Bearing health state prediction method, device, equipment, medium and product

Abstract

The invention discloses a bearing health state prediction method comprising the following steps: collecting a bearing vibration signal of a to-be-predicted bearing, and carrying out denoising processing on the bearing vibration signal; splitting the denoised bearing vibration signal into a trend signal and a residual signal; performing feature extraction on the residual signal by adopting time causal convolution to obtain a short-term mutation feature; adopting an attention enhancement mechanism to carry out multiple rounds of circulation enhancement on the short-term mutation feature to obtain an enhanced residual feature; and fusing the trend signal and the enhanced residual feature, and outputting a final bearing vibration signal prediction value. According to the method, high-precision prediction can be carried out on the long-term trend and the short-term sudden change of the bearing vibration signal, and the precision and the reliability of bearing health state prediction are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of bearing health condition technology, and in particular to a method, apparatus, equipment, medium and product for predicting bearing health condition. Background Technology

[0002] Belt conveyors are core equipment in coal mine production. From the underground mining face to the surface washing and coal storage yards, coal transportation mainly relies on belt conveyors. Idler rollers are the main load-bearing equipment of belt conveyors, and bearings, as the core components of idler rollers, bear the function of supporting rotation. More than 70% of idler roller failures are caused by bearing failures. By predicting the vibration of idler roller bearing operating signals, it is possible to effectively judge their operating status trend, thereby realizing the prediction of the health status of belt conveyor idler roller bearings. At present, the prediction methods for idler roller bearings are mainly divided into physical model-based methods and data-driven methods. Among them, the data-driven deep learning method has become the most suitable bearing condition prediction method for the field because it can automatically extract features and has a strong fitting ability for nonlinear and non-stationary signals, and has strong robustness.

[0003] Due to the high background noise in the working environment of belt conveyor rollers, the original vibration signal collected is affected by the noise, making it difficult to identify early damage characteristics of the bearings. Wavelet denoising is a relatively convenient and commonly used method. However, due to the influence of the selection of relevant parameters such as wavelet basis, decomposition level and threshold of the denoising algorithm, the denoising effect on industrial background noise is not ideal.

[0004] Current deep learning prediction models still have some limitations: First, they perform poorly when faced with signals in the field with strong background noise or signals that change abruptly due to various operating conditions, making it difficult to effectively predict field data; second, they cannot take into account both long-term trend changes and short-term abrupt changes. Traditional deep learning methods and recurrent neural network structures based on sliding windows are limited by the window size, resulting in limited ability to predict the long-term operating trend of equipment. While encoder-decoder structure models represented by Transformer have strong long-term trend prediction capabilities, they usually ignore short-term abnormal changes in signals, making it difficult to respond in a timely manner to sudden bearing deterioration caused by unexpected situations.

[0005] Therefore, there is an urgent need for a more accurate method for predicting bearing health status that can effectively avoid on-site noise interference and take into account both long-term trend prediction and short-term mutation prediction. Summary of the Invention

[0006] This invention provides a bearing health condition prediction method that can effectively avoid on-site noise interference and make high-precision predictions of both long-term trends and short-term abrupt changes in bearing vibration signals, significantly improving the accuracy and reliability of bearing health condition prediction.

[0007] In a first aspect, embodiments of the present invention provide a method for predicting the health status of a bearing, comprising: Collect the bearing vibration signal of the bearing to be predicted, and perform noise reduction processing on the bearing vibration signal; The denoised bearing vibration signal is split into a trend signal and a residual signal; The residual signal is used to extract features to obtain short-term mutation features; An attention enhancement mechanism is used to perform multiple rounds of iterative enhancement on the short-term mutation features to obtain enhanced residual features. The trend signal and the enhanced residual features are fused to output the final predicted value of the bearing vibration signal.

[0008] Furthermore, the noise reduction processing of the bearing vibration signal includes: Obtain the bearing's own parameters, and calculate the bearing's fault characteristic frequency based on the own parameters; Based on the bearing vibration signal and a pre-set set of adaptive wavelet basis functions, the optimal wavelet basis and the optimal number of decomposition layers are automatically selected based on the maximization of information metric criteria. Based on the optimal wavelet basis, the optimal number of decomposition layers, and the fault characteristic frequency, the lifting wavelet algorithm is used to filter out noise in the bearing vibration signal to obtain the denoised bearing vibration signal.

[0009] Furthermore, the step of automatically selecting the optimal wavelet basis and the optimal number of decomposition levels based on the bearing vibration signal and a pre-set set of adaptive wavelet basis functions, according to the maximization of information metric criterion, includes: Calculate the mutual information value, kurtosis, and Shannon entropy based on the bearing vibration signal; Based on the mutual information value, the kurtosis, and the Shannon entropy, calculate the maximum information measure using different wavelet bases and decomposition levels; Based on the maximized information measure, the optimal wavelet basis and the optimal number of decomposition layers are determined.

[0010] Furthermore, the step of using temporal causal convolution to extract features from the residual signal to obtain short-term mutation features includes: Perform a first-time causal convolution operation on the residual signal to extract local abrupt change features in the residual signal; Perform weight normalization processing on the convolution kernel weights corresponding to the local mutation features; The feature signal after weight normalization is input into the linear rectified activation function to filter out invalid signals; Perform Dropout regularization on the activated feature signals; A second dilation convolution operation is performed on the regularized feature signal to expand the receptive field of feature extraction by increasing the dilation rate, thereby obtaining short-term mutation features.

[0011] Furthermore, the attention enhancement mechanism is employed to perform multiple rounds of cyclical enhancement on the short-term mutation features to obtain enhanced residual features, including: Step 1: Perform max pooling on the short-term mutation features, input the max pooled feature signal into a one-dimensional convolutional layer for feature refinement, and then filter out invalid features through a linear rectified activation function to obtain the initially focused pulse features; Step 2: Perform average pooling and max pooling operations on the initially focused pulse features to obtain global trend features and local peak features. After concatenating the two pooling features according to the channel dimension, input them into a one-dimensional convolutional layer for feature fusion. Then, use the Sigmoid activation function to map the fused features to attention weights in the [0,1] interval to obtain the pulse feature weight matrix. Step 3: Multiply the initially focused pulse features and the pulse feature weight matrix element by element to output the attention-enhanced short-term mutation feature signal; Step 4: Perform weight normalization, linear rectified activation function and Dropout regularization on the short-term mutation feature signal in sequence; Step 5: Repeat steps 1 to 4 above until the preset number of iterations is reached to obtain the enhanced residual features.

[0012] Furthermore, the process of fusing the trend signal and the enhanced residual features to output the final predicted bearing vibration signal includes: The trend signal and the enhanced residual feature are element-wise added to obtain the fused feature; The fused features are input into a multilayer perceptron module, and the fused features are nonlinearly transformed and deeply integrated through a multilayer fully connected network to output the final predicted value of the bearing vibration signal.

[0013] Secondly, embodiments of the present invention provide a bearing health condition prediction device, comprising: The noise reduction module is used to acquire the bearing vibration signal of the bearing to be predicted and to perform noise reduction processing on the bearing vibration signal. The sequence decomposition module is used to separate the denoised bearing vibration signal into a trend signal and a residual signal. The feature extraction module is used to extract features from the residual signal using temporal causal convolution to obtain short-term mutation features; The attention enhancement module is used to perform multiple rounds of cyclical enhancement on the short-term mutation features using an attention enhancement mechanism to obtain enhanced residual features. The feature fusion and prediction output module is used to fuse the trend signal and the enhanced residual features to output the final predicted value of the bearing vibration signal.

[0014] Thirdly, embodiments of the present invention provide an electronic device, comprising: Memory, used to store computer programs; A processor for executing the computer program; Wherein, when the processor executes the computer program, it implements the bearing health status prediction method described in any of the first aspects above.

[0015] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program that, when executed, implements the bearing health status prediction method described in any of the first aspects above.

[0016] Fifthly, embodiments of the present invention provide a computer program product, including computer instructions, which, when executed by a processor, implement the bearing health status prediction method described in any of the first aspects above.

[0017] Compared with existing technologies, the bearing health status prediction method provided in this invention has the following advantages: It collects the bearing vibration signal of the bearing to be predicted and performs denoising processing on the bearing vibration signal; it decomposes the denoised bearing vibration signal into a trend signal and a residual signal; it extracts features from the residual signal using temporal causal convolution to obtain short-term abrupt change features; it uses an attention enhancement mechanism to perform multiple rounds of iterative enhancement on the short-term abrupt change features to obtain enhanced residual features; and it fuses the trend signal and the enhanced residual features to output the final predicted value of the bearing vibration signal. This invention can accurately predict both the long-term trend and short-term abrupt changes of bearing vibration signals, significantly improving the accuracy and reliability of bearing health status prediction. Attached Figure Description

[0018] To more clearly illustrate the technical features of the embodiments of the present invention, the drawings used in the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating a bearing health status prediction method provided in an embodiment of the present invention; Figure 2 This is an actual effect diagram of a bearing health status prediction method provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a bearing health status prediction device provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0020] 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 embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, and the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0022] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein is for the purpose of describing embodiments of the invention only and is not intended to limit the invention.

[0023] In a first aspect, embodiments of the present invention provide a method for predicting the health status of bearings, see [link to previous document]. Figure 1 This is a flowchart illustrating an embodiment of a bearing health condition prediction method provided by the present invention.

[0024] like Figure 1 As shown, the method includes the following steps: S1: Collect the bearing vibration signal of the bearing to be predicted, and perform noise reduction processing on the bearing vibration signal; S2: Decompose the denoised bearing vibration signal into a trend signal and a residual signal; S3: Use temporal causal convolution to extract features from the residual signal to obtain short-term mutation features; S4: The short-term mutation feature is enhanced in multiple rounds using an attention enhancement mechanism to obtain the enhanced residual feature; S5: The trend signal and the enhanced residual features are fused together to output the final predicted value of the bearing vibration signal.

[0025] In practice, bearings, as the core transmission components of rotating machinery, contain a wealth of health status information in their vibration signals. By accurately predicting the vibration signals, the deterioration trend of the bearing can be predicted in advance. This step uses sensors to collect the vibration signals of the bearing to be predicted during operation, and uses adaptive noise reduction processing to remove noise interference, thereby obtaining high-quality bearing vibration signals, laying the foundation for subsequent processing.

[0026] Furthermore, a sequence decomposition algorithm is used to split the denoised vibration signal into a trend signal and a residual signal, thereby separating the features at different time scales. The trend signal contains long-term degradation trend information of the data, while the residual signal contains short-term abrupt changes and background noise. The trend signal is directly retained, while the residual signal needs to be further processed in subsequent steps. The sequence decomposition algorithm can be wavelet decomposition, VMD variational mode decomposition, or any other decomposition algorithm, without specific limitations.

[0027] For the residual signal, a temporal causal convolution method is used for feature extraction to avoid interference from future signals on the current feature extraction, thus obtaining short-term mutation features. An attention mechanism is then used to perform multiple rounds of iterative enhancement on the short-term mutation features, ultimately obtaining enhanced residual features. For example, SAM (Spatial Attention Mechanism), CBAM (Convolutional Block Attention Module), or other attention mechanisms can be used, without specific limitations here. Finally, through feature fusion and deep integration with MLP (Multilayer Perceptron), the final predicted value of the bearing vibration signal is output.

[0028] For example, see Figure 2The figure shows the actual effect of the bearing health status prediction method of the present invention. The horizontal axis represents continuous sampling points in time, and the vertical axis represents the vibration intensity at each moment. As can be seen from the figure, the deviation between the predicted value and the actual value is small. In the stable stage of 0~250, the vibration amplitude fluctuates between 1.5~2.0g, which is the periodic vibration generated by the normal rotation of the bearing. From 250 to 300, the vibration amplitude rises continuously from about 1.8g to 3.0g, reflecting the long-term increase in vibration intensity caused by the aggravation of bearing wear. The peak value of 3.0g that appears at 300 reflects the instantaneous vibration of the bearing failure. It can be seen that the bearing health status prediction method of the present invention can accurately fit the real vibration signal and accurately predict periodic trends, long-term upward trends and short-term signal mutations, and can provide reliable prediction data for bearing health status monitoring and early fault warning.

[0029] In summary, this invention decomposes the trend signal and residual signal, processes the two signals in parallel, and can handle long-term trend information and short-term abrupt change information respectively. It uses an attention mechanism to extract key features from the residual signal, and then fuses the trend signal and the processed residual signal in the output stage. This invention can simultaneously have the ability to predict long-term trends and the ability to efficiently predict short-term abrupt changes, significantly improving the accuracy and reliability of bearing health status prediction.

[0030] In one optional implementation, the noise reduction processing of the bearing vibration signal includes: Obtain the bearing's own parameters, and calculate the bearing's fault characteristic frequency based on the own parameters; Based on the bearing vibration signal and a pre-set set of adaptive wavelet basis functions, the optimal wavelet basis and the optimal number of decomposition layers are automatically selected based on the maximization of information metric criteria. Based on the optimal wavelet basis, the optimal number of decomposition layers, and the fault characteristic frequency, the lifting wavelet algorithm is used to filter out noise in the bearing vibration signal to obtain the denoised bearing vibration signal.

[0031] Specifically, the bearing's own parameters are obtained, including but not limited to core parameters such as bearing model, inner diameter, outer diameter, number of rolling elements, rolling element diameter, contact angle, and rated speed. Based on the above parameters, the existing mature fault characteristic frequency calculation formula is used to calculate the bearing's fault characteristic frequency.

[0032] The pre-defined adaptive wavelet basis function set contains a variety of commonly used wavelet bases, all of which have good time-domain localization characteristics and are suitable for processing non-stationary signals such as bearing vibration. Based on the bearing vibration signal and the adaptive wavelet basis function set, the optimal wavelet base and the optimal number of decomposition layers are automatically selected based on the maximization of information measure criterion.

[0033] Based on the optimal wavelet basis and the optimal number of decomposition levels, the original bearing vibration signal is decomposed into multiple scales to obtain low-frequency approximate components (corresponding to the steady vibration characteristics of the bearing) and high-frequency detail components (corresponding to clutter and fault mutation characteristics) at different scales. Combined with the calculated fault characteristic frequency, the high-frequency detail components are thresholded and filtered to retain the high-frequency components (fault mutation characteristics) corresponding to the fault characteristic frequency and remove the high-frequency components (clutter) that do not have the fault characteristic frequency. The filtered high-frequency detail components and low-frequency approximate components are then subjected to inverse lifting wavelet transform to reconstruct the denoised bearing vibration signal.

[0034] This embodiment can adaptively select the number of wavelet decomposition layers and the type of wavelet basis according to the equipment conditions. Moreover, the wavelet basis selected using the proposed maximum information metric can eliminate on-site noise more accurately than the wavelet basis selected by other indicators, and can achieve the highest signal-to-noise ratio and the lowest error value.

[0035] In one optional implementation, the step of automatically selecting the optimal wavelet basis and the optimal number of decomposition levels based on the bearing vibration signal and a pre-set set of adaptive wavelet basis functions, according to the maximization of information metric criterion, includes: Calculate the mutual information value, kurtosis, and Shannon entropy based on the bearing vibration signal; Based on the mutual information value, the kurtosis, and the Shannon entropy, calculate the maximum information measure using different wavelet bases and decomposition levels; Based on the maximized information measure, the optimal wavelet basis and the optimal number of decomposition layers are determined.

[0036] Specifically, the preset set of wavelet basis functions and the range of decomposition levels are traversed. For each combination of wavelet basis function and decomposition level, the acquired bearing vibration signal is subjected to preliminary denoising to obtain the corresponding denoised signal. Calculate the denoised signal The corresponding mutual information value, kurtosis, and Shannon entropy are substituted into the formula for maximizing the information measure to calculate the maximum information measure value. The formula for maximizing the information measure is as follows: ; in, To maximize information measurement, For mutual information value, As a reference signal, in this embodiment it can be a theoretical signal generated based on the characteristic frequency of bearing failure. For ravine, For Shannon entropy.

[0037] The combination of "wavelet basis - number of decomposition layers" that maximizes the information metric value is selected as the optimal wavelet basis and the optimal number of decomposition layers for this denoising process.

[0038] This embodiment calculates three dimensions of indices: mutual information, kurtosis, and Shannon entropy. By maximizing the information measure, it automatically selects the wavelet basis and decomposition level most suitable for the current bearing vibration signal, ensuring that the denoised signal retains fault characteristics (high mutual information, high kurtosis) to the greatest extent while eliminating noise (low entropy). In one optional implementation, the step of using temporal causal convolution to extract features from the residual signal to obtain short-term mutation features includes: Perform a first-time causal convolution operation on the residual signal to extract local abrupt change features in the residual signal; Perform weight normalization processing on the convolution kernel weights corresponding to the local mutation features; The feature signal after weight normalization is input into the linear rectified activation function to filter out invalid signals; Perform Dropout regularization on the activated feature signals; A second dilation convolution operation is performed on the regularized feature signal to expand the receptive field of feature extraction by increasing the dilation rate, thereby obtaining short-term mutation features.

[0039] Specifically, the first convolution operation is performed on the residual signal to extract local abrupt changes in the residual signal in a targeted manner, avoiding interference from future signals on the current feature extraction. This step is implemented using dilated causal convolution. The calculation formula for the first convolution operation is as follows: ; in, Where d is the kernel size and d is the dilation rate. These are the convolution kernel weights, which are parameters that can be adaptively updated during model training. At the current time point, The input residual signal, This is the output local mutation feature signal.

[0040] Through the above convolution operation, the scattered local mutation information in the residual signal is transformed into a continuous feature sequence, realizing the preliminary extraction and aggregation of local mutation features.

[0041] Furthermore, the convolution kernel weights In the early stages of model training, there may be amplitude imbalance issues, leading to model instability and slow convergence. Therefore, weight normalization is performed on the convolutional kernel weights to optimize the model training process by standardizing the weight distribution. The formula for weight normalization is as follows: ; in, This is the normalized weight vector. These are trainable scalar parameters. For unnormalized vector parameters, Unnormalized vector parameters The L2 norm.

[0042] By normalizing the weights, the convolutional kernel weights are kept within a reasonable range of amplitude, which not only accelerates the convergence speed of model training but also improves the stability of the training process, providing a reliable parameter basis for subsequent feature processing.

[0043] Furthermore, bearing vibration signals are typical nonlinear signals, and the correlation between local abrupt changes in the residual signal and the bearing fault state is also nonlinear. By using the Rectified Linear Activation Function (ReLU), the model's ability to fit nonlinear signals is enhanced, while invalid feature signals are filtered out, improving subsequent computational efficiency. The formula for calculating the ReLU activation function is: ; in, The core logic of this function is to retain only the valid feature components that are greater than 0 in the feature signal after weight normalization, and to directly set the invalid feature components that are less than or equal to 0 to 0.

[0044] Furthermore, during model training, if there is an over-reliance on some effective features extracted by the ReLU activation function, overfitting is likely to occur. Dropout regularization is needed to break the over-reliance in the feature extraction process and improve the model's generalization ability.

[0045] The second convolution operation is performed on the regularized feature signal. By increasing the dilation rate, the sampling interval of the convolution kernel on the feature signal is further expanded, and the sensing range covers a wider range of time nodes, thereby capturing short-term mutation features with a large span that were not covered by the first convolution.

[0046] This embodiment first extracts local subtle mutation features through a first-time causal convolution, and then expands the receptive field through a second dilated convolution to capture multi-scale mutation association features, which can achieve comprehensive and accurate extraction of short-term mutation features.

[0047] In one optional implementation, the attention enhancement mechanism is used to perform multiple rounds of cyclical enhancement on the short-term mutation features to obtain enhanced residual features, including: Step 1: Perform max pooling on the short-term mutation features, input the max pooled feature signal into a one-dimensional convolutional layer for feature refinement, and then filter out invalid features through a linear rectified activation function to obtain the initially focused pulse features; Step 2: Perform average pooling and max pooling operations on the initially focused pulse features to obtain global trend features and local peak features. After concatenating the two pooling features according to the channel dimension, input them into a one-dimensional convolutional layer for feature fusion. Then, use the Sigmoid activation function to map the fused features to attention weights in the [0,1] interval to obtain the pulse feature weight matrix. Step 3: Multiply the initially focused pulse features and the pulse feature weight matrix element by element to output the attention-enhanced short-term mutation feature signal; Step 4: Perform weight normalization, linear rectified activation function and Dropout regularization on the short-term mutation feature signal in sequence; Step 5: Repeat steps 1 to 4 above until the preset number of iterations is reached to obtain the enhanced residual features.

[0048] Specifically, step one performs max pooling on the input short-term mutation features to extract the maximum value within each local time window, highlighting the pulse peak features. The max-pooled features are then input into a one-dimensional convolutional layer, where the convolutional kernel fuses and refines the features around the peak to make the features more representative. Finally, the ReLU activation function filters out invalid features less than or equal to 0, resulting in preliminarily focused pulse features. This step can extract the most representative pulse peak information from the short-term mutation features, laying the foundation for subsequent attention enhancement.

[0049] Step 2 performs average pooling and max pooling operations on the initially focused pulse features to obtain global trend features and local peak features. The two pooling features are concatenated along the channel dimension and then input into a one-dimensional convolutional layer for feature fusion. The fused features are then mapped to the [0,1] interval through the Sigmoid activation function to obtain the pulse feature weight matrix. The higher the weight, the more important the pulse feature at the corresponding position.

[0050] Step 3 involves multiplying the initially focused pulse features and the pulse feature weight matrix element by element to enhance the key mutation features. Through attention enhancement, important pulse features are amplified, irrelevant noise is suppressed, and the enhanced short-term mutation feature signal is finally output.

[0051] To ensure the stability and generalization ability of the model training, step four sequentially performs weight normalization, linear rectified activation function and Dropout regularization on the short-term mutation feature signals to ensure that the feature quality is stable and reliable after each round of iteration, providing a good foundation for subsequent iterations.

[0052] To fully exploit the subtle abrupt changes in the residual signal, this invention employs a multi-round cyclic enhancement strategy, repeatedly executing steps one through four until a preset number of cycles is reached. Each cycle further focuses, enhances, and optimizes the enhanced features based on the previous cycle, continuously improving the expressive power of the features and ultimately obtaining the enhanced residual features.

[0053] This embodiment significantly improves the accuracy and reliability of bearing health status prediction through an attention enhancement mechanism and a multi-round cyclic reinforcement method.

[0054] In one optional implementation, fusing the trend signal and the enhanced residual features to output the final predicted bearing vibration signal includes: The trend signal and the enhanced residual feature are element-wise added to obtain the fused feature; The fused features are input into a multilayer perceptron module, and the fused features are nonlinearly transformed and deeply integrated through a multilayer fully connected network to output the final predicted value of the bearing vibration signal.

[0055] Specifically, the trend signal and the enhanced residual features are element-wise added to obtain the fused features. The fused features are then deeply nonlinearly integrated through a multilayer perceptron (MLP) module to uncover the complex correlations between the features and output accurate predicted values ​​of the bearing vibration signal.

[0056] This embodiment fully integrates the long-term characteristics of trend signals with the short-term abrupt change characteristics of residual features, and enhances the feature representation capability through deep nonlinear transformation, ultimately achieving high-precision bearing vibration signal prediction.

[0057] Secondly, embodiments of the present invention provide a bearing health condition prediction device, see [link to relevant documentation]. Figure 3 This is a schematic diagram of an embodiment of a bearing health condition prediction device provided by the present invention.

[0058] like Figure 3 As shown, the device includes: The noise reduction module 21 is used to collect the bearing vibration signal of the bearing to be predicted and to perform noise reduction processing on the bearing vibration signal. Sequence decomposition module 22 is used to decompose the denoised bearing vibration signal into a trend signal and a residual signal; Feature extraction module 23 is used to extract features from the residual signal using temporal causal convolution to obtain short-term mutation features; Attention enhancement module 24 is used to perform multiple rounds of cyclic enhancement on the short-term mutation features using an attention enhancement mechanism to obtain enhanced residual features; The feature fusion and prediction output module 25 is used to fuse the trend signal and the enhanced residual features to output the final bearing vibration signal prediction value.

[0059] In one optional implementation, the noise reduction processing of the bearing vibration signal includes: Obtain the bearing's own parameters, and calculate the bearing's fault characteristic frequency based on the own parameters; Based on the bearing vibration signal and a pre-set set of adaptive wavelet basis functions, the optimal wavelet basis and the optimal number of decomposition layers are automatically selected based on the maximization of information metric criteria. Based on the optimal wavelet basis, the optimal number of decomposition layers, and the fault characteristic frequency, the lifting wavelet algorithm is used to filter out noise in the bearing vibration signal to obtain the denoised bearing vibration signal.

[0060] In one optional implementation, the step of automatically selecting the optimal wavelet basis and the optimal number of decomposition levels based on the bearing vibration signal and a pre-set set of adaptive wavelet basis functions, according to the maximization of information metric criterion, includes: Calculate the mutual information value, kurtosis, and Shannon entropy based on the bearing vibration signal; Based on the mutual information value, the kurtosis, and the Shannon entropy, calculate the maximum information measure using different wavelet bases and decomposition levels; Based on the maximized information measure, the optimal wavelet basis and the optimal number of decomposition layers are determined.

[0061] In one optional implementation, the step of using temporal causal convolution to extract features from the residual signal to obtain short-term mutation features includes: Perform a first-time causal convolution operation on the residual signal to extract local abrupt change features in the residual signal; Perform weight normalization processing on the convolution kernel weights corresponding to the local mutation features; The feature signal after weight normalization is input into the linear rectified activation function to filter out invalid signals; Perform Dropout regularization on the activated feature signals; A second dilation convolution operation is performed on the regularized feature signal to expand the receptive field of feature extraction by increasing the dilation rate, thereby obtaining short-term mutation features.

[0062] In one optional implementation, the attention enhancement mechanism is used to perform multiple rounds of cyclical enhancement on the short-term mutation features to obtain enhanced residual features, including: Step 1: Perform max pooling on the short-term mutation features, input the max pooled feature signal into a one-dimensional convolutional layer for feature refinement, and then filter out invalid features through a linear rectified activation function to obtain the initially focused pulse features; Step 2: Perform average pooling and max pooling operations on the initially focused pulse features to obtain global trend features and local peak features. After concatenating the two pooling features according to the channel dimension, input them into a one-dimensional convolutional layer for feature fusion. Then, use the Sigmoid activation function to map the fused features to attention weights in the [0,1] interval to obtain the pulse feature weight matrix. Step 3: Multiply the initially focused pulse features and the pulse feature weight matrix element by element to output the attention-enhanced short-term mutation feature signal; Step 4: Perform weight normalization, linear rectified activation function and Dropout regularization on the short-term mutation feature signal in sequence; Step 5: Repeat steps 1 to 4 above until the preset number of iterations is reached to obtain the enhanced residual features.

[0063] In one optional implementation, fusing the trend signal and the enhanced residual features to output the final predicted bearing vibration signal includes: The trend signal and the enhanced residual feature are element-wise added to obtain the fused feature; The fused features are input into a multilayer perceptron module, and the fused features are nonlinearly transformed and deeply integrated through a multilayer fully connected network to output the final predicted value of the bearing vibration signal.

[0064] It should be noted that the bearing health condition prediction device provided in this embodiment of the invention is used to execute all the process steps of the bearing health condition prediction method in the above embodiment. The working principle and beneficial effect of the two are one-to-one, so they will not be described again.

[0065] Thirdly, embodiments of the present invention provide an electronic device, see [link to previous document]. Figure 4 The diagram shown is a structural schematic of an electronic device provided in an embodiment of the present invention.

[0066] like Figure 4 As shown, the device includes: Memory 31 is used to store computer programs; Processor 32 is used to execute the computer program; When the processor 32 executes the computer program, it implements the bearing health status prediction method as described in any of the above embodiments.

[0067] For example, the computer program may be divided into one or more modules / units, which are stored in the memory 31 and executed by the processor 32 to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0068] The processor 32 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0069] The memory 31 can be used to store the computer programs and / or modules. The processor 32 implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory 31 and calling the data stored in the memory 31. The memory 31 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory 31 may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital card (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0070] It should be noted that the aforementioned electronic devices include, but are not limited to, processors and memory, as will be understood by those skilled in the art. Figure 4 The structural diagram is merely an example of the electronic device described above and does not constitute a limitation on the electronic device. It may include more components than shown in the diagram, or combine certain components, or use different components.

[0071] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing a computer program that, when executed, implements the bearing health status prediction method described in any of the above embodiments.

[0072] It should be understood that the present invention can implement all or part of the processes in the above-described bearing health condition prediction method by instructing related hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the above-described bearing health condition prediction method. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0073] Fifthly, embodiments of this application also provide a computer program product, which is stored in a storage medium and executed by at least one processor to implement the bearing health condition prediction method described in any of the above embodiments.

[0074] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. It should be noted that, for those skilled in the art, several equivalent obvious modifications and / or equivalent substitutions can be made without departing from the technical principles of the present invention, and these obvious modifications and / or equivalent substitutions should also be considered within the scope of protection of the present invention.

Claims

1. A method for predicting the health status of bearings, characterized in that, include: Collect the bearing vibration signal of the bearing to be predicted, and perform noise reduction processing on the bearing vibration signal; The denoised bearing vibration signal is split into a trend signal and a residual signal; The residual signal is used to extract features to obtain short-term mutation features; An attention enhancement mechanism is used to perform multiple rounds of iterative enhancement on the short-term mutation features to obtain enhanced residual features. The trend signal and the enhanced residual features are fused to output the final predicted value of the bearing vibration signal.

2. The bearing health status prediction method as described in claim 1, characterized in that, The noise reduction process for the bearing vibration signal includes: Obtain the bearing's own parameters, and calculate the bearing's fault characteristic frequency based on the own parameters; Based on the bearing vibration signal and a pre-set set of adaptive wavelet basis functions, the optimal wavelet basis and the optimal number of decomposition layers are automatically selected based on the maximization of information metric criteria. Based on the optimal wavelet basis, the optimal number of decomposition layers, and the fault characteristic frequency, the lifting wavelet algorithm is used to filter out noise in the bearing vibration signal to obtain the denoised bearing vibration signal.

3. The bearing health status prediction method as described in claim 2, characterized in that, The step of automatically selecting the optimal wavelet basis and optimal decomposition level based on the bearing vibration signal and a pre-set set of adaptive wavelet basis functions, according to the maximization of information metric criterion, includes: Calculate the mutual information value, kurtosis, and Shannon entropy based on the bearing vibration signal; Based on the mutual information value, the kurtosis, and the Shannon entropy, calculate the maximum information measure using different wavelet bases and decomposition levels; Based on the maximized information measure, the optimal wavelet basis and the optimal number of decomposition layers are determined.

4. The bearing health status prediction method as described in claim 1, characterized in that, The step of using temporal causal convolution to extract features from the residual signal to obtain short-term mutation features includes: Perform a first-time causal convolution operation on the residual signal to extract local abrupt change features in the residual signal; Perform weight normalization processing on the convolution kernel weights corresponding to the local mutation features; The feature signal after weight normalization is input into the linear rectified activation function to filter out invalid signals; Perform Dropout regularization on the activated feature signals; A second dilation convolution operation is performed on the regularized feature signal to expand the receptive field of feature extraction by increasing the dilation rate, thereby obtaining short-term mutation features.

5. The bearing health status prediction method as described in claim 1, characterized in that, The attention enhancement mechanism is used to perform multiple rounds of cyclical enhancement on the short-term mutation features to obtain enhanced residual features, including: Step 1: Perform max pooling on the short-term mutation features, input the max pooled feature signal into a one-dimensional convolutional layer for feature refinement, and then filter out invalid features through a linear rectified activation function to obtain the initially focused pulse features; Step 2: Perform average pooling and max pooling operations on the initially focused pulse features to obtain global trend features and local peak features. After concatenating the two pooling features according to the channel dimension, input them into a one-dimensional convolutional layer for feature fusion. Then, use the Sigmoid activation function to map the fused features to attention weights in the [0,1] interval to obtain the pulse feature weight matrix. Step 3: Multiply the initially focused pulse features and the pulse feature weight matrix element by element to output the attention-enhanced short-term mutation feature signal; Step 4: Perform weight normalization, linear rectified activation function and Dropout regularization on the short-term mutation feature signal in sequence; Step 5: Repeat steps 1 to 4 above until the preset number of iterations is reached to obtain the enhanced residual features.

6. The bearing health condition prediction method as described in claim 1, characterized in that, The step of fusing the trend signal and the enhanced residual features to output the final predicted bearing vibration signal includes: The trend signal and the enhanced residual feature are element-wise added to obtain the fused feature; The fused features are input into a multilayer perceptron module, and the fused features are nonlinearly transformed and deeply integrated through a multilayer fully connected network to output the final predicted value of the bearing vibration signal.

7. A bearing health condition prediction device, characterized in that, include: The noise reduction module is used to acquire the bearing vibration signal of the bearing to be predicted and to perform noise reduction processing on the bearing vibration signal. The sequence decomposition module is used to separate the denoised bearing vibration signal into a trend signal and a residual signal. The feature extraction module is used to extract features from the residual signal using temporal causal convolution to obtain short-term mutation features; The attention enhancement module is used to perform multiple rounds of cyclical enhancement on the short-term mutation features using an attention enhancement mechanism to obtain enhanced residual features. The feature fusion and prediction output module is used to fuse the trend signal and the enhanced residual features to output the final predicted value of the bearing vibration signal.

8. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program; The processor executes the computer program to implement the bearing health status prediction method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed, implements the bearing health condition prediction method as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, It includes computer instructions that, when executed by a processor, implement the bearing health condition prediction method as described in any one of claims 1 to 6.

Images