Bearing fault detection method, system, medium, and electronic device

Abstract

The application discloses a bearing fault detection method, system, medium and electronic equipment, and the method comprises the following steps: collecting one-dimensional time series vibration signals of a rolling bearing to be diagnosed under a running state; reconstructing the one-dimensional time series vibration signals into a phase space trajectory matrix representing system dynamics trajectories; converting the phase space trajectory matrix into a two-dimensional gray-scale image; inputting the two-dimensional gray-scale image into a preset deep image denoising convolutional neural network to output a feature image corresponding to the two-dimensional gray-scale image after noise reduction; inputting the feature image after noise reduction into a preset deep convolutional classification neural network for feature extraction and classification, and outputting a fault state category corresponding to the rolling bearing to be diagnosed. Therefore, by adopting the embodiments of the application, the robustness and accuracy of bearing fault diagnosis can be significantly enhanced, and high-reliability fault state determination is realized.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a bearing fault detection method, system, medium, and electronic device. Background Technology

[0002] Rolling bearings are critical transmission components in various rotating machinery (such as aircraft engines, wind turbines, and industrial machine tools), and their operating condition directly affects the safety, stability, and service life of the equipment. In industrial settings, due to the high-speed rotation of equipment, variable loads, and the complexity of the environment (electromagnetic, vibration, and temperature), the bearing vibration signals collected by sensors inevitably become contaminated with a large amount of random, non-stationary, and strong noise interference. This severely obscures the dynamic characteristics that characterize early, subtle bearing faults, posing a significant challenge to accurate condition monitoring and fault diagnosis.

[0003] In related technologies, there are two main technical approaches for fault diagnosis of rolling bearings. The first is traditional methods based on signal processing. Vibration signal analysis is currently a common method for bearing health management. However, in practical applications, vibration signals are often contaminated by the inherent vibrations of mechanical equipment, environmental noise, and electromagnetic interference. This leads to limitations in traditional methods (such as Fourier transform and wavelet transform) or modern decomposition methods (such as EMD and VMD) when dealing with complex non-stationary noise, such as mode aliasing and weak instantaneous noise removal capabilities, resulting in a significant decrease in diagnostic accuracy. The second is modern methods based on deep learning. While existing one-dimensional signal processing techniques using deep learning (such as 1D-CNN and Transformer) demonstrate good feature extraction capabilities, their ability to capture original temporal dynamic correlation features is limited when dealing with high-noise environments (such as low signal-to-noise ratio conditions like -6dB), making them prone to misjudgment. Summary of the Invention

[0004] This application provides a bearing fault detection method, system, medium, and electronic device. To provide a basic understanding of some aspects of the disclosed embodiments, a brief summary is given below. This summary is not intended as a general description, nor is it intended to identify key / important components or describe the scope of protection of these embodiments. Its sole purpose is to present some concepts in a simple form as a prelude to the detailed description that follows.

[0005] In a first aspect, embodiments of this application provide a bearing fault detection method, the method comprising: Acquire one-dimensional time-series vibration signals of the rolling bearing to be diagnosed under operating conditions; The one-dimensional time-series vibration signal is reconstructed into a phase space trajectory matrix representing the dynamic trajectory of the system; the phase space trajectory matrix is ​​then converted into a two-dimensional grayscale image. A two-dimensional grayscale image is input into a pre-defined depth image denoising convolutional neural network, and the output is a denoised feature image corresponding to the two-dimensional grayscale image. The denoised feature image is input into a pre-defined depth convolutional neural network for feature extraction and classification, and the fault state category corresponding to the rolling bearing to be diagnosed is output.

[0006] Optionally, the one-dimensional time-series vibration signal is reconstructed into a phase space trajectory matrix characterizing the system's dynamic trajectory, including: Obtain the total number of sampling points and amplitude of a one-dimensional time-series vibration signal; Based on the preset autocorrelation function, the total number of sampling points, and the amplitude, calculate the autocorrelation coefficient of the signal at each delay time. The optimal delay time is defined as the time when the autocorrelation coefficient first drops to zero or close to zero. The pseudo-nearest neighbor method is used to calculate the embedding dimension required to map a one-dimensional signal to a high-dimensional phase space; By adjusting the time delay and embedding dimension, the phase space of a one-dimensional time-series vibration signal is reconstructed to generate a phase space trajectory matrix that characterizes the dynamic trajectory of the system.

[0007] Optionally, the default function expression for the autocorrelation function is:

[0008] in, The signal has a delay time of The autocorrelation coefficient at time; It is a time delay, representing the time interval between one state component and the next state component when reconstructing the phase space; It is the total number of sampling points for a one-dimensional time-series vibration signal; It is the time index of the signal. Is The amplitude of the vibration signal collected at any time. Is The amplitude of the vibration signal collected at any time. It is the average value of a one-dimensional time-series vibration signal; The formula for calculating the threshold in the pseudo-nearest neighbor method is:

[0009] in, It is the ratio of pseudo-nearest neighbors; It is the current embedding dimension; yes In the dimensional embedding space, the first A state vector; yes In the dimensional embedding space, with The nearest neighbor with the smallest Euclidean distance; Is In the dimensional embedding space, with The same state vector; Is In the dimensional embedding space, with The same state vector; The first state vector of the phase space trajectory matrix for: ; The second state vector is shifted forward by one sampling interval from the starting point of time. for: .

[0010] Optionally, the phase space trajectory matrix is ​​converted into a two-dimensional grayscale image, including: Calculate the cosine similarity between any two row vectors in the phase space trajectory matrix to obtain a symmetric cosine similarity matrix; The cosine similarity matrix is ​​linearly normalized to map it to the pixel value range of a standard grayscale image, thus obtaining a two-dimensional grayscale image.

[0011] Optionally, the preset depth image denoising convolutional neural network includes a feature extraction layer, an intermediate layer, an output layer, and a residual network based on a residual mechanism; A two-dimensional grayscale image is input into a pre-defined depth image denoising convolutional neural network, which outputs a denoised feature image corresponding to the two-dimensional grayscale image, including: The feature extraction layer extracts features from the two-dimensional grayscale image to obtain an initial feature map; The intermediate layer alternately performs convolution, batch normalization, and activation operations on the initial feature map to obtain the mapping result; The output layer predicts the noise distribution map based on the mapping results; The residual network subtracts the noise distribution map from the two-dimensional grayscale image to obtain the denoised feature image corresponding to the two-dimensional grayscale image.

[0012] Optionally, the preset deep convolutional neural network includes multiple densely connected blocks, transition layers connecting the densely connected blocks, adaptive average pooling layers, and fully connected layers; The denoised feature image is input into a pre-defined deep convolutional neural network for feature extraction and classification, and the output is the fault state category corresponding to the rolling bearing to be diagnosed, including: Dense connectable blocks employ a cross-layer cascaded feature transfer mechanism to process the denoised feature image and obtain a stitched feature map; where the input of the l-th composite layer is the stitching of the feature maps output by all layers preceding the l-th composite layer. The transition layer downsamples and compresses the spliced ​​feature map output by the previous dense connection block to obtain a compressed feature map; The spliced ​​feature map output from the last densely connected block is passed sequentially through an adaptive average pooling layer and a fully connected layer to output the probability of each fault category, thus obtaining the probability distribution. Based on the probability distribution, the category with the highest probability is determined as the fault condition category of the rolling bearing to be diagnosed.

[0013] Optionally, the delay time is an integer within the range of [35, 45], and the embedding dimension is an integer within the range of [40, 50].

[0014] Secondly, embodiments of this application provide a bearing fault detection system, the system comprising: The vibration signal acquisition module is used to acquire one-dimensional time-series vibration signals of the rolling bearing under diagnosis during operation. The matrix transformation module is used to reconstruct a one-dimensional time-series vibration signal into a phase space trajectory matrix that characterizes the dynamic trajectory of the system; and to convert the phase space trajectory matrix into a two-dimensional grayscale image. The denoising module is used to input a two-dimensional grayscale image into a pre-defined depth image denoising convolutional neural network and output the denoised feature image corresponding to the two-dimensional grayscale image. The fault status category output module is used to input the denoised feature image into a pre-defined depth convolutional neural network for feature extraction and classification, and output the fault status category corresponding to the rolling bearing to be diagnosed.

[0015] Thirdly, embodiments of this application provide a computer storage medium storing multiple instructions adapted for loading and execution of the above-described method steps by a processor.

[0016] Fourthly, embodiments of this application provide an electronic device that may include: a processor and a memory; wherein the memory stores a computer program adapted to be loaded by the processor and executed by the above-described method steps.

[0017] The technical solutions provided in this application embodiment may include the following beneficial effects: In this embodiment, on the one hand, phase space reconstruction is used to map the one-dimensional time-series signal to a high-dimensional space, generating a state matrix that can characterize the complete dynamic trajectory of the system, which is then mapped to a grayscale image. In the generated two-dimensional grayscale image, the ordered dynamic features characterizing health or different fault modes are manifested as regular textures, stripes, or patch structures, while background noise is manifested as random, discrete pixels. This transformation makes the fault features and noise have a clear physical boundary visually, thus providing the subsequent neural network with visual input features rich in physical meaning and with far superior discriminative power compared to traditional methods. On the other hand, the preset deep image denoising convolutional neural network can accurately extract noise from textured feature images. Even under extremely low signal-to-noise ratio conditions such as -6dB, it can still effectively recover clear fault feature images, ensuring the accuracy of subsequent classification. Its noise resistance and robustness significantly surpass existing mainstream models. Furthermore, the preset deep convolutional classification neural network's dense connection mechanism ensures that the weak fault features retained after denoising can be repeatedly accessed and reused in the deep layers of the network, avoiding gradient degradation problems. It enhances the network's ability to extract and distinguish subtle texture differences, enabling the model to fully utilize the enhanced fault information and achieve highly reliable fault state determination.

[0018] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0019] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0020] Figure 1 This is a schematic flowchart of a bearing fault detection method provided in an embodiment of this application; Figure 2 This is a schematic diagram of the processing procedure of a pre-defined depth image denoising convolutional neural network provided in an embodiment of this application; Figure 3 This is a schematic diagram of a preset depth convolutional neural network processing procedure provided in an embodiment of this application; Figure 4 This is a schematic diagram of a bearing fault detection process provided in an embodiment of this application; Figure 5 This is a schematic diagram of the structure of a bearing fault detection system provided in an embodiment of this application; Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0021] The following description and accompanying drawings fully illustrate specific embodiments of this application to enable those skilled in the art to practice them.

[0022] It should be understood that the described embodiments are merely some, not all, of the embodiments in this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort are within the scope of protection of this application.

[0023] In the following description, when referring to the accompanying drawings, the same numbers in different drawings denote the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of systems and methods consistent with some aspects of this application as detailed in the appended claims.

[0024] In the description of this application, it should be understood that the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances. Furthermore, in the description of this application, unless otherwise stated, "multiple" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship.

[0025] This application provides a bearing fault detection method, system, medium, and electronic device to address the problems existing in the aforementioned related technologies. In the embodiments of this application, on the one hand, phase space reconstruction is used to map a one-dimensional time-series signal to a high-dimensional space, generating a state matrix that can characterize the complete dynamic trajectory of the system, which is then mapped into a grayscale image. In the generated two-dimensional grayscale image, ordered dynamic features characterizing health or different fault modes are manifested as regular textures, stripes, or patch structures, while background noise is manifested as random, discrete pixels. This transformation makes the fault features and noise have a clear physical boundary visually, thus providing the subsequent neural network with visual input features rich in physical meaning and with far superior discriminative power compared to traditional methods. On the other hand, a pre-defined deep image denoising convolutional neural network can accurately extract noise from textured feature images. Even under extremely low signal-to-noise ratio conditions such as -6dB, it can still effectively recover clear fault feature images, ensuring the accuracy of subsequent classification. Its noise resistance and robustness significantly surpass existing mainstream models. On the other hand, the pre-defined dense connection mechanism of the deep convolutional neural network ensures that the subtle fault features retained after denoising can be repeatedly accessed and reused in the deeper layers of the network, avoiding gradient degradation. This enhances the network's ability to extract and distinguish subtle texture differences, enabling the model to fully utilize the enhanced fault information and achieve highly reliable fault state determination. A detailed description using exemplary embodiments follows.

[0026] The following will be combined with the appendix Figure 1 -Appendix Figure 4 This application provides a detailed description of the bearing fault detection method provided in its embodiments. This method can be implemented using a computer program and can run on a bearing fault detection system based on the von Neumann architecture. The computer program can be integrated into the application or run as a standalone utility application.

[0027] Please see Figure 1 This is a flowchart illustrating a bearing fault detection method provided in an embodiment of this application. Figure 1 As shown, the method in this application embodiment may include the following steps: S101, Acquire one-dimensional time-series vibration signal of the rolling bearing to be diagnosed under operating conditions; Among them, the rolling bearing to be diagnosed refers to a rolling bearing (including components such as the inner ring, outer ring, rolling elements, and cage) that is in operation and requires fault detection and condition assessment. The operating state is the real-time dynamic process of the bearing under actual working conditions, including load, speed, and temperature. A one-dimensional time-series vibration signal refers to a set of discrete vibration data sampled sequentially along a single physical dimension (such as radial acceleration).

[0028] In one possible implementation, when performing bearing fault detection, it is necessary to first acquire a one-dimensional time-series vibration signal of the rolling bearing to be diagnosed under operating conditions.

[0029] S102 reconstructs the one-dimensional time-series vibration signal into a phase space trajectory matrix characterizing the system's dynamic trajectory; then converts the phase space trajectory matrix into a two-dimensional grayscale image. In some embodiments of this application, the specific process of reconstructing a one-dimensional time-series vibration signal into a phase space trajectory matrix representing the system's dynamic trajectory includes: obtaining the total number of sampling points and amplitude of the one-dimensional time-series vibration signal; calculating the autocorrelation coefficient of the signal at each delay time based on a preset autocorrelation function, the total number of sampling points, and the amplitude; taking the delay time when the autocorrelation coefficient first drops to zero or close to zero as the optimal delay time; using the pseudo-nearest neighbor method to calculate the embedding dimension required to map the one-dimensional signal to a high-dimensional phase space; and reconstructing the phase space of the one-dimensional time-series vibration signal through time delay and embedding dimension to generate a phase space trajectory matrix representing the system's dynamic trajectory.

[0030] Specifically, the delay time is an integer within the range of [35, 45], and the embedding dimension is an integer within the range of [40, 50].

[0031] Specifically, the predefined autocorrelation function expression is as follows:

[0032] in, The signal has a delay time of The autocorrelation coefficient at time; It is a time delay, representing the time interval between one state component and the next state component when reconstructing the phase space; It is the total number of sampling points for a one-dimensional time-series vibration signal; It is the time index of the signal. Is The amplitude of the vibration signal collected at any time. Is The amplitude of the vibration signal collected at any time. It is the average value of a one-dimensional time-series vibration signal.

[0033] Specifically, the formula for calculating the threshold in the pseudo-nearest neighbor method is as follows:

[0034] in, It is the ratio of pseudo-nearest neighbors; It is the current embedding dimension; yes In the dimensional embedding space, the first A state vector; yes In the dimensional embedding space, with The nearest neighbor with the smallest Euclidean distance; Is In the dimensional embedding space, with The same state vector; Is In the dimensional embedding space, with The same state vector.

[0035] Specifically, the first state vector of the phase space trajectory matrix for: ; The second state vector is shifted forward by one sampling interval from the starting point of time. for: .

[0036] It should be noted that the preferred range for time delay is set to an integer within the interval [35, 45] (41 being optimal), and the preferred range for embedding dimension is set to an integer within the interval [40, 50] (46 being optimal).

[0037] In some embodiments of this application, the specific process of converting a phase space trajectory matrix into a two-dimensional grayscale image includes: calculating the cosine similarity between any two row vectors in the phase space trajectory matrix to obtain a symmetrical cosine similarity matrix; and performing linear normalization on the cosine similarity matrix to map the cosine similarity matrix to the pixel value range of a standard grayscale image to obtain a two-dimensional grayscale image.

[0038] Specifically, the expression for calculating cosine similarity is:

[0039] in, It is the first in the phase space trajectory matrix and Cosine similarity between them It is the phase space trajectory matrix The row vectors It is the phase space trajectory matrix The Row vectors.

[0040] Specifically, the formula for converting a two-dimensional grayscale image is:

[0041] in, It is the final generated two-dimensional grayscale image. In the middle, located in the first , No. The pixel grayscale values ​​of the column. It is a cosine similarity matrix The Middle line, number Column elements, It is a cosine similarity matrix The maximum value of all elements in the set. It is a cosine similarity matrix The minimum value of all elements in the set. It is a rounding function.

[0042] It should be noted that, in addition to using phase space reconstruction combined with cosine similarity matrix, equivalent time-to-image methods such as Markov transition field (MTF), Gram angle field (GAF), or recurrence plot (RP) can also be used.

[0043] S103: Input the two-dimensional grayscale image into the preset depth image denoising convolutional neural network and output the denoised feature image corresponding to the two-dimensional grayscale image. The preset depth image denoising convolutional neural network includes a feature extraction layer, an intermediate layer, an output layer, and a residual network based on a residual mechanism.

[0044] In some embodiments of this application, the specific process of inputting a two-dimensional grayscale image into a pre-defined depth image denoising convolutional neural network and outputting a denoised feature image corresponding to the two-dimensional grayscale image includes: a feature extraction layer extracting features from the two-dimensional grayscale image to obtain an initial feature map; an intermediate layer alternately performing convolution, batch normalization, and activation operations on the initial feature map to obtain a mapping result; an output layer predicting a noise distribution map based on the mapping result; and a residual network subtracting the noise distribution map from the two-dimensional grayscale image to obtain the denoised feature image corresponding to the two-dimensional grayscale image.

[0045] Specifically, the expression for the feature extraction layer is:

[0046] in, It is a two-dimensional grayscale image. and These represent the kernel weights and biases of this layer, respectively. This represents the convolution operation. It is a linear rectified activation function. This is the initial feature map extracted.

[0047] Specifically, the expression for the intermediate layer is:

[0048] Here, BN(-) represents batch normalization, which is used to accelerate network training and improve stability. The intermediate layers jointly learn a complex mapping from noisy features to noisy features. It is the first The deep feature map output by the layer activation function. It is the first The convolution kernel weight matrix of the convolutional layer, It is the first Feature maps output by the layer It is the first The bias vector of each convolutional layer;

[0049] in, This is a noise distribution map. It is the first The convolution kernel weight matrix of the convolutional layer, It is the first The bias vector of each convolutional layer It is the first Feature maps output by the layer;

[0050] in, It is a two-dimensional grayscale image. It is the denoised feature image corresponding to the two-dimensional grayscale image.

[0051] It should be noted that the first layer is the feature extraction layer. Layers 2 through 16 are intermediate layers. The last layer is the output layer.

[0052] It should be noted that, in addition to using the DnCNN network, the preset depth image denoising convolutional neural network can also use FFDNet (Fast and Flexible Denoising Network) or a denoising model based on generative adversarial networks (such as CycleGAN) to achieve noise removal from two-dimensional feature images.

[0053] For example Figure 2 As shown, Figure 2 This is a schematic diagram of the processing procedure of a pre-defined depth image denoising convolutional neural network provided in this application. First, the noisy grayscale image (i.e., a two-dimensional grayscale image) is processed by convolution + ReLU activation function to extract primary features. Then, it enters a multi-layered repeated convolution + batch normalization (BN) + ReLU activation function module to gradually refine the feature representation in the deep network and obtain the predicted noise distribution map. At the same time, the noisy grayscale image and the predicted noise distribution map are subtracted through the bottom jump connection (residual path) to obtain the denoised feature image corresponding to the two-dimensional grayscale image.

[0054] S104: Input the denoised feature image into a pre-defined depth convolutional neural network for feature extraction and classification, and output the fault state category corresponding to the rolling bearing to be diagnosed.

[0055] The preset deep convolutional neural network includes multiple densely connected blocks, transition layers connecting the densely connected blocks, adaptive average pooling layers, and fully connected layers.

[0056] In some embodiments of this application, the specific process of inputting the denoised feature image into a preset depth convolutional neural network for feature extraction and classification, and outputting the fault state category corresponding to the rolling bearing to be diagnosed, includes: the dense connection block uses a cross-layer cascaded feature transfer mechanism to process the denoised feature image to obtain a spliced ​​feature map; wherein, the input of the l-th composite layer is the splicing of the feature maps output by all layers before the l-th composite layer; the transition layer downsamples and compresses the spliced ​​feature map output by the previous dense connection block to obtain a compressed feature map; the spliced ​​feature map output by the last dense connection block is passed sequentially through an adaptive average pooling layer and a fully connected layer to output the probability of each fault category, thus obtaining a probability distribution; according to the probability distribution, the category with the highest probability is determined as the fault state category of the rolling bearing to be diagnosed.

[0057] Specifically, the operation mapping function for the l-th composite layer is:

[0058] in, It is the first Operations of a composite layer It is a convolution operation performed using a 3x3 kernel. It is a linear rectified activation function. It is a batch normalization operation; The output of the concatenated feature map is defined as:

[0059] in, In a densely connected block, the first The feature map output after calculation of each composite layer It is the first Operational mapping functions for each composite layer, It's a splicing operation. It is the initial input , and the first within the dense connecting block layer to the first All feature maps output by the layer.

[0060] It should be noted that the transition layer achieves dimensionality reduction through 1x1 convolution and 2x2 average pooling.

[0061] Specifically, the operations of the transition layer can be represented as:

[0062] in, It is the output feature map after processing by the transition layer. It is a convolution operation performed using a 1x1 size convolution kernel. It is a 2x2 average pooling operation.

[0063] It should be noted that the preferred range for the number of dense connection blocks is 10 to 60 layers, and the preferred range for the network channel growth rate is [24, 32] (30 is optimal).

[0064] For example Figure 3 As shown, Figure 3 This is a schematic diagram of the pre-defined depth convolutional neural network processing procedure provided in this application. The denoised feature image first extracts initial features through a convolutional layer, then reduces the dimensionality and retains key features through a max pooling layer, and then enters dense block 1. Next, it passes through transition layer 1 (containing convolution and average pooling) to compress the feature dimension and connect to the next stage. Then, it enters dense block 2 to further deepen feature extraction, and then passes through transition layer 2 (containing convolution and average pooling) to adjust the feature structure. Finally, it enters dense block 3 to complete the final feature aggregation. Finally, the aggregated features are input into the classification module to output the bearing fault state category.

[0065] It should be noted that, in addition to DenseNet, other deep convolutional networks with residual connections or feature reuse mechanisms, such as ResNet (Residual Network) or MobileNet, can also be used to achieve the same purpose of extracting and classifying the features of the denoised image.

[0066] For example Figure 4 As shown, Figure 4This is a schematic diagram of a bearing fault detection process provided in this application. First, in the data preprocessing stage, Gaussian white noise is added to the original vibration signal to simulate strong noise interference in a real industrial environment, providing a data foundation for subsequent noise resistance verification. Then, in the phase space reconstruction stage, a phase space trajectory matrix is ​​constructed using the original vibration signal and noisy data, and it is mapped to a two-dimensional grayscale image through cosine similarity, realizing the transformation from one-dimensional time-series signal to structured visual features, thereby preserving the dynamic evolution characteristics of the vibration signal. Next, in the data denoising module, the generated grayscale image is input into a preset deep image denoising convolutional neural network (DnCNN), which accurately predicts and removes noise with the help of its residual learning mechanism, and outputs the denoised feature image. Finally, in the bearing fault classification stage, the denoised feature image is input into a deep convolutional classification neural network (such as DenseNet), which extracts weak fault features and completes classification through a multi-scale feature reuse mechanism, and finally outputs fault states such as "Category 1" to "Category 4", realizing a closed loop of the entire process from noisy signal to accurate diagnosis.

[0067] In this embodiment, on the one hand, phase space reconstruction is used to map the one-dimensional time-series signal to a high-dimensional space, generating a state matrix that can characterize the complete dynamic trajectory of the system, which is then mapped to a grayscale image. In the generated two-dimensional grayscale image, the ordered dynamic features characterizing health or different fault modes are manifested as regular textures, stripes, or patch structures, while background noise is manifested as random, discrete pixels. This transformation makes the fault features and noise have a clear physical boundary visually, thus providing the subsequent neural network with visual input features rich in physical meaning and with far superior discriminative power compared to traditional methods. On the other hand, the preset deep image denoising convolutional neural network can accurately extract noise from textured feature images. Even under extremely low signal-to-noise ratio conditions such as -6dB, it can still effectively recover clear fault feature images, ensuring the accuracy of subsequent classification. Its noise resistance and robustness significantly surpass existing mainstream models. Furthermore, the preset deep convolutional classification neural network's dense connection mechanism ensures that the weak fault features retained after denoising can be repeatedly accessed and reused in the deep layers of the network, avoiding gradient degradation problems. It enhances the network's ability to extract and distinguish subtle texture differences, enabling the model to fully utilize the enhanced fault information and achieve highly reliable fault state determination.

[0068] The following are system embodiments of this application, which can be used to execute the method embodiments of this application. For details not disclosed in the system embodiments of this application, please refer to the method embodiments of this application.

[0069] Please see Figure 5This illustration shows a schematic diagram of a bearing fault detection system provided in an exemplary embodiment of this application. The bearing fault detection system can be implemented as all or part of an electronic device through software, hardware, or a combination of both. System 1 includes a vibration signal acquisition module 10, a matrix conversion module 20, a noise reduction module 30, and a fault status category output module 40.

[0070] Vibration signal acquisition module 10 is used to acquire one-dimensional time-series vibration signals of the rolling bearing to be diagnosed under operating conditions. Matrix transformation module 20 is used to reconstruct a one-dimensional time-series vibration signal into a phase space trajectory matrix characterizing the dynamic trajectory of the system; and to convert the phase space trajectory matrix into a two-dimensional grayscale image. The denoising module 30 is used to input a two-dimensional grayscale image into a pre-defined depth image denoising convolutional neural network and output a denoised feature image corresponding to the two-dimensional grayscale image. The fault status category output module 40 is used to input the noise-reduced feature image into a pre-defined depth convolutional neural network for feature extraction and classification, and output the fault status category corresponding to the rolling bearing to be diagnosed.

[0071] It should be noted that the bearing fault detection system provided in the above embodiments is only illustrated by the division of the above functional modules when executing the bearing fault detection method. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the equipment can be divided into different functional modules to complete all or part of the functions described above. In addition, the bearing fault detection system and the bearing fault detection method embodiments provided in the above embodiments belong to the same concept, and the implementation process is detailed in the method embodiments, which will not be repeated here.

[0072] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0073] In this embodiment, on the one hand, phase space reconstruction is used to map the one-dimensional time-series signal to a high-dimensional space, generating a state matrix that can characterize the complete dynamic trajectory of the system, which is then mapped to a grayscale image. In the generated two-dimensional grayscale image, the ordered dynamic features characterizing health or different fault modes are manifested as regular textures, stripes, or patch structures, while background noise is manifested as random, discrete pixels. This transformation makes the fault features and noise have a clear physical boundary visually, thus providing the subsequent neural network with visual input features rich in physical meaning and with far superior discriminative power compared to traditional methods. On the other hand, the preset deep image denoising convolutional neural network can accurately extract noise from textured feature images. Even under extremely low signal-to-noise ratio conditions such as -6dB, it can still effectively recover clear fault feature images, ensuring the accuracy of subsequent classification. Its noise resistance and robustness significantly surpass existing mainstream models. Furthermore, the preset deep convolutional classification neural network's dense connection mechanism ensures that the weak fault features retained after denoising can be repeatedly accessed and reused in the deep layers of the network, avoiding gradient degradation problems. It enhances the network's ability to extract and distinguish subtle texture differences, enabling the model to fully utilize the enhanced fault information and achieve highly reliable fault state determination.

[0074] This application also provides a computer-readable medium having program instructions stored thereon, which, when executed by a processor, implement the bearing fault detection method provided in the above-described method embodiments.

[0075] This application also provides a computer program product containing instructions that, when run on a computer, causes the computer to execute the bearing fault detection methods of the various method embodiments described above.

[0076] Please see Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 6 As shown, the electronic device 1000 may include: at least one processor 1001, at least one network interface 1004, a user interface 1003, a memory 1005, and at least one communication bus 1002.

[0077] The communication bus 1002 is used to realize the connection and communication between these components.

[0078] The user interface 1003 may include a display screen and a camera. Optionally, the user interface 1003 may also include a standard wired interface and a wireless interface.

[0079] The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0080] The processor 1001 may include one or more processing cores. The processor 1001 connects to various parts within the electronic device 1000 using various interfaces and lines. It executes various functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in the memory 1005, and by calling data stored in the memory 1005. Optionally, the processor 1001 may be implemented using at least one hardware form selected from Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), and Programmable Logic Array (PLA). The processor 1001 may integrate one or more of the following: a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), and a modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content to be displayed on the screen; and the modem handles wireless communication. It is understood that the modem may also be implemented as a separate chip, without being integrated into the processor 1001.

[0081] The memory 1005 may include random access memory (RAM) or read-only memory. Optionally, the memory 1005 may include a non-transitory computer-readable storage medium. The memory 1005 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 1005 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 1005 may also be at least one storage system located remotely from the aforementioned processor 1001. Figure 6 As shown, the memory 1005, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and a bearing fault detection application.

[0082] exist Figure 6In the illustrated electronic device 1000, the user interface 1003 is mainly used to provide an input interface for the user and to acquire user input data; while the processor 1001 can be used to call the bearing fault detection application stored in the memory 1005 and specifically perform the following operations: Acquire one-dimensional time-series vibration signals of the rolling bearing to be diagnosed under operating conditions; The one-dimensional time-series vibration signal is reconstructed into a phase space trajectory matrix representing the dynamic trajectory of the system; the phase space trajectory matrix is ​​then converted into a two-dimensional grayscale image. A two-dimensional grayscale image is input into a pre-defined depth image denoising convolutional neural network, and the output is a denoised feature image corresponding to the two-dimensional grayscale image. The denoised feature image is input into a pre-defined depth convolutional neural network for feature extraction and classification, and the fault state category corresponding to the rolling bearing to be diagnosed is output.

[0083] In one embodiment, when processor 1001 reconstructs a one-dimensional time-series vibration signal into a phase space trajectory matrix characterizing the system's dynamic trajectory, it specifically performs the following operations: Obtain the total number of sampling points and amplitude of a one-dimensional time-series vibration signal; Based on the preset autocorrelation function, the total number of sampling points, and the amplitude, calculate the autocorrelation coefficient of the signal at each delay time. The optimal delay time is defined as the time when the autocorrelation coefficient first drops to zero or close to zero. The pseudo-nearest neighbor method is used to calculate the embedding dimension required to map a one-dimensional signal to a high-dimensional phase space; By adjusting the time delay and embedding dimension, the phase space of a one-dimensional time-series vibration signal is reconstructed to generate a phase space trajectory matrix that characterizes the dynamic trajectory of the system.

[0084] In one embodiment, when the processor 1001 performs the operation of converting the phase space trajectory matrix into a two-dimensional grayscale image, the following specific operations are performed: Calculate the cosine similarity between any two row vectors in the phase space trajectory matrix to obtain a symmetric cosine similarity matrix; The cosine similarity matrix is ​​linearly normalized to map it to the pixel value range of a standard grayscale image, thus obtaining a two-dimensional grayscale image.

[0085] In one embodiment, when the processor 1001 executes the operation of inputting a two-dimensional grayscale image into a pre-defined depth image denoising convolutional neural network and outputting a denoised feature image corresponding to the two-dimensional grayscale image, it specifically performs the following operations: The feature extraction layer extracts features from the two-dimensional grayscale image to obtain an initial feature map; The intermediate layer alternately performs convolution, batch normalization, and activation operations on the initial feature map to obtain the mapping result; The output layer predicts the noise distribution map based on the mapping results; The residual network subtracts the noise distribution map from the two-dimensional grayscale image to obtain the denoised feature image corresponding to the two-dimensional grayscale image.

[0086] In one embodiment, when the processor 1001 performs feature extraction and classification by inputting the denoised feature image into a preset depth convolutional neural network and outputting the fault state category corresponding to the rolling bearing to be diagnosed, it specifically performs the following operations: Dense connectable blocks employ a cross-layer cascaded feature transfer mechanism to process the denoised feature image and obtain a stitched feature map; where the input of the l-th composite layer is the stitching of the feature maps output by all layers preceding the l-th composite layer. The transition layer downsamples and compresses the spliced ​​feature map output by the previous dense connection block to obtain a compressed feature map; The spliced ​​feature map output from the last densely connected block is passed sequentially through an adaptive average pooling layer and a fully connected layer to output the probability of each fault category, thus obtaining the probability distribution. Based on the probability distribution, the category with the highest probability is determined as the fault condition category of the rolling bearing to be diagnosed.

[0087] In this embodiment, on the one hand, phase space reconstruction is used to map the one-dimensional time-series signal to a high-dimensional space, generating a state matrix that can characterize the complete dynamic trajectory of the system, which is then mapped to a grayscale image. In the generated two-dimensional grayscale image, the ordered dynamic features characterizing health or different fault modes are manifested as regular textures, stripes, or patch structures, while background noise is manifested as random, discrete pixels. This transformation makes the fault features and noise have a clear physical boundary visually, thus providing the subsequent neural network with visual input features rich in physical meaning and with far superior discriminative power compared to traditional methods. On the other hand, the preset deep image denoising convolutional neural network can accurately extract noise from textured feature images. Even under extremely low signal-to-noise ratio conditions such as -6dB, it can still effectively recover clear fault feature images, ensuring the accuracy of subsequent classification. Its noise resistance and robustness significantly surpass existing mainstream models. Furthermore, the preset deep convolutional classification neural network's dense connection mechanism ensures that the weak fault features retained after denoising can be repeatedly accessed and reused in the deep layers of the network, avoiding gradient degradation problems. It enhances the network's ability to extract and distinguish subtle texture differences, enabling the model to fully utilize the enhanced fault information and achieve highly reliable fault state determination.

[0088] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The bearing fault detection program can be stored in a computer-readable storage medium, and when executed, the program can include the processes of the embodiments of the above methods. The storage medium for the bearing fault detection program can be a magnetic disk, optical disk, read-only memory, or random access memory, etc.

[0089] The above-disclosed embodiments are merely preferred embodiments of this application and should not be construed as limiting the scope of this application. Therefore, any equivalent variations made in accordance with the claims of this application shall still fall within the scope of this application.

Claims

1. A bearing fault detection method characterized by, The method includes: Acquire one-dimensional time-series vibration signals of the rolling bearing to be diagnosed under operating conditions; The one-dimensional time-series vibration signal is reconstructed into a phase space trajectory matrix characterizing the dynamic trajectory of the system; the phase space trajectory matrix is ​​then converted into a two-dimensional grayscale image. The two-dimensional grayscale image is input into a pre-defined depth image denoising convolutional neural network, and the denoised feature image corresponding to the two-dimensional grayscale image is output. The denoised feature image is input into a pre-defined deep convolutional neural network for feature extraction and classification, and the fault state category corresponding to the rolling bearing to be diagnosed is output.

2. The method of claim 1, wherein, The process of reconstructing the one-dimensional time-series vibration signal into a phase space trajectory matrix characterizing the system's dynamic trajectory includes: Obtain the total number of sampling points and amplitude of the one-dimensional time-series vibration signal; The autocorrelation coefficient of the signal at each delay time is calculated based on the preset autocorrelation function, the total number of sampling points, and the amplitude. The delay time at which the autocorrelation coefficient first drops to zero or close to zero is taken as the optimal delay time; The pseudo-nearest neighbor method is used to calculate the embedding dimension required to map a one-dimensional signal to a high-dimensional phase space; By using the time delay and the embedding dimension, the phase space of the one-dimensional time-series vibration signal is reconstructed to generate a phase space trajectory matrix characterizing the dynamic trajectory of the system.

3. The method of claim 2, wherein, The function expression of the preset autocorrelation function is: wherein, is the autocorrelation coefficient of the signal at a delay time ; is the time delay, representing the time interval between one state component to the next state component when reconstructing the phase space; is the total number of sampling points of the one-dimensional time series vibration signal; is the time index of the signal, is the amplitude of the vibration signal collected at ; is the amplitude of the vibration signal collected at ; is the mean value of the one-dimensional time series vibration signal; The formula for calculating the threshold of the pseudo-nearest neighbor method is as follows: in, It is the ratio of pseudo-nearest neighbors; It is the current embedding dimension; yes In the dimensional embedding space, the first A state vector; yes In the dimensional embedding space, with The nearest neighbor with the smallest Euclidean distance; Is In the dimensional embedding space, with The same state vector; Is In the dimensional embedding space, with The same state vector; The first state vector of the phase space trajectory matrix for: ； The second state vector is shifted forward by one sampling interval from the starting point of time. for: 。 4. The method according to claim 1, characterized in that, The step of converting the phase space trajectory matrix into a two-dimensional grayscale image includes: Calculate the cosine similarity between any two row vectors in the phase space trajectory matrix to obtain a symmetric cosine similarity matrix; The cosine similarity matrix is ​​linearly normalized to map it to the pixel value range of a standard grayscale image, thus obtaining a two-dimensional grayscale image.

5. The method according to claim 1, characterized in that, The preset depth image denoising convolutional neural network includes a feature extraction layer, an intermediate layer, an output layer, and a residual network based on a residual mechanism. The step of inputting the two-dimensional grayscale image into a pre-defined depth image denoising convolutional neural network and outputting the denoised feature image corresponding to the two-dimensional grayscale image includes: The feature extraction layer extracts features from the two-dimensional grayscale image to obtain an initial feature map; The intermediate layer alternately performs convolution, batch normalization, and activation operations on the initial feature map to obtain the mapping result; The output layer predicts the noise distribution map based on the mapping result; The residual network subtracts the noise distribution map from the two-dimensional grayscale image to obtain the denoised feature image corresponding to the two-dimensional grayscale image.

6. The method according to claim 1, characterized in that, The preset deep convolutional neural network includes multiple densely connected blocks, a transition layer connecting the densely connected blocks, an adaptive average pooling layer, and a fully connected layer; The step involves inputting the denoised feature image into a preset depth convolutional neural network for feature extraction and classification, and outputting the fault state category corresponding to the rolling bearing to be diagnosed, including: The dense connection block uses a cross-layer cascaded feature transfer mechanism to process the denoised feature image to obtain a stitched feature map; wherein, the input of the l-th composite layer is the stitching of the feature maps output by all layers before the l-th composite layer; The transition layer downsamples and compresses the spliced ​​feature map output by the previous dense connection block to obtain a compressed feature map; The spliced ​​feature map output from the last densely connected block is passed sequentially through the adaptive average pooling layer and the fully connected layer to output the probability of each fault category, thus obtaining the probability distribution. Based on the probability distribution, the category with the highest probability is determined as the fault state category of the rolling bearing to be diagnosed.

7. The method according to claim 2, characterized in that, The delay time is an integer within the range of [35, 45], and the embedding dimension is an integer within the range of [40, 50].

8. A bearing fault detection system, characterized in that, The system includes: The vibration signal acquisition module is used to acquire one-dimensional time-series vibration signals of the rolling bearing under diagnosis during operation. The matrix transformation module is used to reconstruct the one-dimensional time-series vibration signal into a phase space trajectory matrix characterizing the system's dynamic trajectory; and to convert the phase space trajectory matrix into a two-dimensional grayscale image. The denoising module is used to input the two-dimensional grayscale image into a preset depth image denoising convolutional neural network and output the denoised feature image corresponding to the two-dimensional grayscale image. The fault status category output module is used to input the noise-reduced feature image into a preset depth convolutional neural network for feature extraction and classification, and output the fault status category corresponding to the rolling bearing to be diagnosed.

9. A computer storage medium, characterized in that, The computer storage medium stores a plurality of instructions adapted for loading by a processor and executing the method as described in any one of claims 1-7.

10. An electronic device, characterized in that, include: A processor and a memory; wherein the memory stores a computer program adapted to be loaded by the processor and executed as described in any one of claims 1-7.

Images