Rolling bearing residual life prediction method based on bidirectional time convolution network and storage medium

Abstract

The invention relates to a rolling bearing residual life prediction method based on a bidirectional time convolution network and a storage medium, and relates to the technical field of fault prediction and health management. According to the scheme, original signals of a bearing are collected and preprocessed to obtain time series data, and the time series data are divided into multiple pieces of window data based on a preset local window; performing forward and backward time convolution processing on each piece of window data in the window; performing self-attention processing on each forward time convolution processing result and each backward time convolution processing result to obtain self-attention weighted features; and performing multi-layer perceptron processing on each self-attention weighted feature to obtain a corresponding life prediction value. The forward and backward time convolution processing is carried out on each window data in the window, so that the strong correlation of the time sequence can be kept while the local information can be fully mined, and the causal relationship of the whole time sequence can be better kept.

Description

Technical Field

[0001] The present invention relates to the technical field of fault prediction and health management, and in particular to a method and a storage medium for predicting the remaining life of a rolling bearing based on a bidirectional temporal convolutional network. Background Art

[0002] Rolling bearings are an important part of mechanical equipment. Since bearings operate in extremely challenging working environments, they are easily affected by factors such as wear and corrosion. Their failure can lead to equipment shutdown, production interruption, economic loss and safety hazards. Therefore, real-time health monitoring and remaining useful life (RUL) prediction of rolling bearings during operation are crucial. Accurate RUL prediction helps preventive maintenance, avoid sudden failures, and improve equipment reliability and availability.

[0003] At present, the methods used to predict the RUL of bearings mainly include methods based on physical models, methods based on statistical models, and data-driven methods. Physics-based models predict the RUL based on the working principle and wear mechanism of bearings through physical equations and models. These methods require a deep understanding of the physical properties and working environment of bearings. Statistical-based models require the use of historical fault data and life distribution models to estimate the RUL of bearings. These models usually require a large amount of historical data to produce reliable results. Data-driven models use machine learning and deep learning algorithms to predict RUL by analyzing sensor data such as vibration, temperature, and sound, which has been a hot topic in recent years. Temporal convolutional network (TCN) is a method based on the convolutional neural network architecture. It can calculate data of all time steps in parallel and effectively model long-term dependencies with fewer parameters. TCN has been widely used in various fields. However, TCN is strictly causal, which means that they only rely on current and past information when processing time series data, without considering future information. At the same time, local data dependencies are crucial in the prediction of bearing RUL prediction. Therefore, how to mine local important information while ensuring causality has become a trend in current methods. Summary of the invention

[0004] The technical problem to be solved by the present application is to provide a rolling bearing remaining life prediction method and storage medium based on a bidirectional time convolutional network, which has the characteristics of being able to more fully mine local information while maintaining the strong correlation of the time series and better maintaining the causal relationship of the overall time series.

[0005] In a first aspect, an embodiment provides a method for predicting the remaining life of a rolling bearing based on a bidirectional temporal convolutional network, comprising: The original signal of the bearing is collected and preprocessed to obtain time series data, and the time series data is divided into multiple window data based on the preset local window; Perform forward and backward temporal convolution processing on each window data within the window; Perform self-attention processing on each forward and backward temporal convolution result to obtain self-attention weighted features; Each self-attention weighted feature is processed by a multi-layer perceptron to obtain the corresponding lifespan prediction value.

[0006] In one embodiment, the original signal includes a vibration acceleration signal, a pressure signal and a torque signal.

[0007] In one embodiment, the time series data is divided into a plurality of window data based on a preset local window, including: Based on the preset local window, the time series data is divided into multiple window data so that the step length between windows is smaller than the length of the local window.

[0008] In one embodiment, the forward and backward time convolution processing is performed on each window data in the window, and the forward and backward time convolution processing is performed on any window data in the window, including: Performing forward time convolution processing on the window data within the window to obtain first feature data; Performing a backward time convolution process on the window data within the window to obtain second feature data; The first data feature and the second data feature are concatenated to obtain third feature data.

[0009] In one embodiment, for any window data, a method for performing forward time convolution processing includes: The window data is used as the input of the forward time convolution network, each signal data in the window data is sequentially subjected to forward time convolution processing, and each signal data after the forward time convolution processing is merged to obtain the first feature data corresponding to the window data; For any window data, the method of performing backward time convolution processing includes: The window data is used as the input of the backward time convolution network. After the time series data is reversed, each signal data in the reversed time series data is sequentially subjected to forward time convolution processing, and each signal data after the forward time convolution processing is merged to obtain the second feature data corresponding to the window data.

[0010] In one embodiment, the forward time convolution processing includes multiple layers of residual processing, each layer of residual processing includes two layers of dilated convolution layers; for any signal data, any layer of residual processing includes: , , , in, represents the output of the first layer of dilated convolution processing, l represents the index of the residual processing layer, t represents the index of the signal data in the window data, 1≤t≤w, w represents the length of the local window; represents the output of the second layer of dilated convolution, represents the input of the lth layer residual processing, Represents the preset weight matrix of the lth layer residual processing; Represents the output of the l-th layer residual processing; represents the weight parameter of the first layer of dilated convolution, represents the weight parameter of the second layer of dilated convolution, represents the dilation factor of the first layer of dilated convolution, Represents the dilation factor of the second layer of dilated convolution; DilatedConv1D represents one-dimensional dilated convolution processing, WN represents weight normalization processing, ReLU represents ReLU activation function, and Dropout represents regularization processing.

[0011] In one embodiment, the step of concatenating the first feature data and the second feature data to obtain the third feature data includes: , in, represents the third characteristic data, represents the first feature data, represents the second feature data, and n represents the index of the window data.

[0012] In one embodiment, for any forward and backward temporal convolution processing results, self-attention processing is performed to obtain self-attention weighted features, including: , in, Represents the self-attention weighted feature corresponding to any forward and backward temporal convolution processing result, represents self-attention processing, represents normalization processing, represents the dimension of the key matrix; represents the query matrix, , Represents the weight matrix of the query; represents the key matrix, , The weight matrix representing the key; represents the value matrix, , represents the weight matrix of the value, and T represents the transpose.

[0013] In one embodiment, for any self-attention weighted feature, multi-layer perceptron processing is performed to obtain the corresponding life prediction value, including: , , , in, Represents the output of the first fully connected layer in the multilayer perceptron, ReLU represents the ReLU activation function, represents the weight matrix of the first fully connected layer, Represents the bias vector of the first fully connected layer; represents the output of the second fully connected layer in the multilayer perceptron, represents the weight matrix of the second fully connected layer, Represents the bias vector of the second fully connected layer; represents the life expectancy prediction value, represents the weight matrix of the third fully connected layer in the multilayer perceptron, Represents the bias vector of the third fully connected layer.

[0014] In a second aspect, an embodiment provides a computer-readable storage medium, in which a program is stored, and the program can be loaded by a processor to execute the method for predicting the remaining life of a rolling bearing described in any one of the above embodiments.

[0015] The beneficial effects of the present invention are: The original signal of the bearing is collected and preprocessed to obtain time series data, and the time series data is divided into multiple window data based on the preset local window. Since each window data is processed with forward and backward time convolution within the window, the local information can be more fully mined while maintaining the strong correlation of the time series and better maintaining the causal relationship of the overall time series. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a schematic flow chart of a method for predicting the remaining life of a rolling bearing based on a bidirectional temporal convolutional network according to an embodiment of the present application; Figure 2 It is a flowchart of a method for performing forward and backward time volume processing on any window data according to an embodiment of the present application. DETAILED DESCRIPTION

[0017] The present invention is further described in detail below by specific embodiments in conjunction with the accompanying drawings. Wherein similar elements in different embodiments adopt associated similar element numbers. In the following embodiments, many detailed descriptions are for making the present application better understood. However, those skilled in the art can easily recognize that some features can be omitted in different situations, or can be replaced by other elements, materials, methods. In some cases, some operations related to the present application are not shown or described in the specification, this is to avoid the core part of the present application being overwhelmed by too much description, and for those skilled in the art, it is not necessary to describe these related operations in detail, and they can fully understand the related operations according to the description in the specification and the general technical knowledge in the art.

[0018] In addition, the features, operations or characteristics described in the specification can be combined in any appropriate manner to form various implementations. At the same time, the steps or actions in the method description can also be interchanged or adjusted in a manner that is obvious to those skilled in the art. Therefore, the various sequences in the specification and the drawings are only for the purpose of clearly describing a certain embodiment and are not meant to be a required sequence, unless otherwise specified that a certain sequence must be followed.

[0019] The serial numbers assigned to the components herein, such as "first", "second", etc., are only used to distinguish the objects described and do not have any order or technical meaning. The "connection" and "coupling" mentioned in this application, unless otherwise specified, include direct and indirect connections (couplings).

[0020] To facilitate the description of the inventive concept of the present application, the following briefly describes the rolling bearing remaining life prediction technology based on a bidirectional time convolutional network.

[0021] In the current technology, there is also a method for predicting the remaining life of rolling bearings based on a bidirectional time convolution network, but its bidirectional convolution is not based on a divided local window, and the bidirectional time convolution is performed on the window data in the window. Instead, it is more about performing only forward time convolution or ordinary convolution operations on the window data in the window. However, the applicant found in the study that bidirectional time convolution of the window data in the window is more conducive to fully mining local information. While more fully mining local information, it can maintain the strong correlation of the time series and better maintain the causal relationship of the overall time series.

[0022] In view of this, an embodiment of the present application provides a rolling bearing remaining life prediction method and storage medium based on a bidirectional time convolution network, which collects the original signal of the bearing and preprocesses it to obtain time series data, and divides the time series data into multiple window data based on a preset local window, and performs forward and backward time convolution processing on each window data within the window, so that local information can be more fully mined while maintaining the strong correlation of the time series, thereby better maintaining the causal relationship of the overall time series.

[0023] A method for predicting the remaining life of rolling bearings based on bidirectional temporal convolutional networks, please refer to Figure 1 ,include: Step S10, collecting the original signal of the bearing and preprocessing it to obtain time series data, and dividing the time series data into a plurality of window data based on a preset local window.

[0024] In one embodiment, it is assumed that the obtained time series data is , where m represents the index of the moment, 1≤m≤M, and M represents the total length of the time series. The signal data representing the time m may include vibration acceleration signal data, pressure signal data, torque signal data, and the like.

[0025] Assuming the length of each local window is w, the time series X can be divided into multiple window data, and the step length between windows is s. If the step length s=w, there is no overlap between windows. If s<w, there will be overlapping areas between windows. If s>w, there will be gaps between windows, and some data may not be covered. Therefore, in one embodiment, based on a preset local window, the step length between windows is less than the length of the local window, that is, s<w, and the time series data is divided into multiple window data. For the nth window, the obtained nth window data can be expressed as , where t represents the index of the signal data in the window data, Represents the t-th signal data in the n-th window data, 1≤t≤w.

[0026] Step S20, performing forward and backward time convolution processing on each window data within the window.

[0027] In each local time window, the window data is processed by forward and backward bidirectional temporal convolution. Two independent temporal convolution (TCN) branches are used, including a forward TCN branch and a backward TCN branch, to perform bidirectional temporal convolution. For any window data, the forward and backward temporal convolution processing may include: Step S201, performing forward time convolution processing on window data within the window to obtain first feature data.

[0028] In one embodiment, for any window data, a method for performing forward time convolution processing includes: The window data is used as the input of the forward time convolution network, and each signal data in the window data is sequentially subjected to forward time convolution processing, and each signal data after the forward time convolution processing is merged to obtain the first feature data corresponding to the window data.

[0029] In one embodiment, the forward temporal convolution processing includes multiple layers of residual processing, each layer of residual processing includes two dilated convolution layers.

[0030] Multi-layer residual processing can effectively capture long dependencies in time series. Each dilated convolution layer can expand the receptive field of the convolution kernel by the dilation factor, so that the convolution operation can cover a wider range of time steps.

[0031] In one embodiment, for any signal data, the residual processing of any layer includes: , , , in, represents the output of the first layer of dilated convolution processing, l represents the index of the residual processing layer, represents the output of the second layer of dilated convolution, represents the input of the lth layer residual processing, Represents the preset weight matrix of the lth layer residual processing; It represents the output of the residual processing of the lth layer and the input of the residual processing of the l+1th layer; represents the weight parameter of the first layer of dilated convolution, represents the weight parameter of the second layer of dilated convolution, represents the dilation factor of the first layer of dilated convolution, Represents the dilation factor of the second layer of dilated convolution; DilatedConv1D represents one-dimensional dilated convolution processing, WN represents weight normalization processing, ReLU represents ReLU activation function, and Dropout represents regularization processing.

[0032] The signal data after the last layer of residual processing is merged to obtain the first feature data obtained by forward time convolution processing of the input window data.

[0033] For any window data , after forward time convolution processing, the corresponding first feature data can be obtained .

[0034] Step S202, performing backward time convolution processing on the window data within the window to obtain second feature data.

[0035] In one embodiment, for any window data, a method for performing backward time convolution processing includes: The window data is used as the input of the backward time convolution network. After the time series data is reversed, each signal data in the reversed time series data is sequentially subjected to forward time convolution processing, and each signal data after the forward time convolution processing is merged to obtain the second feature data corresponding to the window data.

[0036] For any window data , first reverse the time series data to obtain the reversed window data , and then the inverted window data After forward time convolution processing, the corresponding second feature data can be obtained , then the second feature data can be expressed as: , in, represents the backward temporal convolution, .

[0037] Step S203, concatenating the first feature data and the second feature data to obtain third feature data.

[0038] In one embodiment, the third characteristic data can be expressed as: , in, Represents the third feature data. In this way, the previous and next information can be captured simultaneously in each window, while ensuring the causality of the overall time series.

[0039] In one embodiment, the third feature data is taken as the result of forward and backward temporal convolution processing.

[0040] Step S30, performing self-attention processing on each forward and backward temporal convolution processing result to obtain a self-attention weighted feature.

[0041] In one embodiment, for any forward and backward temporal convolution processing results, self-attention processing is performed, and the obtained self-attention weighted features can be expressed as: , in, Represents the self-attention weighted feature corresponding to any forward and backward temporal convolution processing result, represents self-attention processing, represents normalization processing, represents the dimension of the key matrix; represents the query matrix, , Represents the weight matrix of the query; represents the key matrix, , The weight matrix representing the key; represents the value matrix, , represents the weight matrix of the value, and T represents the transpose.

[0042] By applying the self-attention mechanism to each local window, the complex dependencies and key features in the local time series can be captured more effectively.

[0043] Step S40, performing multi-layer perceptron processing on each self-attention weighted feature to obtain a corresponding lifespan prediction value.

[0044] In one embodiment, for any self-attention weighted feature, multi-layer perceptron processing is performed to obtain the corresponding life prediction value, which can be expressed as: , , , in, Represents the output of the first fully connected layer in the multilayer perceptron, ReLU represents the ReLU activation function, represents the weight matrix of the first fully connected layer, Represents the bias vector of the first fully connected layer; represents the output of the second fully connected layer in the multilayer perceptron, represents the weight matrix of the second fully connected layer, Represents the bias vector of the second fully connected layer; represents the life expectancy prediction value, represents the weight matrix of the third fully connected layer in the multilayer perceptron, Represents the bias vector of the third fully connected layer.

[0045] Based on the above embodiment, the life prediction value of each window data is obtained by inputting the self-attention weighted features into three fully connected layers of the multilayer perceptron.

[0046] In any of the above embodiments, by dividing the local time window and creatively applying the bidirectional time convolutional network in the local window, the bidirectional information in the local window is used for feature extraction, the local information in the bearing sequence is fully mined and the overall causal relationship is maintained, thereby improving the prediction performance of the model. At the same time, the self-attention mechanism is introduced to automatically assign attention weights to the data of each time step, so that it can focus on more important information, further improving the performance of the network.

[0047] The method for predicting the remaining life of a rolling bearing based on a bidirectional time convolution network in the above-mentioned embodiment is implemented based on a trained rolling bearing remaining life prediction network, which is a neural network obtained by training based on a constructed basic neural network architecture combined with a constructed loss function. Among them, the basic neural network architecture includes a bidirectional time convolution subnetwork, a self-attention processing subnetwork and a multi-layer perceptron. Among them, the bidirectional time convolution subnetwork includes a forward TCN branch and a backward TCN branch, which are used to perform forward TCN processing and backward TCN processing on the input window data respectively, and the bidirectional time convolution subnetwork also includes a splicing processing layer for splicing the forward TCN processing results and the backward TCN processing results. The self-attention processing subnetwork is used to perform self-attention processing on each forward and backward time convolution processing result to obtain a self-attention weighted feature. The multi-layer perceptron is used to perform multi-layer perceptron processing on each self-attention weighted feature to obtain a predicted life value. The loss function calculates the distance between the predicted life value and the true life value and provides feedback until the network training meets the requirements.

[0048] In order to verify the effectiveness of the rolling bearing remaining life prediction method based on the bidirectional time convolution network in the above embodiment, in one embodiment, an aviation bearing data set is used for verification experiments. The hardware structure of the civil aircraft bearing component test bench consists of eight parts: a drive motor, a gearbox, a rolling bearing, a connecting shaft, a disc flywheel, a pulley, a coupling, and an adjustable load. The sensors used to collect the characteristics of the bearing from normal operation to failure include a vibration acceleration sensor, a torque sensor, a temperature sensor, a pressure sensor, and an air temperature and humidity sensor. The experiment of this embodiment uses a vibration acceleration sensor, a torque sensor, and a pressure sensor to collect data signals. The sampling interval of the test bench is 10 seconds, each acquisition cycle is 0.1 seconds, and each bearing sample generates 2500 data points.

[0049] In order to verify the accuracy and robustness of the proposed prediction method, two evaluation indicators, MAE and MSE, are introduced; MAE is used to reflect the true error. MSE is more sensitive to larger errors in the sample and can detect larger and less common errors. Taken together, these two indicators can well reflect the prediction performance. For the sake of completeness, the network structure of this application is compared with the classic network structure for predicting the RUL of bearings. The selected network structures include FCN, LSTM, Bi-LSTM and TCN. The performance results of bearing 1 and bearing 2 are shown in Table 1: Table 1 , The experimental results show that the performance of the proposed method is better than other existing models in terms of any evaluation indicators. Analyzing the average evaluation indicators of the two bearings, when compared with the TCN model, FCN model, LSTM model and Bi-LSTM in the existing models, the proposed method reduces the MAE by 5.1%, 21.5%, 16.7% and 13.8% in bearing 1, and the MSE by 1.4%, 16.7%, 7.2% and 4.3% respectively.

[0050] The above is only a specific implementation method of the present invention, but the protection scope of the present invention is not limited to this. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

[0051] In one embodiment of the present application, a computer-readable storage medium is provided, on which a program is stored. The stored program includes a method that can be loaded by a processor and process any of the above embodiments.

[0052] Those skilled in the art will appreciate that all or part of the functions of the various methods in the above-mentioned embodiments can be implemented by hardware or by computer programs. When all or part of the functions in the above-mentioned embodiments are implemented by computer programs, the program can be stored in a computer-readable storage medium, and the storage medium can include: read-only memory, random access memory, disk, optical disk, hard disk, etc., and the program is executed by a computer to implement the above-mentioned functions. For example, the program is stored in the memory of the device, and when the program in the memory is executed by the processor, all or part of the above-mentioned functions can be implemented. In addition, when all or part of the functions in the above-mentioned embodiments are implemented by computer programs, the program can also be stored in a storage medium such as a server, another computer, disk, optical disk, flash disk or mobile hard disk, and can be downloaded or copied and saved in the memory of the local device, or the system of the local device is updated, and when the program in the memory is executed by the processor, all or part of the functions in the above-mentioned embodiments can be implemented.

[0053] The above specific examples are used to illustrate the present invention, which is only used to help understand the present invention and is not intended to limit the present invention. For those skilled in the art, according to the concept of the present invention, some simple deductions, modifications or substitutions can be made.

Claims

1. A method for predicting the remaining life of a rolling bearing based on a bidirectional temporal convolutional network, characterized in that: include: The original signal of the bearing is collected and preprocessed to obtain time series data, and the time series data is divided into multiple window data based on the preset local window; Perform forward and backward temporal convolution processing on each window data within the window; Perform self-attention processing on each forward and backward temporal convolution result to obtain self-attention weighted features; Each self-attention weighted feature is processed by a multi-layer perceptron to obtain the corresponding lifespan prediction value.

2. The method for predicting the remaining life of a rolling bearing according to claim 1, characterized in that: The original signal includes a vibration acceleration signal, a pressure signal and a torque signal.

3. The method for predicting the remaining life of a rolling bearing according to claim 1, characterized in that: The time series data is divided into a plurality of window data based on the preset local window, including: Based on the preset local window, the time series data is divided into multiple window data so that the step length between windows is smaller than the length of the local window.

4. The method for predicting the remaining life of a rolling bearing according to claim 1, characterized in that: The forward and backward time convolution processing is performed on each window data in the window, and the forward and backward time convolution processing is performed on any window data in the window, including: Performing forward time convolution processing on the window data within the window to obtain first feature data; Performing a backward time convolution process on the window data within the window to obtain second feature data; The first data feature and the second data feature are concatenated to obtain third feature data.

5. The method for predicting the remaining life of a rolling bearing according to claim 4, characterized in that: For any window data, the method of forward time convolution processing includes: The window data is used as the input of the forward time convolution network, each signal data in the window data is sequentially subjected to forward time convolution processing, and each signal data after the forward time convolution processing is merged to obtain the first feature data corresponding to the window data; For any window data, the method of performing backward time convolution processing includes: The window data is used as the input of the backward time convolution network. After the time series data is reversed, each signal data in the reversed time series data is sequentially subjected to forward time convolution processing, and each signal data after the forward time convolution processing is merged to obtain the second feature data corresponding to the window data.

6. The method for predicting the remaining life of a rolling bearing according to claim 5, characterized in that: The forward time convolution processing includes multiple layers of residual processing, each layer of residual processing includes two layers of dilated convolution layers; for any signal data, any layer of residual processing includes: ， ， ， in, represents the output of the first layer of dilated convolution processing, l represents the index of the residual processing layer, t represents the index of the signal data in the window data, 1≤t≤w, w represents the length of the local window; represents the output of the second layer of dilated convolution, represents the input of the lth layer residual processing, Represents the preset weight matrix of the lth layer residual processing; Represents the output of the l-th layer residual processing; represents the weight parameter of the first layer of dilated convolution, represents the weight parameter of the second layer of dilated convolution, represents the dilation factor of the first layer of dilated convolution, Represents the dilation factor of the second layer of dilated convolution; DilatedConv1D represents one-dimensional dilated convolution processing, WN represents weight normalization processing, ReLU represents ReLU activation function, and Dropout represents regularization processing.

7. The method for predicting the remaining life of a rolling bearing according to claim 5, characterized in that: The step of concatenating the first feature data and the second feature data to obtain the third feature data includes: ， in, represents the third characteristic data, represents the first feature data, represents the second feature data, and n represents the index of the window data.

8. The method for predicting the remaining life of a rolling bearing according to claim 7, characterized in that: For any forward and backward temporal convolution processing results, self-attention processing is performed to obtain self-attention weighted features, including: ， in, Represents the self-attention weighted feature corresponding to any forward and backward temporal convolution processing result, represents self-attention processing, represents normalization processing, represents the dimension of the key matrix; represents the query matrix, , Represents the weight matrix of the query; represents the key matrix, , The weight matrix representing the key; represents the value matrix, , represents the weight matrix of the value, and T represents the transpose.

9. The method for predicting the remaining life of a rolling bearing according to claim 8, characterized in that: For any self-attention weighted feature, multi-layer perceptron processing is performed to obtain the corresponding lifespan prediction value, including: ， ， ， in, Represents the output of the first fully connected layer in the multilayer perceptron, ReLU represents the ReLU activation function, represents the weight matrix of the first fully connected layer, Represents the bias vector of the first fully connected layer; represents the output of the second fully connected layer in the multilayer perceptron, represents the weight matrix of the second fully connected layer, Represents the bias vector of the second fully connected layer; represents the life expectancy prediction value, represents the weight matrix of the third fully connected layer in the multilayer perceptron, Represents the bias vector of the third fully connected layer.

10. A computer-readable storage medium, characterized in that: The medium stores a program, which can be loaded by a processor and execute the method for predicting the remaining life of a rolling bearing as described in any one of claims 1 to 9.

Images