Bearing anomaly detection method and system based on time sequence correlation difference

Abstract

The invention discloses a bearing anomaly detection method and system based on time sequence correlation difference, and the method comprises the steps: constructing a convolutional neural network composed of a plurality of CNN modules, and extracting the multi-dimensional features of a bearing in a time domain and a frequency domain from different scales through each CNN module; respectively extracting adjacent time sequence features and non-adjacent time sequence features in the bearing sliding window through an adjacent attention module and a non-adjacent attention module; a bearing time sequence correlation difference mechanism based on symmetric KL divergence is adopted to calculate correlation differences between adjacent time sequence features and non-adjacent time sequence features; inputting the association difference into a convolutional neural network for training, increasing the association difference between the normal state and the abnormal state of the bearing, and enhancing the distinction degree of feature representation; and mapping the multi-dimensional features of the bearing into probability distribution by using the trained convolutional neural network, thereby realizing the detection of the abnormal state of the bearing. According to the method, the accuracy of the bearing anomaly detection task is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of fault detection in the intelligent manufacturing industry, and in particular to a bearing anomaly detection method and system based on time series correlation differences. Background Art

[0002] As the core component of rotating machinery, the operating status of bearings directly affects the reliability and safety of the equipment. Traditional bearing anomaly detection methods include methods based on vibration analysis and methods based on sound analysis. Among them, the method based on vibration analysis diagnoses the bearing status by monitoring the vibration signal of the bearing during operation, uses an accelerometer to capture the vibration data of the bearing, and uses signal processing technology to identify the characteristic frequency of the bearing fault. However, in the process of capturing data, it is difficult to capture the temporal correlation between multiple physical fields, resulting in a high rate of missed detection of early weak anomalies. The method based on sound analysis detects anomalies by analyzing the sound signals generated by the bearing during operation. This method can use sound sensors such as microphones to capture data, and use sound processing technologies such as Mel frequency cepstral coefficients to analyze the abnormal characteristics of the signal, but it is affected by environmental noise and has low accuracy.

[0003] In recent years, bearing anomaly detection methods based on deep learning have become a research hotspot. By training deep neural network models, abnormal patterns in bearing data can be identified. Bearing anomaly detection methods based on deep learning can process large amounts of data and learn complex feature representations from them, thereby improving the accuracy and reliability of bearing fault detection. However, in the process of bearing detection, existing methods only learn the timing characteristics of the bearings, and cannot accurately obtain the adjacent timing characteristics and non-adjacent timing characteristics of the bearings, resulting in low accuracy in bearing anomaly detection.

[0004] Therefore, how to improve the accuracy of bearing abnormality detection is a technical problem that needs to be solved urgently in this field. Summary of the invention

[0005] In order to solve the above technical problems, this application proposes the following technical solutions: In a first aspect, an embodiment of the present application provides a bearing abnormality detection method based on time series correlation difference, comprising: Construct a convolutional neural network composed of multiple CNN modules, and use each CNN module to extract multi-dimensional features of the bearing in the time domain and frequency domain from different scales; The adjacent time series features and non-adjacent time series features in the bearing sliding window are extracted respectively through the adjacent attention module and the non-adjacent attention module; The correlation difference between the adjacent time series features and the non-adjacent time series features is calculated using a bearing time series correlation difference mechanism based on symmetric KL divergence; The correlation difference is input into a convolutional neural network for training, so as to increase the correlation difference between the normal state and the abnormal state of the bearing and enhance the discrimination of the feature representation; The trained convolutional neural network is used to map the multi-dimensional features of the bearing into probability distribution to detect abnormal conditions of the bearing.

[0006] In a possible implementation, extracting adjacent time series features and non-adjacent time series features in the bearing sliding window respectively by using an adjacent attention module and a non-adjacent attention module includes: The input bearing feature is evenly divided into the first sub-feature and the second sub-feature according to the channel dimension; When the first sub-feature and the second sub-feature pass through the neighboring attention module and the non-neighboring attention module respectively, the bearing neighboring time series features and non-neighboring time series features are extracted.

[0007] In a possible implementation, when the first sub-feature passes through the neighboring attention module, the bearing neighboring temporal features are extracted, including: The first sub-feature is input into the neighboring attention module, and the query-key relationship between the current moment and multiple historical moments is defined, and the cross attention module is used to calculate the correlation features between the current moment and the neighboring moments step by step; the calculation formula of the cross attention module is: In the formula, , , Represent the query vector, key vector, and value vector respectively. express , The dimension size of The associated features of the multiple historical moments are fused to form a comprehensive adjacent temporal feature representation.

[0008] In a possible implementation, by defining a query-key relationship between the current moment and multiple historical moments, the cross-attention module is used to calculate the correlation features between the current moment and the adjacent moments step by step, including: Define Tn as the current moment, Tn-1 as the previous moment, Tn-2 as the previous two moments, and so on until Tn-t is the previous t moments; The current moment Tn is used as the query vector, and the previous moment Tn-1 is used as the key vector and value vector, which are input into the cross attention module to obtain the associated features of the moment Tn relative to the previous moment; The current moment Tn is used as the query vector, and the previous two moments Tn-1 and Tn-2 are used as the key vector and value vector, which are input into the cross attention module to obtain the correlation features of the moment Tn relative to the previous two moments; The current moment Tn is used as the query vector, and the previous t moments Tn-1, Tn-2, …, Tn-t are used as key vectors and value vectors, which are input into the cross-attention module to obtain the associated features of moment Tn relative to the previous t moments.

[0009] In a possible implementation, when the second sub-feature passes through the non-adjacent attention module, the bearing non-adjacent time series feature is extracted, including: The non-neighboring attention module performs sliding window sampling on the second sub-feature; Use sparse sampling to obtain a sparse representation of the input bearing features; The sparse features of the bearing are converted into query volume, key vector and value vector through the first linear layer, and the correlation between the query volume and the key vector is calculated by matrix multiplication; Apply the Softmax activation function to convert the correlation into a probability distribution vector; The probability distribution vector is matrix multiplied with the value vector, and converted through the second linear layer to obtain the non-adjacent time series features of the current moment of the bearing and the entire bearing sequence.

[0010] In a possible implementation, the calculation of the correlation difference between the adjacent time series features and the non-adjacent time series features by using the bearing time series correlation difference mechanism based on symmetric KL divergence includes: KL divergence is used to calculate the difference between adjacent time series features and non-adjacent time series features. The calculation formula is: in, and Represent the probability distribution of adjacent time series features and non-adjacent time series features respectively; KL divergence is used to calculate the difference between non-adjacent time series features and adjacent time series features. The calculation formula is: The symmetric KL divergence is used to calculate the point-by-point correlation difference between adjacent time series features and non-adjacent time series features. The calculation formula is: in, Priori association Associated with sequence The KL divergence calculated between two discrete distributions for each row, yes Compared with multi-layer prior association Associated with sequence The point-by-point correlation difference.

[0011] In a possible implementation, the calculation formula for increasing the correlation difference between the normal state and the abnormal state of the bearing is: in, It represents the correlation difference between the prior correlation P and the sequence correlation S in the same sliding window; is a hyperparameter; Represents the value of the loss function.

[0012] In a second aspect, an embodiment of the present application provides a bearing abnormality detection system based on time series correlation difference, including: The feature extraction module is a convolutional neural network composed of multiple CNN modules. Each CNN module is used to extract multi-dimensional features of the bearing in the time domain and frequency domain from different scales. A time series feature learning module, used for extracting adjacent time series features and non-adjacent time series features in the bearing sliding window respectively through an adjacent attention module and a non-adjacent attention module; A time series correlation difference calculation module, used for calculating the correlation difference between the adjacent time series features and the non-adjacent time series features based on the bearing time series correlation difference mechanism of symmetric KL divergence; A network training module, used for inputting the correlation difference into a convolutional neural network for training, increasing the correlation difference between the normal state and the abnormal state of the bearing, and enhancing the discrimination of the feature representation; The anomaly detection module uses a trained convolutional neural network to map the multi-dimensional features of the bearing into a probability distribution to detect abnormal conditions of the bearing.

[0013] Compared with the prior art, the beneficial effects of this application are: This application designs neighboring attention modules and non-neighboring attention modules at the level of extracting neighboring time series features of bearings and the level of extracting non-neighboring time series features of bearings. Among them, the neighboring attention module enables the network to better identify potential fault modes by focusing on the features of adjacent time points, effectively captures the dynamic changes of bearings in a local range, and enhances the model's sensitivity and accuracy to changes in bearing states. The non-neighboring attention module can efficiently capture the dynamic changes of bearings on a larger time scale, which is particularly important for fault features that are not obvious in the short term but gradually appear in long-term operation.

[0014] In the bearing timing correlation difference processing level, this application designs a bearing timing correlation difference mechanism based on symmetric KL divergence, and calculates adjacent timing features and non-adjacent timing features in the same sliding window based on the bearing timing correlation difference mechanism based on symmetric KL divergence. During training, the distinguishing ability of feature representation is improved by increasing the correlation difference between normal state and abnormal state, so as to effectively distinguish normal points and abnormal points of the bearing. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A flow chart of a bearing abnormality detection method based on time series correlation difference provided in an embodiment of the present application; Figure 2 The overall block diagram of the convolutional neural network provided in the embodiment of the present application; Figure 3 A schematic diagram of the bearing Transformer structure provided in an embodiment of the present application; Figure 4 A schematic diagram of a bearing structure provided in an embodiment of the present application; Figure 5 A schematic diagram of the structure of a neighboring attention module provided in an embodiment of the present application; Figure 6 A schematic diagram of the structure of a non-adjacent attention module provided in an embodiment of the present application; Figure 7 A schematic diagram of a bearing abnormality detection system based on time series correlation differences provided in an embodiment of the present application. DETAILED DESCRIPTION

[0016] The present solution is described below in conjunction with the accompanying drawings and specific implementation methods.

[0017] Figure 1 A flow chart of a bearing abnormality detection method based on time series correlation difference provided in an embodiment of the present application is shown in FIG. Figure 1 In this embodiment, a bearing abnormality detection method based on time series correlation difference includes: S101, construct a convolutional neural network composed of multiple CNN modules, and use each CNN module to extract multi-dimensional features of the bearing in the time domain and frequency domain from different scales.

[0018] In the bearing anomaly detection model, the extraction of time domain features is crucial because the richness of these features directly affects the accuracy of anomaly detection. Figure 2 As shown in the figure, this embodiment constructs a convolutional neural network composed of five CNN blocks. Each CNN module consists of 1D convolution, bearing transformer, ReLU, batch normalization and maximum pooling, which are used to capture the characteristics of bearings at different frequencies and time domains, extract multi-dimensional features of bearings at different scales, obtain multi-scale representation of bearing features, and enhance the model's sensitivity to bearing state changes. Among them, the core of the CNN module is the bearing transformer, and its architecture is as follows: Figure 3As shown in Figure 1. The design of the bearing transformer consists of two key parts: the neighboring attention module and the non-neighboring attention module. These two modules are responsible for extracting the neighboring time series features and non-neighboring time series features in the bearing data respectively. The neighboring attention module focuses on capturing the relationship between the time points and the local patterns and short-term dependencies in the bearing signal. These features are usually closely related to the immediate state of the bearing. Its architecture is shown in Figure 1. Figure 5 The non-neighboring attention module aims to identify the relationship between time points and long-term dependencies and global patterns in the bearing signal, which is crucial for understanding the long-term operating status and potential degradation trends of the bearing. Its architecture is shown in Figure 6 As shown. By combining these two types of attention modules, the bearing Transformer module is able to comprehensively extract the time domain features of the bearing. 1D convolution is used to extract bearing features, and nonlinearity is introduced through the ReLU activation function to help the model learn complex patterns. Batch normalization is used to stabilize the learning process, speed up convergence, and help reduce internal covariate shifts and improve the generalization ability of the model. The maximum pooling layer is used to reduce the spatial dimension of the features while retaining the most important information. This pooling strategy not only reduces the number of parameters, but also improves the robustness of the model to small changes and disturbances.

[0019] S102, extracting adjacent time series features and non-adjacent time series features in the bearing sliding window respectively through a neighboring attention module and a non-neighboring attention module.

[0020] In this embodiment, in the CNN block, a bearing transformer is designed to obtain the adjacent time series features and non-adjacent time series features of the bearing. First, the input bearing features are evenly divided into two sub-features according to the channel dimension. The first sub-feature is extracted through the adjacent attention module to extract the adjacent time series features of the bearing, and the second sub-feature is extracted through the non-adjacent attention module to extract the non-adjacent time series features of the bearing. When the adjacent time series features are fused with the non-adjacent time series features, the adjacent time series features obtained through the adjacent attention module and the non-adjacent time series features obtained through the non-adjacent attention module are spliced ​​in the channel dimension, and the spliced ​​features are restored to the input video frame feature dimension using a linear layer. A feedforward network is used to extract useful feature representations from the spliced ​​features to obtain the fusion results of the adjacent time series features and the non-adjacent time series features, which are used for subsequent network learning of bearing features.

[0021] Among them, when the first sub-feature is extracted from the bearing neighboring time series features through the neighboring attention module, the method of defining the time relationship, feature extraction, and feature fusion is used to extract the bearing neighboring time series features. When defining the time relationship, let Tn be the current time, Tn-1 be the previous time, Tn-2 be the previous two times, and so on until Tn-t is the previous t times. When extracting features, the current time Tn is used as the query vector Q, the previous time Tn-1 is used as the key vector K and the value vector V, and is input into the cross attention module to obtain the associated features of the Tn time relative to its previous time. Extend this process, use Tn as the query vector Q, the previous two times Tn-1 and Tn-2 as the key vector K and the value vector V, and input them into the cross attention module to obtain the associated features of the Tn time relative to its previous two times. Continue this process until Tn is used as the query vector Q, and the previous t moments Tn-1, Tn-2, ..., Tn-t are used as the key vector K and the value vector V, which are input into the cross-attention module to obtain the associated features of the moment Tn relative to its previous t moments. When fusion of features, all the features obtained above are input into the cross-attention module, which automatically fuses the above associated features to form a comprehensive neighboring temporal feature representation.

[0022] When the second sub-feature is extracted through the non-neighboring attention module, the input feature is first sampled in a sliding window, and a sparse sampling method is used to achieve a sparse representation of the input bearing feature. The linear layer converts the sparse features of the bearing into a query vector (Q), a key vector (K), and a value vector (V). Then, matrix multiplication is used to calculate the association between the query vector features and the key vector features. In order to adaptively adjust these associations, the attention weight scaling method is used to introduce a learnable scaling factor, allowing the model to dynamically adjust the strength of the association between features according to the specific needs of the task. Subsequently, the Softmax activation function is applied to convert the association into a probability distribution vector. The obtained probability distribution vector is matrix multiplied with the value vector V, and then transformed through another linear layer to obtain the non-neighboring time series features of the bearing Tn moment and the bearing sequence in the sliding window.

[0023] All the features obtained above are input into the cross attention module, which will automatically fuse the associated features of the Tn moment and its adjacent moments to form a comprehensive adjacent temporal feature representation. In this embodiment, the calculation formula of the cross attention module is: In the formula, , , Represent the query vector, key vector, and value vector respectively. express , The dimension size of . S103, using a bearing time series correlation difference mechanism based on symmetric KL divergence to calculate the correlation difference between adjacent time series features and non-adjacent time series features.

[0024] See also Figure 4 In this embodiment, a bearing time series correlation difference mechanism based on symmetric KL divergence is used to calculate the correlation difference between the adjacent time series features and the non-adjacent time series features of the bearing, which not only considers the positive correlation difference, but also considers the reverse correlation difference, thereby providing a comprehensive correlation difference metric. First, the difference between the adjacent time series features and the non-adjacent time series features is calculated using the KL divergence, and the calculation formula is: in, and Represent the probability distribution of adjacent time series features and non-adjacent time series features respectively; KL divergence is used to calculate the difference between non-adjacent time series features and adjacent time series features. The calculation formula is: In order to solve the problem of asymmetric KL divergence, symmetric KL divergence is used to calculate the point-by-point correlation difference between adjacent time series features and non-adjacent time series features, ensuring that the feature difference at each time point can be accurately measured, thereby providing rich time series information for bearing anomaly detection. The calculation formula is: in, Priori association Associated with sequence The KL divergence calculated between two discrete distributions for each row, yes Compared with multi-layer prior association Associated with sequence Given the point-by-point association differences, abnormal time points will present larger AssDis(P,S;X) than normal time points, which makes AssDis inherently distinguishable.

[0025] S104, inputting the correlation difference into a convolutional neural network for training, increasing the correlation difference between the normal state and the abnormal state of the bearing, and enhancing the discrimination of the feature representation.

[0026] In this embodiment, by increasing the correlation difference between the normal state and the abnormal state, the distinguishing ability of the feature representation is improved to effectively distinguish the normal points and abnormal points of the bearing. Specifically, when the correlation difference is increased, the correlation difference between the normal time point of the bearing and the adjacent time point does not change much from the correlation difference between the normal time point of the bearing and the non-adjacent time point, that is, the correlation difference is small, because the normal point of the bearing usually follows the general mode of bearing operation. In the time series, the normal point does not change much from the adjacent point, and the change with the non-adjacent point is not significant, which shows that the normal point shows a high consistency in the sequence. However, the correlation difference between the abnormal time point of the bearing and the adjacent time point changes greatly from the correlation difference between the abnormal time point of the bearing and the non-adjacent time point, that is, the correlation difference is large, which reflects the isolation and inconsistency of the abnormal point in the time series. Therefore, increasing the correlation difference can better realize the detection of abnormal points, so that the normal points and abnormal points of the bearing can be effectively distinguished. During the training process, the calculation formula for increasing the correlation difference between the normal state and the abnormal state of the bearing is: in, It represents the correlation difference between the prior correlation P and the sequence correlation S in the same sliding window; is a hyperparameter used to adjust the weight of the associated difference; Represents the value of the loss function, the network adjusts The accuracy of bearing abnormality detection can be improved.

[0027] S105, using the trained convolutional neural network to map the multi-dimensional features of the bearing into a probability distribution, so as to detect the abnormal state of the bearing.

[0028] Corresponding to the bearing abnormality detection method based on time series correlation difference provided in the above embodiment, the present application also provides an embodiment of a bearing abnormality detection system based on time series correlation difference.

[0029] See also Figure 7 , a bearing abnormality detection system 20 based on time series correlation difference provided in an embodiment of the present application includes: The feature extraction module 201 is a convolutional neural network composed of multiple CNN modules, and each CNN module is used to extract multi-dimensional features of the bearing in the time domain and frequency domain from different scales.

[0030] The time series feature learning module 202 is used to extract the neighboring time series features and non-neighboring time series features in the bearing sliding window respectively through the neighboring attention module and the non-neighboring attention module.

[0031] The time series correlation difference calculation module 203 is used to calculate the correlation difference between the adjacent time series features and the non-adjacent time series features based on the bearing time series correlation difference mechanism of the symmetric KL divergence.

[0032] The network training module 204 is used to input the correlation difference into the convolutional neural network for training, increase the correlation difference between the normal state and the abnormal state of the bearing, and enhance the discrimination of the feature representation.

[0033] The abnormality detection module 205 uses the trained convolutional neural network to map the multi-dimensional features of the bearing into a probability distribution to detect the abnormal state of the bearing.

[0034] In the embodiments of the present application, "multiple" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent the existence of A alone, the existence of A and B at the same time, and the existence of B alone. A and B can be singular or plural. The character " / " generally indicates that the associated objects before and after are in an "or" relationship.

[0035] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0036] The above is only a specific implementation of the present application. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. The protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A bearing abnormality detection method based on time series correlation difference, characterized in that: include: Construct a convolutional neural network composed of multiple CNN modules, and use each CNN module to extract multi-dimensional features of the bearing in the time domain and frequency domain from different scales; The adjacent time series features and non-adjacent time series features in the bearing sliding window are extracted respectively through the adjacent attention module and the non-adjacent attention module; The correlation difference between the adjacent time series features and the non-adjacent time series features is calculated using a bearing time series correlation difference mechanism based on symmetric KL divergence; The correlation difference is input into a convolutional neural network for training, so as to increase the correlation difference between the normal state and the abnormal state of the bearing and enhance the discrimination of the feature representation; The trained convolutional neural network is used to map the multi-dimensional features of the bearing into probability distribution to detect abnormal conditions of the bearing.

2. The bearing abnormality detection method based on time series correlation difference according to claim 1 is characterized in that: The extracting of adjacent time series features and non-adjacent time series features in the bearing sliding window respectively by using the adjacent attention module and the non-adjacent attention module includes: The input bearing feature is evenly divided into the first sub-feature and the second sub-feature according to the channel dimension; When the first sub-feature and the second sub-feature pass through the neighboring attention module and the non-neighboring attention module respectively, the bearing neighboring time series features and non-neighboring time series features are extracted.

3. The bearing abnormality detection method based on time series correlation difference according to claim 2 is characterized in that: When the first sub-feature passes through the neighboring attention module, the bearing neighboring temporal features are extracted, including: The first sub-feature is input into the neighboring attention module, and the query-key relationship between the current moment and multiple historical moments is defined, and the cross attention module is used to calculate the correlation features between the current moment and the neighboring moments step by step; the calculation formula of the cross attention module is: In the formula, , , Represent the query vector, key vector, and value vector respectively. express , The dimension size of The associated features of the multiple historical moments are fused to form a comprehensive adjacent temporal feature representation.

4. The bearing abnormality detection method based on time series correlation difference according to claim 3 is characterized in that: The method defines the query-key relationship between the current moment and multiple historical moments, and uses the cross attention module to calculate the correlation features between the current moment and the adjacent moments step by step, including: Define Tn as the current moment, Tn-1 as the previous moment, Tn-2 as the previous two moments, and so on until Tn-t is the previous t moments; The current moment Tn is used as the query vector, and the previous moment Tn-1 is used as the key vector and value vector, which are input into the cross attention module to obtain the associated features of the moment Tn relative to the previous moment; The current moment Tn is used as the query vector, and the previous two moments Tn-1 and Tn-2 are used as the key vector and value vector, which are input into the cross attention module to obtain the correlation features of the moment Tn relative to the previous two moments; The current moment Tn is used as the query vector, and the previous t moments Tn-1, Tn-2, …, Tn-t are used as key vectors and value vectors, which are input into the cross-attention module to obtain the associated features of moment Tn relative to the previous t moments.

5. The bearing abnormality detection method based on time series correlation difference according to claim 2 is characterized in that: When the second sub-feature passes through the non-adjacent attention module, the non-adjacent time series features of the bearing are extracted, including: The non-neighboring attention module performs sliding window sampling on the second sub-feature; Use sparse sampling to obtain a sparse representation of the input bearing features; The sparse features of the bearing are converted into query volume, key vector and value vector through the first linear layer, and the correlation between the query volume and the key vector is calculated by matrix multiplication; Apply the Softmax activation function to convert the correlation into a probability distribution vector; The probability distribution vector is matrix multiplied with the value vector, and converted through the second linear layer to obtain the non-adjacent time series features of the current moment of the bearing and the entire bearing sequence.

6. The bearing abnormality detection method based on time series correlation difference according to claim 1 is characterized in that: The calculation of the correlation difference between the adjacent time series features and the non-adjacent time series features by using the bearing time series correlation difference mechanism based on symmetric KL divergence includes: KL divergence is used to calculate the difference between adjacent time series features and non-adjacent time series features. The calculation formula is: in, and Represent the probability distribution of adjacent time series features and non-adjacent time series features respectively; KL divergence is used to calculate the difference between non-adjacent time series features and adjacent time series features. The calculation formula is: The symmetric KL divergence is used to calculate the point-by-point correlation difference between adjacent time series features and non-adjacent time series features. The calculation formula is: in, Priori association Associated with sequence The KL divergence calculated between two discrete distributions for each row, yes Compared with multi-layer prior association Associated with sequence The point-by-point correlation difference.

7. The bearing abnormality detection method based on time series correlation difference according to claim 1 is characterized in that: The calculation formula to increase the correlation difference between the normal state and the abnormal state of the bearing is: in, It represents the correlation difference between the prior correlation P and the sequence correlation S in the same sliding window; is a hyperparameter; Represents the value of the loss function.

8. A bearing anomaly detection system based on time series correlation difference, characterized in that: include: The feature extraction module is a convolutional neural network composed of multiple CNN modules. Each CNN module is used to extract multi-dimensional features of the bearing in the time domain and frequency domain from different scales. A time series feature learning module, used for extracting adjacent time series features and non-adjacent time series features in the bearing sliding window respectively through an adjacent attention module and a non-adjacent attention module; A time series correlation difference calculation module, used for calculating the correlation difference between the adjacent time series features and the non-adjacent time series features based on the bearing time series correlation difference mechanism of symmetric KL divergence; A network training module, used for inputting the correlation difference into a convolutional neural network for training, increasing the correlation difference between the normal state and the abnormal state of the bearing, and enhancing the discrimination of the feature representation; The anomaly detection module uses a trained convolutional neural network to map the multi-dimensional features of the bearing into a probability distribution to detect abnormal conditions of the bearing.

Images