grayscale time-frequency image to obtain the fused edge matrix.

[0026] 4) Convert the fused edge matrix into a binary image and perform image normalization to obtain a black and white binary image, which can be used as training data for a lightweight neural network.

[0027] In step 2) above, Morlet wavelet is used as the mother wavelet, and its waveform is highly similar to the mechanical shock attenuation signal.

[0028] The specific steps for calculating the modulus matrix of wavelet coefficients are as follows: The scale sequence is set to an integer range of 1 to 128; a two-dimensional wavelet coefficient matrix is ​​obtained by convolving the vibration sample sequence with the mother wavelet function at different scales and translation positions. The modulus (i.e., the absolute value) of each complex element in the wavelet coefficient matrix is ​​calculated, resulting in a wavelet modulus matrix containing only non-negative real numbers.

[0029] In step 2) above, the specific implementation process of image preprocessing includes the following steps: 2.1) Normalization and Grayscale Conversion: The modulus matrix is ​​normalized to the [0, 255] interval, and a pseudo-color mapping (Jet color table) is applied to generate a three-channel color time-frequency image (RGB image). To reduce computation and focus on texture features for grayscale conversion, the pseudo-color image is converted into a single-channel grayscale image.

[0030] In this embodiment, the pseudo-color image is converted into a single-channel grayscale image according to the formula Gray=0.299×R+0.587×G+0.114×B.

[0031] Where Gray represents the converted grayscale pixel value (or brightness value); R represents the red channel component value of the corresponding pixel in the original image; G represents the green channel component value of the corresponding pixel in the original image; and B represents the blue channel component value of the corresponding pixel in the original image. This formula sets weighting coefficients based on the differences in human eye sensitivity to red, green, and blue (the human eye is most sensitive to green, followed by red, and least sensitive to blue), thus preserving the image brightness characteristics that conform to human visual perception.

[0032] 2.2) The grayscale image is processed by median filtering using a window with a set number of pixels, and the image is scaled to a uniform size as input for subsequent processing.

[0033] In this embodiment, a 3×3 pixel window is used to perform median filtering on the grayscale image.

[0034] In step 3) above, adaptive edge fusion based on four-level statistical characteristics is performed on the grayscale time-frequency image to obtain the fused edge matrix, including the following steps: 3.1) Apply the Sobel operator (calculating the gradients in the X and Y directions separately and then synthesizing them) and the Canny operator (double threshold detection) to the grayscale time-frequency image to obtain the Sobel edge map. And Canny edge map .

[0035] 3.2) Use the Scharr operator to calculate the gradient magnitude matrix of the grayscale time-frequency image. Statistical gradient magnitude matrix Gradient magnitude of all non-zero pixels Calculate the mean μ (mag_mean) and standard deviation σ (mag_std) to determine the high, medium and low thresholds.

[0036] In this embodiment, the high threshold, medium threshold, and low threshold are as follows: High threshold ; Middle threshold ; low threshold .

[0037] 3.3) Constructing the Canny weight matrix and Sobel weight matrix Based on the current pixel gradient magnitude The interval in question is assigned a Canny weight matrix based on a four-level strategy with high, medium, and low thresholds. and Sobel weight matrix Assign a value.

[0038] In this embodiment, the Canny weight matrix is ​​constructed. and Sobel weight matrix Specifically: .

[0039] Specifically, based on the current pixel gradient magnitude The interval in question is assigned a Canny weight matrix based on a four-level strategy with high, medium, and low thresholds. and Sobel weight matrix Assignment, specifically: The first level is the fine-edge region: when At that time, take , The gradient in this region is extremely high, and it belongs to the edge of a very well-defined fault characteristic.

[0040] The second level is a finer edge region: when At that time, take , The gradient in this region remains high, but a small amount of Sobel components need to be introduced to maintain coherence.

[0041] The third level is a coarser edge region: when At that time, take , In this region, the noise level is below the mean and Canny is prone to producing false edges. Therefore, the Canny weight is reduced and the Sobel weight is increased to take advantage of Sobel's smoothing properties.

[0042] The fourth level is the coarse edge area: when At that time, take , The gradient in this region is extremely low, and it is most likely background noise.

[0043] 3.4) Based on the assigned Canny weight matrix and Sobel weight matrix The edge intensity after fusion is calculated pixel by pixel, and the edge intensity is used as the element of the fused edge matrix to obtain the fused edge matrix.

[0044] In this embodiment, based on the assigned Canny weight matrix and Sobel weight matrix The edge intensity image after fusion is calculated pixel by pixel, specifically as follows:

[0045] In the formula, Let (i,j) be the pixel value of the fused edge intensity image at coordinates (i,j). Canny edge map The pixel value at coordinates (i,j); Sobel edge graph The pixel value at coordinates (i,j).

[0046] In step 4) above, the fused edge matrix is ​​converted into a binary image, and image normalization is performed to obtain a black and white binary image, including the following steps: 4.1) Binarized Image Generation: Due to the obtained fused edge image Its internal pixel values ​​are still continuous grayscale values ​​(Float type). To further reduce data redundancy and highlight the topological structure of the fault texture, it is converted into a binary image. A binarization threshold is set. (That is, half of the grayscale level). Traverse every pixel in the matrix. ,like If the value is 255 (white), then the point is considered a valid edge and assigned a value of 255. If the value is not specified, the point is considered a non-edge and assigned a value of 0 (black). This operation generates a high-contrast black-and-white binary image. .

[0047] 4.2) Image Standardization: 4.2.1) Image scaling: The image is scaled to the standard size specified by the network using bilinear interpolation. Considering the need for lightweight deployment, a size of 128×128 pixels is preferred.

[0048] 4.2.2) Numerical normalization: Mapping pixel values ​​from [0,255] to the range [0,1].

[0049] In one embodiment of the present invention, a lightweight neural network training data construction system based on wavelet adaptive edge feature extraction is provided, comprising: The data acquisition module collects vibration signals for different fault types, different fault degrees under the same fault type, and different operating conditions of the equipment. The long sequence of vibration signals collected is then divided into vibration sample sequences of length N by slicing through a sliding window. The preprocessing module performs continuous wavelet transform on each vibration sample sequence, calculates the modulus matrix of the wavelet coefficients, and performs image preprocessing to obtain a grayscale time-frequency image. The four-level adaptive edge fusion module performs adaptive edge fusion based on four-level statistical characteristics on grayscale time-frequency images to obtain a fused edge matrix; The image conversion output module converts the fused edge matrix into a binary image and performs image normalization to obtain a black and white binary image, which is used as training data for a lightweight neural network.

[0050] In the above embodiments, the specific implementation process of image preprocessing includes: The modulus matrix is ​​normalized to the [0,255] interval, and a three-channel color time-frequency image is generated by applying pseudo-color mapping. The pseudo-color image is then converted into a single-channel grayscale image. The grayscale image is processed by median filtering using a window with a set number of pixels, and the image is then scaled to a uniform size.

[0051] In the above embodiments, adaptive edge fusion based on four-level statistical characteristics is performed on the grayscale time-frequency image to obtain a fused edge matrix, including: The Sobel and Canny operators are applied to the grayscale time-frequency image respectively to obtain the Sobel edge map. And Canny edge map ; The gradient magnitude matrix is ​​obtained by calculating the grayscale time-frequency image using the Scharr operator. Statistical gradient magnitude matrix Gradient magnitude of all non-zero pixels Calculate their mean μ and standard deviation σ to determine the high threshold, medium threshold and low threshold; Constructing the Canny weight matrix and Sobel weight matrix Based on the current pixel gradient magnitude The interval in question is assigned a Canny weight matrix based on a four-level strategy with high, medium, and low thresholds. and Sobel weight matrix Assignment; Based on the assigned Canny weight matrix and Sobel weight matrix The edge intensity after fusion is calculated pixel by pixel, and the edge intensity is used as the element of the fused edge matrix to obtain the fused edge matrix.

[0052] In the above embodiments, the high threshold, medium threshold, and low threshold are respectively: High threshold ; Middle threshold ; low threshold .

[0053] In the above embodiments, the Canny weight matrix is ​​constructed. and Sobel weight matrix Specifically: .

[0054] In the above embodiments, based on the current pixel gradient magnitude The interval in question is assigned a Canny weight matrix based on a four-level strategy with high, medium, and low thresholds. and Sobel weight matrix Assignment, specifically: The first level is the fine-edge region: when At that time, take , ; The second level is a finer edge region: when At that time, take , ; The third level is a coarser edge region: when At that time, take , ; The fourth level is the coarse edge area: when At that time, take , .

[0055] In the above embodiments, based on the assigned Canny weight matrix and Sobel weight matrix The edge intensity after fusion is calculated pixel by pixel, specifically as follows:

[0056] In the formula, Let (i,j) be the pixel value of the fused edge intensity image at coordinates (i,j). Canny edge map The pixel value at coordinates (i,j); Sobel edge graph The pixel value at coordinates (i,j).

[0057] The system provided in this embodiment is used to execute the above-described method embodiments. For specific processes and details, please refer to the above embodiments, which will not be repeated here.

[0058] In one embodiment of the present invention, a computing device is provided. This computing device can be a terminal and may include a processor, a communication interface, memory, a display screen, and an input device. The processor, communication interface, and memory communicate with each other via a communication bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. When the computer programs are executed by the processor, they implement the methods described in the above embodiments. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The communication interface is used for wired or wireless communication with external terminals. Wireless communication can be achieved through Wi-Fi, a management network, NFC (Near Field Communication), or other technologies. The display screen can be a liquid crystal display (LCD) or an e-ink display. The input device can be a touch layer covering the display screen, or buttons, a trackball, or a touchpad mounted on the casing of the computing device, or an external keyboard, touchpad, or mouse. The processor can call logical instructions stored in the memory.

[0059] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and sold or used as independent products, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0060] In one embodiment of the present invention, a computer program product is provided, the computer program product including a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions, and when the program instructions are executed by a computer, the computer is able to perform the methods provided in the above-described method embodiments.

[0061] In one embodiment of the present invention, a non-transitory computer-readable storage medium is provided, which stores server instructions that cause a computer to perform the methods provided in the above embodiments.

[0062] The computer-readable storage medium provided in the above embodiments has a similar implementation principle and technical effect to the above method embodiments, and will not be described again here.

[0063] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations 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, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0064] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0065] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0066] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A lightweight neural network training data construction method based on wavelet adaptive edge feature extraction, characterized in that, include: Vibration signals are collected for different types of equipment failures, different degrees of failure under the same type of failure, and different operating conditions. The collected long-sequence vibration signals are then divided into vibration sample sequences of length N by slicing through a sliding window. For each vibration sample sequence, a continuous wavelet transform is performed, the modulus matrix of the wavelet coefficients is calculated, and image preprocessing is performed to obtain a grayscale time-frequency image. Adaptive edge fusion based on four-level statistical characteristics is performed on grayscale time-frequency images to obtain a fused edge matrix; The fused edge matrix is ​​converted into a binary image and then normalized to obtain a black-and-white binary image, which is used as training data for a lightweight neural network.

2. The lightweight neural network training data construction method based on wavelet adaptive edge feature extraction as described in claim 1, characterized in that, The specific implementation process of image preprocessing includes: The modulus matrix is ​​normalized to the [0,255] interval, and a three-channel color time-frequency image is generated by applying pseudo-color mapping. The pseudo-color image is then converted into a single-channel grayscale image. The grayscale image is processed by median filtering using a window with a set number of pixels, and the image is then scaled to a uniform size.

3. The lightweight neural network training data construction method based on wavelet adaptive edge feature extraction as described in claim 1, characterized in that, Adaptive edge fusion based on four-level statistical properties is performed on grayscale time-frequency images to obtain a fused edge matrix, including: The Sobel and Canny operators are applied to the grayscale time-frequency image respectively to obtain the Sobel edge map. And Canny edge map ; The gradient magnitude matrix is ​​obtained by calculating the grayscale time-frequency image using the Scharr operator. Statistical gradient magnitude matrix Gradient magnitude of all non-zero pixels Calculate their mean μ and standard deviation σ to determine the high threshold, medium threshold and low threshold; Constructing the Canny weight matrix and Sobel weight matrix Based on the current pixel gradient magnitude The interval in question is assigned a Canny weight matrix based on a four-level strategy with high, medium, and low thresholds. and Sobel weight matrix Assignment; Based on the assigned Canny weight matrix and Sobel weight matrix The edge intensity after fusion is calculated pixel by pixel, and the edge intensity is used as the element of the fused edge matrix to obtain the fused edge matrix.

4. The lightweight neural network training data construction method based on wavelet adaptive edge feature extraction as described in claim 3, characterized in that, The high threshold, medium threshold, and low threshold are as follows: High threshold ; Middle threshold ; low threshold .

5. The lightweight neural network training data construction method based on wavelet adaptive edge feature extraction as described in claim 3, characterized in that, Constructing the Canny weight matrix and Sobel weight matrix Specifically: .

6. The lightweight neural network training data construction method based on wavelet adaptive edge feature extraction as described in claim 5, characterized in that, Based on the current pixel gradient magnitude The interval in question is assigned a Canny weight matrix based on a four-level strategy with high, medium, and low thresholds. and Sobel weight matrix Assignment, specifically: The first level is the fine-edge region: when At that time, take , ; The second level is a finer edge region: when At that time, take , ; The third level is a coarser edge region: when At that time, take , ; The fourth level is the coarse edge area: when At that time, take , .

7. The lightweight neural network training data construction method based on wavelet adaptive edge feature extraction as described in claim 3, characterized in that, Based on the assigned Canny weight matrix and Sobel weight matrix The edge intensity after fusion is calculated pixel by pixel, specifically as follows: In the formula, Let (i,j) be the pixel value of the fused edge intensity image at coordinates (i,j). Canny edge map The pixel value at coordinates (i,j); Sobel edge graph The pixel value at coordinates (i,j).

8. A lightweight neural network training data construction system based on wavelet adaptive edge feature extraction, characterized in that, include: The data acquisition module collects vibration signals for different fault types, different fault degrees under the same fault type, and different operating conditions of the equipment. The long sequence of vibration signals collected is then divided into vibration sample sequences of length N by slicing through a sliding window. The preprocessing module performs continuous wavelet transform on each vibration sample sequence, calculates the modulus matrix of the wavelet coefficients, and performs image preprocessing to obtain a grayscale time-frequency image. The four-level adaptive edge fusion module performs adaptive edge fusion based on four-level statistical characteristics on grayscale time-frequency images to obtain a fused edge matrix; The image conversion output module converts the fused edge matrix into a binary image and performs image normalization to obtain a black and white binary image, which is used as training data for a lightweight neural network.

9. A computer-readable storage medium for storing one or more programs, characterized in that, The one or more programs include instructions that, when executed by a computing device, cause the computing device to perform any of the methods described in claims 1 to 7.

10. A computing device, characterized in that, include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including instructions for performing any of the methods described in claims 1 to 7.

Images