Rolling Bearing Remaining Life Prediction Method and System Based on Gaussian Elastic Convolution

Abstract

The present invention provides a method and system for predicting the remaining life of a rolling bearing based on Gaussian elastic convolution, belonging to the technical field of predicting the remaining life of mechanical equipment based on neural networks. The vibration signal and rotational speed signal of the rolling bearing under variable rotational speed conditions are obtained; the obtained vibration signal and rotational speed signal are processed by a pre-trained life prediction model to obtain the life prediction result of the bearing. Among them, in the training of the life prediction model, a time information embedding layer is constructed, the rotational speed information of the bearing and the bearing operation time information are introduced into the representation learning, the degradation features are recalibrated according to different degradation rates, and the time series information for regression analysis of the remaining life prediction is obtained. The present invention can directly extract the degradation features with elastic scales from the variable rotational speed vibration monitoring signals of the rolling bearing and perform feature recalibration, accurately excavate the degradation information most relevant to the health status of the rolling bearing, and realize the prediction of the remaining life of the rolling bearing under variable rotational speed conditions.

Description

Technical Field

[0001] The present invention relates to the technical field of predicting the remaining useful life of mechanical equipment based on neural networks, and in particular to a method and system for predicting the remaining useful life of rolling bearings based on Gaussian elastic convolution. Background Art

[0002] Rolling bearings are widely used in mechanical equipment such as railway vehicles, aircraft engines, and industrial robots, and are key rotating components for fixing structures and transmitting loads. During the long-term service of mechanical equipment, the performance of rolling bearings continuously deteriorates, and various failures are likely to occur, which may lead to unplanned shutdowns or even catastrophic safety accidents. Therefore, it is necessary to predict the remaining useful life of rolling bearings to guide maintenance personnel in formulating reasonable maintenance plans, thereby improving the reliability, availability, maintainability, and safety of mechanical equipment. The popularization and application of industrial Internet of Things and artificial intelligence have brought mechanical health monitoring into the "big data" era, and data-driven remaining useful life prediction methods have thus received extensive attention. However, the complex and variable working conditions during data acquisition also pose more severe challenges to the prediction of the remaining useful life of rolling bearings. Therefore, it is necessary to invent a new data-driven method for predicting the remaining useful life of rolling bearings to ensure the safe operation of mechanical equipment.

[0003] By introducing deep learning theory, data-driven remaining useful life prediction methods can effectively extract the degradation information of rolling bearings from monitoring data, overcoming the shortcomings of traditional methods that rely too much on prior knowledge, and thus have been widely applied. However, in industrial applications, the rotational speed of bearings usually changes continuously. Existing deep learning-based prediction methods usually only focus on predicting the remaining useful life of bearings under constant working conditions, and do not consider the change of bearing rotational speed during network construction and representation learning. Therefore, they cannot effectively extract the degradation information of bearings from variable-speed data, seriously affecting the performance of the deep prediction network. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and system for predicting the remaining useful life of rolling bearings based on Gaussian elastic convolution, which takes the vibration monitoring signal and rotational speed signal of the bearing as inputs, directly extracts degradation features with elastic scales, and can perform feature recalibration on these degradation features based on time series information, with higher prediction accuracy, better stability, and stronger robustness, so as to solve at least one of the technical problems existing in the above background art.

[0005] To achieve the above purpose, the present invention adopts the following technical solutions:

[0006] On the one hand, the present invention provides a method for predicting the remaining useful life of rolling bearings based on Gaussian elastic convolution, including:

[0007] Obtain the vibration signal and rotational speed signal of the rolling bearing under variable rotational speed conditions;

[0008] Use the pre-trained life prediction model to process the obtained vibration signal and rotational speed signal to obtain the life prediction result of the rolling bearing;

[0009] Among them, the training of the life prediction model includes:

[0010] Obtain the vibration signal and rotational speed signal of the rolling bearing under variable rotational speed conditions and perform preprocessing;

[0011] Construct multiple parallel and independent representation learning paths corresponding to the monitoring signals of time steps in the preprocessed signal vibration signal and rotational speed signal, and extract the degradation features with elastic scale in the input data;

[0012] Construct a time information embedding layer, introduce the bearing rotational speed information and bearing operating time information into the representation learning, recalibrate the degradation features according to different degradation rates, and obtain the time series information for regression analysis of remaining life prediction;

[0013] Concatenate the recalibrated degradation features head and tail to obtain a concatenated vector, input the concatenated vector into the fully connected layer to obtain the predicted value of the remaining life of the bearing corresponding to the input data; Based on the Adam optimization algorithm, iteratively update the model parameters to obtain the optimal rolling bearing remaining life prediction model.

[0014] Optionally, the preprocessing of the original vibration signal and rotational speed signal includes: for the vibration monitoring signal x t Perform Z-score normalization processing, and embed the effective time information data before the sampling moment into the current data through the time window embedding strategy:

[0015] Perform normalization processing on the vibration monitoring signal x t through Z-score to obtain the preprocessed monitoring data x Z t ; The Z-score standardization operation expression is as follows:

[0016]

[0017] In the formula, x t is the original signal; is the mean of the original signal; σ x t is the standard deviation of the original signal;

[0018] Set a time window of size S to integrate the data {x Z t ,v t}, and the monitoring signal samples obtained at the previous S - 1 sampling moments, to obtain the signal {x' t , v' t}, where v' t = {v t-S+1 , v t-S+2 , …, v t}.

[0019] Optionally, construct S parallel and independent representation learning paths, corresponding to the monitoring signals of S time steps in the signal {x' t , v' t}, which have the same network structure and hyperparameter settings. The data of each time step is respectively input into the corresponding representation learning path. Each representation learning path is obtained by cross - stacking N Gaussian elastic convolutional layers and N max - pooling layers to extract the degraded features with elastic scales in the input data.

[0020] Optionally, construct a time information embedding layer to introduce the bearing rotation speed information and bearing operation time information into the representation learning, so that the network can perform feature recalibration according to different degradation rates, and provide comprehensive time - series information for the subsequent regression analysis of remaining life prediction. The specific steps of the information aggregation layer are as follows:

[0021] Based on the L 10 life calculation formula of the bearing and the exponential degradation model, calculate the calibration coefficient corresponding to each representation learning path. The calibration coefficient α i corresponding to the i - th representation learning path is calculated as follows:

[0022]

[0023]

[0024] In the formula, v i and t i are respectively the i - th bearing speed value and working time within a time window of size S, is the bearing L i life at the rotational speed v 10 , F C is the basic dynamic load rating, F P is the equivalent dynamic bearing load, and r is the life index;

[0025] Based on the calculated calibration coefficients perform feature recalibration on the degraded features with elastic scales obtained by each representation learning path , and the calculation formula is as follows:

[0026]

[0027] In the formula, represents the re-scaled degraded feature, represents the element-wise product.

[0028] Optionally, the S re-scaled degraded features are concatenated head-to-tail to obtain a vector H, and the calculation expression is as follows: In the formula, represents the vector concatenation operation; the vector H is input into the fully connected layer to obtain the predicted remaining life preRUL of the bearing corresponding to the input data {x' t , v' t}, and the calculation expression is as follows: preRUL = δ(W F H + b F ); in the formula, W F represents the weight matrix; b F represents the bias vector; δ(·) represents the linear rectification unit activation function.

[0029] Optionally, based on the Adam optimization algorithm, the number of iterations E is set, and the parameters of the Gaussian elastic convolution network are iteratively updated to obtain an optimal rolling bearing remaining life prediction model, that is, to minimize the mean square error objective function:

[0030]

[0031] In the formula, actRUL i is the true remaining life value of the sample; B represents the number of mini-batch samples.

[0032] In a second aspect, the present invention provides a rolling bearing remaining life prediction system based on Gaussian elastic convolution, including:

[0033] An acquisition module, configured to acquire the vibration signal and the rotational speed signal of the rolling bearing under variable rotational speed conditions;

[0034] A prediction module, configured to process the acquired vibration signal and rotational speed signal by using a pre-trained life prediction model to obtain a life prediction result of the rolling bearing;

[0035] Among them, the training of the life prediction model includes:

[0036] Acquire the vibration signal and the rotational speed signal of the rolling bearing under variable rotational speed conditions and perform preprocessing;

[0037] Construct a plurality of parallel and independent representation learning paths corresponding to the monitoring signals of the time steps in the preprocessed signal vibration signal and rotational speed signal, and extract the degraded features with elastic scales in the input data;

[0038] Construct a time information embedding layer, introduce the bearing rotation speed information and bearing operation time information into the representation learning, recalibrate the degradation features according to different degradation rates, and obtain the time series information for regression analysis of remaining life prediction;

[0039] Concatenate the head and tail of the recalibrated degradation features to obtain a concatenated vector, input the concatenated vector into the fully connected layer to obtain the predicted value of the remaining life of the bearing corresponding to the input data; Based on the Adam optimization algorithm, iteratively update the model parameters to obtain an optimal rolling bearing remaining life prediction model.

[0040] In a third aspect, the present invention provides a non-transitory computer-readable storage medium, which is used to store computer instructions. When the computer instructions are executed by a processor, the method for predicting the remaining life of a rolling bearing based on Gaussian elastic convolution as described above is implemented.

[0041] In a fourth aspect, the present invention provides a computer program product, including a computer program, which is used to implement the method for predicting the remaining life of a rolling bearing based on Gaussian elastic convolution as described above when running on one or more processors.

[0042] In a fifth aspect, the present invention provides an electronic device, including: a processor, a memory, and a computer program; wherein, the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device runs, the processor executes the computer program stored in the memory, so that the electronic device executes the instruction for implementing the method for predicting the remaining life of a rolling bearing based on Gaussian elastic convolution as described above.

[0043] The beneficial effects of the present invention: It can directly extract degradation features with elastic scales from the variable speed vibration monitoring signals of rolling bearings, and can perform feature recalibration based on the speed signals, and then can accurately mine the degradation information most relevant to the health status of rolling bearings; It overcomes the defect that the existing methods can only extract features at a fixed scale, and realizes the prediction of the remaining life of rolling bearings under variable speed conditions.

[0044] The advantages of the additional aspects of the present invention will be more clearly given in the following description part, or can be understood through the practice of the present invention. Description of the Drawings

[0045] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0046] Figure 1 The flowchart of the rolling bearing remaining life prediction method based on Gaussian elastic convolution according to the embodiment of the present invention.

[0047] Figure 2 The example diagram of the bearing rotation speed changing with the operation time according to the embodiment of the present invention.

[0048] Figure 3 The remaining life prediction result diagram of the variable speed rolling bearing according to the embodiment of the present invention.

[0049] Figure 4 The comparison diagram of the remaining life prediction performance of each method in the ablation experiment according to the embodiment of the present invention. Detailed implementation manners

[0050] The following details the implementation manners of the present invention. The examples of the implementation manners are shown in the drawings, where the same or similar reference numerals represent the same or similar elements or elements with the same or similar functions from beginning to end. The implementation manners described through the drawings are exemplary and are only used to explain the present invention, and cannot be construed as a limitation to the present invention.

[0051] Those skilled in the art of the present technology can understand that, unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meaning as the general understanding of those of ordinary skill in the art in the field to which the present invention belongs.

[0052] It should also be understood that terms such as those defined in a general dictionary should be understood to have a meaning consistent with the meaning in the context of the prior art, and will not be interpreted with an idealized or overly formal meaning unless defined as here.

[0053] Those skilled in the art of the present technology can understand that, unless specifically stated, the singular forms "a", "an", "the" and "said" used herein may also include the plural forms. It should be further understood that the term "comprising" used in the specification of the present invention means the presence of the described features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements and / or their groups.

[0054] In the description of this specification, the descriptions referring to terms such as "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0055] For ease of understanding the present invention, the following further explains the present invention with specific embodiments in conjunction with the accompanying drawings, and the specific embodiments do not constitute a limitation to the embodiments of the present invention.

[0056] Those skilled in the art should understand that the drawings are only schematic diagrams of the embodiments, and the components in the drawings are not necessarily essential for implementing the present invention.

[0057] Embodiment 1

[0058] In this Embodiment 1, first, a rolling bearing remaining life prediction system based on Gaussian elastic convolution is provided, including: an acquisition module for acquiring the vibration signal and speed signal of the rolling bearing under variable speed conditions; a prediction module for processing the acquired vibration signal and speed signal using a pre-trained life prediction model to obtain the life prediction result of the rolling bearing.

[0059] Using the above system, a rolling bearing remaining life prediction method based on Gaussian elastic convolution is realized, including: first using the acquisition module to acquire the vibration signal and speed signal of the rolling bearing under variable speed conditions to be predicted; then using the prediction module to process the acquired vibration signal and speed signal using a pre-trained life prediction model to obtain the life prediction result of the rolling bearing.

[0060] Among them, the training of the life prediction model includes:

[0061] Acquire the vibration signal and speed signal of the rolling bearing under variable speed conditions and perform preprocessing;

[0062] Construct multiple parallel and independent representation learning paths corresponding to the monitoring signals of the time steps in the preprocessed signal vibration signal and speed signal, and extract the degradation features with elastic scale in the input data;

[0063] Construct a time information embedding layer, introduce the bearing speed information and bearing operation time information into the representation learning, recalibrate the degradation features according to different degradation rates, and obtain the time series information for regression analysis of remaining life prediction;

[0064] Concatenate the head and tail of the re-scaled degraded features to obtain a concatenated vector, and input the concatenated vector into a fully connected layer to obtain the predicted remaining life value of the bearing corresponding to the input data; based on the Adam optimization algorithm, iteratively update the model parameters to obtain an optimal remaining life prediction model for rolling bearings.

[0065] Among them, the preprocessing of the original vibration signal and rotational speed signal includes: for the vibration monitoring signal x t Perform Z-score normalization processing, and embed the effective time information data before the sampling moment into the current data through the time window embedding strategy:

[0066] Perform normalization processing on the vibration monitoring signal x through Z-score t to obtain the preprocessed monitoring data x Z t ; The expression of the Z-score standardization operation is as follows:

[0067]

[0068] In the formula, x t is the original signal; is the mean of the original signal; σ x t is the standard deviation of the original signal;

[0069] Set a time window of size S, and integrate the data {x Z t , v t} and the monitoring signal samples obtained at the previous S - 1 sampling moments to obtain the signal {x' t , v' t}, where v' t = {v t-S+1 , v t-S+2 , …, v t}.

[0070] Among them, construct S parallel and independent representation learning paths, corresponding to the monitoring signals at S time steps in the signal {x' t , v' t}, they have the same network structure and hyperparameter settings, and the data at each time step is respectively input into the corresponding representation learning path. Each representation learning path is obtained by cross-stacking N Gaussian elastic convolutional layers and N max-pooling layers to extract the degraded features with elastic scales in the input data.

[0071] Construct a time information embedding layer to introduce the bearing rotation speed information and bearing operation time information into the representation learning, enabling the network to perform feature recalibration according to different degradation rates and providing comprehensive time series information for the subsequent regression analysis of remaining life prediction. The specific steps of the information aggregation layer are as follows:

[0072] Based on the bearing L 10 life calculation formula and the exponential degradation model, calculate the calibration coefficient corresponding to each representation learning path. The calibration coefficient α i corresponding to the i-th representation learning path is calculated as follows:

[0073]

[0074]

[0075] In the formula, v i and t i are respectively the i-th bearing speed value and working time within the time window of size S, is the bearing L i life at the rotational speed v 10 , F C is the basic dynamic load rating, F P is the equivalent dynamic bearing load, and r is the life exponent;

[0076] Based on the calculated calibration coefficient perform feature recalibration on the degradation features with elastic scale obtained from each representation learning path , and the calculation formula is as follows:

[0077]

[0078] In the formula, represents the recalibrated degradation feature, represents the element-wise product.

[0079] Concatenate the S recalibrated degradation features end to end to obtain the vector H, and the calculation formula is as follows: In the formula, represents the vector concatenation operation; input the vector H into the fully connected layer to obtain the predicted remaining life preRUL of the bearing corresponding to the input data {x' t , v' t}, and the calculation formula is as follows: preRUL = δ(W F H + b F ); In the formula, W F represents the weight matrix; b F represents the bias vector; δ(·) represents the linear rectifier unit activation function.

[0080] Based on the Adam optimization algorithm, set the number of iterations E, and iteratively update the parameters of the Gaussian elastic convolutional network to obtain the optimal remaining useful life prediction model of the rolling bearing, that is, minimize the mean square error objective function:

[0081]

[0082] In the formula, actRUL i is the true remaining useful life value of the sample; B represents the number of mini-batch samples.

[0083] Example 2

[0084] In this Example 2, a method for predicting the remaining useful life of a rolling bearing based on Gaussian elastic convolution is proposed. This method takes the vibration monitoring signal and rotational speed signal of the bearing as inputs, directly extracts the degradation features with elastic scales, and can re-calibrate these degradation features based on time series information to predict the remaining useful life of the bearing; this method has the advantages of higher prediction accuracy, better stability, and stronger robustness.

[0085] The method for predicting the remaining useful life of a rolling bearing based on Gaussian elastic convolution provided in this Example 2 includes the following steps:

[0086] 1) Obtain the original vibration signal and rotational speed signal of the rolling bearing under variable rotational speed conditions Among them, represents the original vibration signal obtained at the t-th sampling moment, represents the original rotational speed signal obtained at the t-th sampling moment, Q is the number of signal samples, H1 is the number of data points included in each signal sample, and C1 is the number of vibration acceleration sensors;

[0087] 2) Preprocess the original vibration signal and rotational speed signal: First, perform Z-score normalization on the vibration monitoring signal x t , and then embed the effective time information data before the sampling moment into the current data through the time window embedding strategy. The specific steps are as follows:

[0088] 2.1) Perform normalization on the vibration monitoring signal x t through Z-score to obtain the preprocessed monitoring data x Z t ; The expression of the Z-score standardization operation is as follows:

[0089]

[0090] In the formula, x t is the original signal; is the mean of the original signal; σ xt is the standard deviation of the original signal;

[0091] 2.2) Set a time window of size S, and integrate the data {x Z t , v t} and the monitoring signal samples obtained at the previous S - 1 sampling moments to obtain the signal {x' t , v' t}, where v' t = {v t-S+1 , v t -S+2 , …, v t};

[0092] 3) Construct S parallel and independent representation learning paths corresponding to the monitoring signals at S time steps in the signal {x' t , v' t}. They have the same network structure and hyperparameter settings. The data at each time step is respectively input into the corresponding representation learning path. Each representation learning path is obtained by cross - stacking N Gaussian elastic convolutional layers and N max - pooling layers, and extracts the degradation features with elastic scales in the input data. The specific steps are as follows:

[0093] 3.1) The rotational speed of the bearing changes at any time in industrial applications. This time - varying characteristic directly affects the distribution of key degradation features on the time scale. The Gaussian elastic convolutional layer is used to automatically extract the degradation features with scale changes from the variable - speed monitoring data, overcoming the defect that the traditional convolution operation cannot process the variable - speed monitoring data with a fixed convolution kernel size. Specifically, in the Gaussian elastic convolutional layer, a linear combination of the partial derivatives of the Gaussian kernel is used to approximate the standard convolution kernel to avoid manually setting the size of the convolution kernel. The one - dimensional Gaussian kernel function and its p - th derivative are as follows:

[0094]

[0095]

[0096] In the formula, σ is the standard deviation of the Gaussian kernel function, that is, the scale. In particular, due to the derivative operation leading to a high computational complexity, when solving the derivative of the Gaussian kernel function, the Hermite Polynomials are used to calculate recursively. Then the p - th derivative of the one - dimensional Gaussian kernel function can be expressed as follows:

[0097]

[0098] In the formula, H p(·) represents the Hermite polynomial of order p. If extended from one dimension to two dimensions, the derivative of the two-dimensional Gaussian kernel function can be obtained by the product of the partial derivatives in two dimensions, i.e.:

[0099]

[0100] Specify the discrete set of Gaussian partial derivative function kernels as φ = {G p,q (x, y, σ)|0 ≤ p + q ≤ K}, where |φ| = M, and M is the number of Gaussian partial derivative function kernels used for linear combination; use the self-learnable linear combination of Gaussian partial derivative function kernels in φ to replace the traditional convolution kernel to implement Gaussian elastic convolution operation, and elastically adjust the scale of feature learning through the self-learnable parameter σ and the adaptive linear combination weights, thus avoiding the manual setting of the traditional convolution kernel size. Assume that the input sequence of the l-th Gaussian elastic convolution layer is Then the n-th feature map of the l-th Gaussian elastic convolution layer can be calculated by the following formula:

[0101]

[0102]

[0103] In the formula, H and W are the height and width of the input data respectively; C is the number of input channels; in particular, when I l-1 represents the input data of each representation learning path, H is the number of data points contained in each signal sample; C is the number of vibration acceleration sensors; the value of W is 1; that is, H = H1, C = C1, W = 1; represents the output of the Gaussian elastic convolution operation; δ(·) is the rectified linear unit (ReLU) activation function; * represents the convolution operator; represents the n-th Gaussian elastic convolution kernel, which is obtained by the linear combination of M Gaussian partial derivative function kernels, and is the self-learnable weight coefficient of each Gaussian partial derivative function kernel; is the bias vector. Through the self-learnable parameter σ and the Gaussian elastic convolution layer adaptively and elastically performs representation learning at different scales, enabling the network to effectively extract complete and valuable bearing degradation information from variable-speed vibration signals;

[0104] 3.2) Input the output of the l-th Gaussian elastic convolution layer into the max-pooling layer, obtain the maximum value of the elements in the non-overlapping pooling region, and get the pooled feature The calculation formula is as follows:

[0105]

[0106] Wherein, maxpool(·) represents the max pooling function; p is the pooling size; s is the pooling stride;

[0107] 3.3) Repeat steps 3.1) and 3.2) N times in each representation learning path, that is, cross-stack N Gaussian elastic convolutional layers and N max pooling layers, and then apply the flattening function to the feature map output by the last layer to obtain the degraded feature H with elastic scale i , specifically, for the i-th representation learning path, from the variable speed vibration signal The extracted degraded feature H i The calculation expression is as follows:

[0108] H i = flatten(P i N )

[0109] Wherein, flatten(·) represents the flattening operation function; P i N Is the output feature of the last pooling layer;

[0110] 4) Construct a time information embedding layer to introduce the bearing speed information and bearing operation time information into the representation learning, so that the network can perform feature recalibration according to different degradation rates, and provide comprehensive time series information for the subsequent regression analysis of remaining life prediction. The specific steps of the information aggregation layer are as follows:

[0111] 4.1) Based on the bearing L 10 Life calculation formula and exponential degradation model, calculate the calibration coefficient corresponding to each representation learning path. Specifically, the calibration coefficient α i corresponding to the i-th representation learning path is calculated as follows:

[0112]

[0113]

[0114] Wherein, v i and t i Are respectively the i-th bearing speed value and working time within a time window of size S, Is the bearing L i at the speed v 10 Life, F C Is the basic dynamic load rating, F P Is the equivalent dynamic bearing load, and r is the life index (3 for ball bearings and 10 / 3 for roller bearings);

[0115] 4.2) Based on the calibration coefficient calculated in 4.1) For the degenerate features with elastic scales obtained for each representation learning path Perform feature recalibration, and the calculation expression is as follows:

[0116]

[0117] In the formula, represents the recalibrated degenerate features, represents the element-wise product;

[0118] 5) Concatenate the S recalibrated degenerate features head-to-tail to obtain the vector H, and the calculation expression is as follows:

[0119]

[0120] In the formula, represents the vector concatenation operation;

[0121] 6) Input the vector H into the fully connected layer to obtain the predicted remaining useful life preRUL of the bearing corresponding to the input data {x' t , v' t}, and the calculation expression is as follows:

[0122] preRUL = δ(W F H + b F )

[0123] In the formula, W F represents the weight matrix; b F represents the bias vector; δ(·) represents the linear rectification unit activation function;

[0124] 7) Based on the Adam optimization algorithm, repeat steps 3), 4), 5) and 6), set the number of iterations E, and iteratively update the parameters of the Gaussian elastic convolution network to obtain the optimal rolling bearing remaining useful life prediction model, that is, minimize the mean square error objective function:

[0125]

[0126] In the formula, actRUL i is the true remaining useful life value of the sample; B represents the number of mini-batch samples;

[0127] 8) Input the preprocessed rolling bearing vibration and rotational speed monitoring signals {x' t , v' t} into the optimal remaining useful life prediction model to predict the remaining useful life of the rolling bearing.

[0128] In this Embodiment 2, the effectiveness of the above method is verified based on the variable-speed accelerated degradation experimental dataset of rolling bearings. The variable-speed accelerated life experimental dataset adopted in this embodiment includes a total of 3 subsets, corresponding to the load condition settings of three different radial forces, namely 10 kN, 11 kN, and 12 kN. Among them, the 10-kN load contains the monitoring signals of the full life cycle of 4 rolling bearings, and the remaining load conditions contain the monitoring signals of the full life cycle of 5 rolling bearings. The monitoring signal of each bearing contains a rotational speed signal and two vibration signals. During the data acquisition process, the rotational speed is randomly changed every 25 to 35 minutes, and the rotational speed value is randomly set to one of 2100 rpm, 2400 rpm, and 2700 rpm. An example of the change of the bearing rotational speed with the running time is as Figure 2 shown. When using the method of the present invention to predict the remaining life of the roller bearing, in the data subset of the 10-kN load condition, the data of the first 3 rolling bearings are used as the training dataset, and the data of the last 1 rolling bearing are used as the test dataset. In the data subsets of the 11-kN and 12-kN load conditions, the data of the first 4 rolling bearings are used as the training dataset, and the data of the last 1 rolling bearing are used as the test dataset. The dataset division is shown in Table 1.

[0129] The hyperparameter settings of the Gaussian elastic convolution network are shown in Table 2. Using the method of the present invention to predict the remaining life of the rolling bearing under variable-speed conditions, the prediction results of test bearings 3-5 are as Figure 3 shown. As can be seen from Figure 3 , although the deviation between the true life and the predicted life of the rolling bearing is large in the early stage, with the passage of time, the predicted life of the rolling bearing gradually approaches the true life, which indicates that the method of the present invention can effectively predict the remaining life of the rolling bearing. In order to verify the effectiveness of the Gaussian elastic convolution layer and the information embedding layer in the method of the present invention, an ablation test is carried out on the method of the present invention. In Method 1, the information embedding layer is not used, and the standard convolution layer is used to extract the degradation features; in Method 2, only the Gaussian elastic convolution layer is used for representation learning, and the information embedding layer is not used for feature recalibration; in Method 3, only the information embedding layer is used without using the Gaussian elastic convolution layer. The convergence index (Convergence) and the cumulative relative accuracy (Cumulative Relative Accuracy, CRA) on the three test bearings are as Figure 4As shown, it can be seen that the method of the present invention obtains higher CRA values and lower convergence values compared with other comparative methods, indicating that the introduction of the Gaussian elastic convolution layer and the information embedding layer can improve the performance of the prediction network. At the same time, Method 2 and Method 3 obtain higher CRA values and lower convergence values than Method 1, once again verifying the advantages of Gaussian elastic convolution in multi-scale feature extraction, and the recalibration of degraded features using the information embedding layer helps to improve the performance of the prediction network. To further verify the superiority of the present invention, the method of the present invention is compared with the remaining useful life prediction method based on the gated recurrent unit (Gate Recurrent Unit, GRU) and the remaining useful life prediction method based on the multi-scale convolutional neural network (Multi-scale Convolutional Neural Network, MCNN). The three methods are evaluated using the cumulative relative accuracy and the root mean square error (Root Mean Square Error, RMSE) prediction performance metrics, and the results are shown in Table 3. As can be seen from Table 3, in the remaining useful life prediction of the three test bearings, the method of the present invention obtains the highest CRA value and the lowest RMSE value, indicating that the remaining useful life prediction method of the present invention has higher accuracy, better stability, and stronger robustness.

[0130] Table 1

[0131]

[0132] Table 2

[0133]

[0134] Table 3

[0135]

[0136] Through the above remaining useful life prediction results of the rolling bearing and the comparison of the prediction performance with the two methods, it can be found that the method of the present invention uses the Gaussian elastic convolution layer to directly extract high-level typical features with elastic scales from the variable-speed vibration signal, and uses the information embedding layer to embed time series information for feature recalibration, and then can accurately mine the degradation information most relevant to the health status of the rolling bearing, effectively improving the accuracy of the remaining useful life prediction of the rolling bearing and obtaining more excellent performance.

[0137] Example 3

[0138] This Example 3 provides a non-transitory computer-readable storage medium for storing computer instructions, which when executed by a processor, implement the above-mentioned method for predicting the remaining useful life of a rolling bearing based on Gaussian elastic convolution. The method includes:

[0139] Obtain the vibration signal and rotational speed signal of the rolling bearing under variable rotational speed conditions;

[0140] Process the obtained vibration signal and rotational speed signal using a pre-trained life prediction model to obtain the life prediction result of the rolling bearing;

[0141] Among them, the training of the life prediction model includes:

[0142] Obtain the vibration signal and rotational speed signal of the rolling bearing under variable rotational speed conditions and perform preprocessing;

[0143] Construct multiple parallel and independent representation learning paths corresponding to the monitoring signals of time steps in the preprocessed signal vibration signal and rotational speed signal, and extract the degradation features with elastic scale in the input data;

[0144] Construct a time information embedding layer, introduce the bearing rotational speed information and bearing operation time information into the representation learning, recalibrate the degradation features according to different degradation rates, and obtain the time series information for regression analysis of remaining life prediction;

[0145] Concatenate the recalibrated degradation features head to tail to obtain a concatenated vector, input the concatenated vector into a fully connected layer to obtain the predicted value of the remaining life of the bearing corresponding to the input data; based on the Adam optimization algorithm, iteratively update the model parameters to obtain the optimal rolling bearing remaining life prediction model.

[0146] Embodiment 4

[0147] This Embodiment 4 provides a computer program product, including a computer program, which when running on one or more processors, is used to implement the above-mentioned method for predicting the remaining life of a rolling bearing based on Gaussian elastic convolution. The method includes:

[0148] Obtain the vibration signal and rotational speed signal of the rolling bearing under variable rotational speed conditions;

[0149] Process the obtained vibration signal and rotational speed signal using a pre-trained life prediction model to obtain the life prediction result of the rolling bearing;

[0150] Among them, the training of the life prediction model includes:

[0151] Obtain the vibration signal and rotational speed signal of the rolling bearing under variable rotational speed conditions and perform preprocessing;

[0152] Construct multiple parallel and independent representation learning paths corresponding to the monitoring signals of time steps in the preprocessed signal vibration signal and rotational speed signal, and extract the degradation features with elastic scale in the input data;

[0153] Construct a time information embedding layer, introduce the bearing rotation speed information and bearing operation time information into the representation learning, recalibrate the degradation features according to different degradation rates, and obtain the time series information for regression analysis of remaining life prediction;

[0154] Concatenate the recalibrated degradation features at the head and tail to obtain a concatenated vector, input the concatenated vector into a fully connected layer to obtain the predicted value of the remaining life of the bearing corresponding to the input data; Based on the Adam optimization algorithm, iteratively update the model parameters to obtain an optimal rolling bearing remaining life prediction model.

[0155] Embodiment 5

[0156] This Embodiment 5 provides an electronic device, including: a processor, a memory, and a computer program; wherein, the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device runs, the processor executes the computer program stored in the memory, so that the electronic device executes the instructions for implementing the above-mentioned rolling bearing remaining life prediction method based on Gaussian elastic convolution, and this method includes:

[0157] Obtain the vibration signal and rotation speed signal of the rolling bearing under variable rotation speed conditions;

[0158] Use the pre-trained life prediction model to process the obtained vibration signal and rotation speed signal to obtain the life prediction result of the rolling bearing;

[0159] Among them, the training of the life prediction model includes:

[0160] Obtain the vibration signal and rotation speed signal of the rolling bearing under variable rotation speed conditions and perform preprocessing;

[0161] Construct multiple parallel and independent representation learning paths, corresponding to the monitoring signals of time steps in the preprocessed signal vibration signal and rotation speed signal, and extract the degradation features with elastic scale in the input data;

[0162] Construct a time information embedding layer, introduce the bearing rotation speed information and bearing operation time information into the representation learning, recalibrate the degradation features according to different degradation rates, and obtain the time series information for regression analysis of remaining life prediction;

[0163] Concatenate the recalibrated degradation features at the head and tail to obtain a concatenated vector, input the concatenated vector into a fully connected layer to obtain the predicted value of the remaining life of the bearing corresponding to the input data; Based on the Adam optimization algorithm, iteratively update the model parameters to obtain an optimal rolling bearing remaining life prediction model.

[0164] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0165] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate a means for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0166] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including an instruction means that implements the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0167] These computer program instructions can also be loaded onto a computer or other programmable data processing device to perform a series of operation steps on the computer or other programmable device to generate a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0168] Although the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, it is not a limitation on the protection scope of the present invention. Those skilled in the art should understand that based on the technical solutions disclosed in the present invention, various modifications or deformations that can be made by those skilled in the art without creative efforts should be covered within the protection scope of the present invention.

Claims

1. A method for predicting the remaining useful life of a rolling bearing based on Gaussian elastic convolution, characterized in that Including: Obtain the vibration signal and rotational speed signal of the rolling bearing under variable rotational speed conditions; Process the obtained vibration signal and rotational speed signal using a pre-trained life prediction model to obtain the life prediction result of the rolling bearing; Among them, the training of the life prediction model includes: Obtain the vibration signal and rotational speed signal of the rolling bearing under variable rotational speed conditions and perform preprocessing; Construct multiple parallel and independent representation learning paths corresponding to the monitoring signals of the time steps in the preprocessed signal vibration signal and rotational speed signal, and extract the degradation features with elastic scales in the input data; Construct a time information embedding layer, introduce the bearing rotational speed information and bearing operation time information into the representation learning, recalibrate the degradation features according to different degradation rates, and obtain the time series information for regression analysis of remaining life prediction; Concatenate the recalibrated degradation features head to tail to obtain a concatenated vector, input the concatenated vector into a fully connected layer to obtain the predicted value of the remaining life of the bearing corresponding to the input data; based on the Adam optimization algorithm, iteratively update the model parameters to obtain the optimal rolling bearing remaining life prediction model.

2. The method for predicting the remaining life of a rolling bearing based on Gaussian elastic convolution according to claim 1, characterized in that Preprocessing the original vibration signal and rotational speed signal includes: performing Z-score normalization on the vibration monitoring signal x t and embedding the effective time information data before the sampling moment into the current data through a time window embedding strategy: Normalize the vibration monitoring signal x through Z-score t to obtain the preprocessed monitoring data x Z t ; The operation expression of Z-score standardization is as follows: where x t is the original signal; is the mean of the original signal; σ x t is the standard deviation of the original signal; Set a time window of size S and integrate the data {x Z t , v t} and the monitoring signal samples obtained at the previous S - 1 sampling moments to obtain the signal {x' t , v' t}, where v' t = {v t-S+1 , v t-S+2 , …, v t}.

3. The method for predicting the remaining life of a rolling bearing based on Gaussian elastic convolution according to claim 2, characterized in that Construct S parallel and independent representation learning paths, corresponding to the monitoring signals of S time steps in the signal {x' t , v' t}, which have the same network structure and hyperparameter settings. The data of each time step is respectively input into the corresponding representation learning path. Each representation learning path is obtained by cross-stacking N Gaussian elastic convolutional layers and N max pooling layers to extract the degraded features with elastic scales in the input data.

4. The method for predicting the remaining life of a rolling bearing based on Gaussian elastic convolution according to claim 3, wherein Construct a time information embedding layer, introduce the bearing rotational speed information and bearing operation time information into the representation learning, enable the network to perform feature recalibration according to different degradation rates, and provide comprehensive time series information for subsequent regression analysis of remaining life prediction. The specific steps of the information aggregation layer are as follows: Bearing-based L 10 Based on the bearing life calculation formula and the exponential degradation model, calculate the calibration coefficient corresponding to each representation learning path. The calibration coefficient α corresponding to the i-th representation learning path i The calculation expression is as follows: In the formula, v i and t i are the speed value and working time of the i-th bearing in the time window of size S, is the speed v i Lower bearing L 10 Lifespan, F C is the basic dynamic load level, F P is the equivalent dynamic bearing load, r is the life index; Based on the calculated calibration coefficient For the degradation features of the elastic scale obtained for each representation learning path Perform feature recalibration, and the calculation expression is as follows: In the formula, represents the re-scaled degraded feature, represents the element-wise product.

5. The method for predicting the remaining life of a rolling bearing based on Gaussian elastic convolution according to claim 4, wherein Concatenate the S recalibrated degraded features from beginning to end to obtain a vector H. The calculation expression is as follows: In the formula, denotes the vector concatenation operation; input the vector H into the fully connected layer to obtain the predicted value preRUL of the remaining bearing life corresponding to the input data {x' t , v' t}. The calculation expression is as follows: preRUL = δ(W F H + b F ); in the formula, W F denotes the weight matrix; b F denotes the bias vector; δ(·) denotes the linear rectification unit activation function.

6. The method for predicting the remaining life of a rolling bearing based on Gaussian elastic convolution according to claim 5, wherein Based on the Adam optimization algorithm, set the number of iterations E, and iteratively update the parameters of the Gaussian elastic convolution network to obtain the optimal rolling bearing remaining life prediction model, that is, minimize the mean square error objective function: where actRUL i is the true remaining life value of the sample; B represents the number of small-batch samples.

7. A rolling bearing remaining life prediction system based on Gaussian elastic convolution, characterized in that, Including: An acquisition module for obtaining the vibration signal and rotational speed signal of the rolling bearing under variable rotational speed conditions; A prediction module for processing the obtained vibration signal and rotational speed signal using a pre-trained life prediction model to obtain the life prediction result of the rolling bearing; Among them, the training of the life prediction model includes: Obtain the vibration signal and rotational speed signal of the rolling bearing under variable rotational speed conditions and perform preprocessing; Construct multiple parallel and independent representation learning paths corresponding to the monitoring signals of the time steps in the preprocessed signal vibration signal and rotational speed signal, and extract the degradation features with elastic scales in the input data; Construct a time information embedding layer, introduce the bearing rotational speed information and bearing operation time information into the representation learning, recalibrate the degradation features according to different degradation rates, and obtain the time series information for regression analysis of remaining life prediction; Concatenate the recalibrated degradation features head to tail to obtain a concatenated vector, input the concatenated vector into a fully connected layer to obtain the predicted value of the remaining life of the bearing corresponding to the input data; based on the Adam optimization algorithm, iteratively update the model parameters to obtain the optimal rolling bearing remaining life prediction model.

8. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by a processor, the method for predicting the remaining life of a rolling bearing based on Gaussian elastic convolution as described in any one of claims 1-6 is implemented.

9. A computer program product, characterized in that, It includes a computer program which, when running on one or more processors, is used to implement the rolling bearing remaining life prediction method based on Gaussian elastic convolution as described in any one of claims 1-6.

10. An electronic device, characterized in that, It includes: a processor, a memory, and a computer program; wherein, the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device runs, the processor executes the computer program stored in the memory so that the electronic device executes the instructions for implementing the rolling bearing remaining life prediction method based on Gaussian elastic convolution as described in any one of claims 1-6.

Images