Battery pole piece double-rolling equipment state prediction system and implementation method

Abstract

The invention belongs to the technical field of battery pole piece equipment detection, and particularly relates to a battery pole piece double-rolling equipment state prediction system and an implementation method. The method comprises the following steps of S1, acquiring vibration data of the double-rolling equipment and performing feature extraction; s2, dividing the vibration data into training set vibration data and test set vibration data, wherein the training set vibration data and the test set vibration data are two-dimensional space-time sequence data; s3, a CNN-BiGRU network is selected, the vibration data of the training set in the step S2 is applied to the CNN-BiGRU network, and a trained CNN-BiGRU model is obtained; s4, using the CNN-BiGRU model trained in the step S3 and the test set vibration data in the step S2 to predict the state of the double rolling equipment; according to the method, the CNN module and the BiGRU module are fused, and the local feature extraction capability of the CNN and the advantages of the BiGRU in processing time sequence data are effectively combined, so that the accuracy and efficiency of state prediction of the double rolling equipment are greatly improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of battery pole piece equipment detection, and in particular to a battery pole piece double-roller pressing equipment state prediction system and an implementation method thereof. Background Art

[0002] Faced with the challenges of environmental pollution and energy shortage, the international community is accelerating the development of new energy technologies. Among them, the production efficiency and quality of battery pole pieces, as the core of the battery, are directly affected by the operating status of the battery pole piece rolling mill. At present, pole piece rolling mills are developing towards high speed and large scale, but this also brings about the increased complexity of equipment maintenance. The double-roller pressing equipment for battery pole pieces directly determines the production quality and production efficiency of battery pole pieces.

[0003] Although deep learning models have shown significant potential in predicting equipment operating status, there are still many problems to be solved in the practical application of these models due to the limitations of computing resources and processing power of embedded platforms. This invention aims to explore the efficient hardware deployment of double-roller press equipment status prediction models on embedded platforms. The goal is to achieve high efficiency and high-precision prediction of the model to ensure the continuous and stable operation of the equipment, thereby improving the efficiency of the production process and the quality of the final product.

[0004] Therefore, the present application proposes a battery pole piece double-roll pressing equipment state prediction system and implementation method. Summary of the invention

[0005] In order to make up for the deficiencies of the prior art and solve the technical problems existing in the background technology, the present invention proposes a battery pole piece double-roll pressing equipment state prediction system and implementation method.

[0006] The present invention is achieved through the following technical solutions:

[0007] A method for realizing state prediction of a battery pole piece double-roll pressing device comprises the following steps:

[0008] S1: Obtain vibration data of double roller pressing equipment and perform feature extraction;

[0009] S2: dividing the vibration data into training set vibration data and test set vibration data, wherein both the training set vibration data and the test set vibration data are two-dimensional spatiotemporal sequence data;

[0010] S3: Select the CNN-BiGRU network and apply the training set vibration data in S2 to the CNN-BiGRU network to obtain the trained CNN-BiGRU model;

[0011] S4: Use the CNN-BiGRU model trained in S3 and the test set vibration data in S2 to predict the state of the double-roller press.

[0012] Preferably, the vibration data in S1 includes vibrations in the radial, vertical and horizontal directions of the bearing; the bearing vibration data is used as a timing signal to form one-dimensional time series data, and the data collected by multiple sensors are arranged in parallel at the same sampling rate to form two-dimensional space-time series data.

[0013] Preferably, the feature extraction in S1 is obtained by a time-frequency analysis method, which comprises the following steps:

[0014] P1: Extract key feature information in the time domain, frequency domain and time-frequency domain of vibration data;

[0015] P2: The extracted feature information is screened by applying monotonicity and correlation indicators to establish a sensitive degradation feature set that can effectively reflect the degradation state of the roll bearing.

[0016] Preferably, the time-frequency analysis method is based on wavelet transform, which decomposes the original signal layer by layer through wavelet function and scaling function to obtain low-frequency and high-frequency energy characteristics, thereby improving the prediction ability of the bearing degradation state.

[0017] Preferably, the CNN-BiGRU model includes:

[0018] An input layer, the input layer is used to standardize signal data of time-frequency domain features;

[0019] CNN layer, the CNN layer is used to obtain local spatial characteristics, input the data processed by the input layer into the CNN layer, and extract the spatial local characteristics of the data layer by layer;

[0020] A BiGRU layer, which captures the dynamic characteristics of time series data by analyzing the relevant vector information in both forward and backward directions;

[0021] The output layer is used to output the features learned by aggregating the vector information extracted in the BiGRU layer through the fully connected layer to obtain a prediction result.

[0022] Preferably, the CNN layer comprises:

[0023] The convolution layer is used to traverse the input data information, perform convolution operation with the local sequence using the convolution kernel weight, and obtain a preliminary feature matrix;

[0024] The pooling layer adopts maximum pooling and is responsible for data dimensionality reduction of the extracted feature matrix. By reducing the number of output feature vectors, the complexity and number of parameters of the model are reduced, thereby effectively avoiding overfitting of the model and retaining the main feature information.

[0025] A battery pole piece double roller pressing equipment state prediction system applied to the above includes:

[0026] A data set acquisition module, wherein the data set acquisition module uses an acceleration sensor, which is installed at corresponding positions of the roller bearings on the operating side and the transmission side to collect vibration data;

[0027] A storage module, wherein the storage module uses off-chip DRAM storage to store the data collected by the data collection module;

[0028] A data feature extraction and screening module, wherein the data feature extraction and screening module screens out key feature information from the collected data and establishes a sensitive degradation feature set that effectively reflects the degradation state of the roller bearing;

[0029] A prediction module, wherein the prediction module is used to train the CNN-BiGRU model using the feature set data to obtain a trained CNN-BiGRU model, and then test and evaluate the model;

[0030] A control module is used to manage data transmission between the FPGA off-chip storage module and the prediction module, including read and write operations of input data, weight data and intermediate calculation data.

[0031] The beneficial effects of the present invention are:

[0032] The present invention integrates the CNN module with the BiGRU module, effectively combining the local feature extraction capability of CNN and the advantage of BiGRU in processing time series data, thereby greatly improving the accuracy and efficiency of double-roller pressing equipment state prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is a diagram of a method for predicting the state of a double-roll pressing device in the present invention;

[0034] Figure 2 It is a diagram of the installation position of the vibration sensor in the present invention;

[0035] Figure 3 It is a flow chart of feature extraction of roller bearing vibration signal in the present invention;

[0036] Figure 4 It is a flowchart of CNN-BiGRU model training in the present invention;

[0037] Figure 5 It is a schematic diagram of the CNN structure in the present invention;

[0038] Figure 6 It is a schematic diagram of the BiGRU structure in the present invention;

[0039] Figure 7 This is a flow chart of the prediction system in the present invention. DETAILED DESCRIPTION

[0040] The present invention will be further described below in conjunction with specific examples. It should be understood that these examples are only used to illustrate the present invention and are not intended to limit the scope of the present invention. The experimental methods in the following examples without specifying specific conditions are usually carried out under conventional conditions or under conditions recommended by the manufacturer.

[0041] Unless otherwise defined, all professional and scientific terms used herein have the same meanings as those familiar to those skilled in the art. The reagents or raw materials used in the present invention can be purchased through conventional channels. Unless otherwise specified, the reagents or raw materials used in the present invention are used in a conventional manner in the art or in accordance with the product instructions. In addition, any method and material similar to or equivalent to the described content can be applied to the method of the present invention. The present invention is further described according to the accompanying drawings and specific embodiments of the present invention, and the preferred implementation methods and materials described in the present invention are only for demonstration purposes.

[0042] Embodiment 1:

[0043] like Figure 1 As shown, an embodiment of the present invention provides a data-driven double-roller pressing equipment state prediction method.

[0044] (1) The vibration data of the roller bearing in the radial, vertical and horizontal directions of the training set are used to extract sensitive degradation features to reduce the influence of interference signals and ensure the richness of degradation features.

[0045] (2) The feature signals extracted from the two directions of the bearing are standardized to ensure that the model can learn each feature at the same scale.

[0046] (3) Input the standardized feature data into the model, use CNN to perform deep feature mining on the data, and extract spatial features. Ensure data connection between CNN and BiGRU, and the extracted feature data will be input into the BiGRU layer, which comprehensively learns the past and future moments by considering the forward and backward relationship in the time series data. Finally, the model integrates and maps the output results of BiGRU through the fully connected layer and outputs the prediction results.

[0047] (4) The health index HI is constructed based on the fusion features. The model predicts the future status of the equipment by processing the relationship between HI and the time series. The model parameters are adjusted in iterative training to realize the construction of the model.

[0048] (5) Input the test set into the constructed model and perform the test according to steps 3 and 4. The RUL of the roller bearing is predicted and the prediction curve is obtained. Finally, quantitative analysis is performed using evaluation indicators such as RMSE to accurately evaluate the prediction performance of the constructed model.

[0049] Embodiment 2:

[0050] A method for realizing state prediction of a battery pole piece double-roll pressing device comprises the following steps:

[0051] S1: Obtain vibration data of the double-roller press and perform feature extraction.

[0052] S2: Divide the vibration data into training set vibration data and test set vibration data, wherein both the training set vibration data and the test set vibration data are two-dimensional spatiotemporal sequence data.

[0053] The present invention focuses on the vibration of the bearing in the radial vertical and horizontal directions. Therefore, the data collected by the acceleration sensors installed at the corresponding positions of the roller bearings on the operating side and the transmission side of the roller bearings of the double-roll press equipment are selected for analysis. Figure 2 As shown, the bearing vibration data as a time series signal will form one-dimensional time series data. The data collected by multiple sensors are arranged in parallel at the same sampling rate to form two-dimensional space-time series data.

[0054] The original high-dimensional vibration data is extracted by time-frequency analysis method to extract key feature information in the time domain, frequency domain and time-frequency domain of the vibration data. Then, the extracted features are screened by applying indicators such as monotonicity and correlation to establish a sensitive degradation feature set that can effectively reflect the degradation state of the roller bearing. The specific process is as follows Figure 3 As shown:

[0055] The original high-dimensional data obtained uses a time-frequency domain feature extraction method based on wavelet transform. This method decomposes the original signal layer by layer through wavelet function and scaling function to obtain low-frequency and high-frequency energy characteristics, thereby improving the prediction ability of bearing degradation state.

[0056] Band wavelet energy expression:

[0057]

[0058] The expression of frequency band wavelet energy ratio is:

[0059]

[0060] By performing wavelet packet decomposition on the bearing vibration signal and analyzing the frequency band performance at different levels, it is found that the optimal wavelet decomposition level of the vibration signal is 3 layers. The amplitude change of wavelet energy and wavelet energy ratio between the running-in period and the degradation period of the bearing can be used to determine the starting point of bearing degradation. This process lays a solid foundation for identifying the starting point of degradation and building a subsequent prediction model.

[0061] Then, the indicators with unclear degradation trends are screened out through monotonicity index, correlation index and comprehensive index to reduce the redundancy of feature sets.

[0062] Monotonicity index formula:

[0063]

[0064] Among them, X is the characteristic index value; x is t n The degradation index when N is the total length of the characteristic sequence; No.ofdd x >0 and No.ofdd x <0 is the number of numerical point derivatives in the indicator that are greater than 0 and less than 0, respectively. The closer the monotonicity value is to 1, the stronger the monotonicity of the feature.

[0065] Correlation indicator formula:

[0066]

[0067] Among them, x n t n The degradation index value at ; is the mean value of degradation index; is the mean of the operating cycle. The closer the correlation index is to 1, the stronger the correlation of the features is.

[0068] Comprehensive indicator formula:

[0069] F(X)=w1Mon(X)+w2Corr(X,T)

[0070] Where w1 and w2 are the corresponding weights of the two indicators. The larger the value of the comprehensive indicator, the more sensitive the feature is to the degradation process.

[0071] This method can not only solve the problem that recurrent neural networks are difficult to handle long-span data, but also greatly improve the accuracy and efficiency of predictions.

[0072] S3: Select the CNN-BiGRU network and apply the training set vibration data in S2 to the CNN-BiGRU network to obtain the trained CNN-BiGRU model.

[0073] The flowchart for training the CNN-BiGRU model is as follows Figure 4 As shown:

[0074] Input layer: This layer is mainly responsible for standardizing the signal data of time-frequency domain characteristics to ensure that the data has a uniform format and scale before entering the model.

[0075] The schematic diagram of CNN layer structure is as follows Figure 5 As shown in the figure, through the stacking of convolutional layers and pooling layers, the feature information of the input data is deeply mined layer by layer.

[0076] The convolution layer constitutes the core of CNN and is responsible for in-depth feature extraction of data. The formula for convolution calculation is as follows:

[0077]

[0078] In the formula, represents the convolution result of the jth channel of the lth layer; M j The feature vector representing the input; Represents the input of the i-th channel of the l-1th layer; Represents the weight matrix of the convolution kernel; represents the bias; f represents the activation function; * represents the convolution operation.

[0079] The pooling layer plays a key role in dimensionality reduction in CNN. It mainly reduces the complexity and number of parameters of the model by reducing the number of output feature vectors. Maximum pooling can better retain the significant information in the degraded features while reducing the sensitivity to background noise. Its calculation formula is:

[0080]

[0081] In the formula, j represents the number of pooling steps; k represents the pooling kernel size.

[0082] The CNN layer can effectively prevent the model from overfitting by reducing the dimension of the data operation of the input layer while retaining the key feature information.

[0083] BiGRU layers such as Figure 6 As shown in the figure, it can realize bidirectional processing of time series data, thereby fully capturing time series information. Its structural diagram is shown in Figure 3 As shown in Figure 1, the BiGRU model combines the forward GRU calculation and the reverse GRU calculation to make more comprehensive use of time series data. The forward GRU model makes predictions based on past information, denoted as The reverse GRU network uses future information to make predictions, denoted as BiGRU generates the final output by integrating the prediction features of these two directions. Its calculation expression can be expressed as:

[0084]

[0085] Among them, is the weight of the forward prediction result, and is the weight of the reverse prediction result.

[0086] S4: Use the CNN-BiGRU model trained in S3 and the test set vibration data in S2 to predict the state of the double-roller press.

[0087] The processed roller bearing sensitive degradation feature data is two-dimensional spatiotemporal sequence data, so CNN is used to achieve feature dimensionality reduction and extract deep degradation features; the GRU model has the RNN time feature extraction capability and can retain long-term historical information. The BiGRU structure enables the model to extract data features with long-distance dependence from both positive and negative directions and mine degradation information in the full operating cycle vibration data. Through such a hierarchical structure, the CNN-BiGRU model can effectively integrate spatiotemporal features and improve the accuracy of predicting the operating status of the lithium battery pole piece double roller pressing equipment.

[0088] Embodiment three:

[0089] like Figure 7 As shown, a battery pole piece double roller pressing equipment state prediction system applied to the above-mentioned battery pole piece double roller pressing equipment includes:

[0090] A data set acquisition module, wherein the data set acquisition module uses an acceleration sensor, which is installed at corresponding positions of the roller bearings on the operating side and the transmission side to collect vibration data;

[0091] A storage module, wherein the storage module uses off-chip DRAM storage to store the data collected by the data collection module;

[0092] A data feature extraction and screening module, wherein the data feature extraction and screening module screens out key feature information from the collected data and establishes a sensitive degradation feature set that effectively reflects the degradation state of the roller bearing;

[0093] A prediction module, wherein the prediction module is used to train the CNN-BiGRU model using the feature set data to obtain a trained CNN-BiGRU model, and then test and evaluate the model;

[0094] A control module is used to manage data transmission between the FPGA off-chip storage module and the prediction module, including read and write operations of input data, weight data and intermediate calculation data.

[0095] Any matters not described in the present invention are applicable to the prior art.

Claims

1. A method for realizing state prediction of battery pole piece double roller pressing equipment, characterized in that: The following steps are involved: S1: Obtain vibration data of double roller pressing equipment and perform feature extraction; S2: dividing the vibration data into training set vibration data and test set vibration data, wherein both the training set vibration data and the test set vibration data are two-dimensional spatiotemporal sequence data; S3: Select the CNN-BiGRU network and apply the training set vibration data in S2 to the CNN-BiGRU network to obtain the trained CNN-BiGRU model; S4: Use the CNN-BiGRU model trained in S3 and the test set vibration data in S2 to predict the state of the double-roller press.

2. The method for realizing state prediction of a battery pole piece double-roll pressing device according to claim 1, characterized in that: The vibration data in S1 includes the vibrations in the radial, vertical and horizontal directions of the bearing. The bearing vibration data is used as a timing signal to form one-dimensional time series data, and the data collected by multiple sensors are arranged in parallel at the same sampling rate to form two-dimensional space-time series data.

3. The method for realizing state prediction of a battery pole piece double-roll pressing device according to claim 1, characterized in that: The feature extraction in S1 is obtained by a time-frequency analysis method, which includes the following steps: P1: Extract key feature information in the time domain, frequency domain and time-frequency domain of vibration data; P2: The extracted feature information is screened by applying monotonicity and correlation indicators to establish a sensitive degradation feature set that can effectively reflect the degradation state of the roll bearing.

4. The method for realizing state prediction of a battery pole piece double-roll pressing device according to claim 3, characterized in that: The time-frequency analysis method is based on wavelet transform, which decomposes the original signal layer by layer through wavelet function and scaling function, thereby obtaining low-frequency and high-frequency energy characteristics, thereby improving the prediction ability of the bearing degradation state.

5. The method for realizing state prediction of battery pole piece double-roll pressing equipment according to claim 1, characterized in that: The CNN-BiGRU model includes: An input layer, the input layer is used to standardize signal data of time-frequency domain features; CNN layer, the CNN layer is used to obtain local spatial characteristics, input the data processed by the input layer into the CNN layer, and extract the spatial local characteristics of the data layer by layer; A BiGRU layer, which captures the dynamic characteristics of time series data by analyzing the relevant vector information in both the forward and backward directions; The output layer is used to output the features learned by aggregating the vector information extracted in the BiGRU layer through the fully connected layer to obtain a prediction result.

6. A method for realizing state prediction of a battery pole piece double-roll pressing device according to claim 5, characterized in that: The CNN layers include: The convolution layer is used to traverse the input data information, perform convolution operation with the local sequence using the convolution kernel weight, and obtain a preliminary feature matrix; The pooling layer adopts maximum pooling and is responsible for data dimensionality reduction of the extracted feature matrix. By reducing the number of output feature vectors, the complexity and number of parameters of the model are reduced, thereby effectively avoiding overfitting of the model and retaining the main feature information.

7. A battery pole piece double-roll pressing equipment state prediction system applied to claims 1-6, characterized in that: include: A data set acquisition module, wherein the data set acquisition module uses an acceleration sensor, which is installed at corresponding positions of the roller bearings on the operating side and the transmission side to collect vibration data; A storage module, wherein the storage module uses off-chip DRAM storage to store the data collected by the data collection module; A data feature extraction and screening module, wherein the data feature extraction and screening module screens out key feature information from the collected data and establishes a sensitive degradation feature set that effectively reflects the degradation state of the roller bearing; A prediction module, wherein the prediction module is used to train the CNN-BiGRU model using the feature set data to obtain a trained CNN-BiGRU model, and then test and evaluate the model; A control module is used to manage data transmission between the FPGA off-chip storage module and the prediction module, including read and write operations of input data, weight data and intermediate calculation data.

Images