Method for predicting residual life of wind power rolling bearing by applying incremental mapping and recursive estimation

Abstract

The invention discloses a method for predicting the residual life of a wind power rolling bearing by applying incremental mapping and recursive estimation. The invention discloses a method for predicting the residual life of a wind power rolling bearing, and relates to renewable energy equipment state monitoring. The method comprises the following steps: firstly, obtaining a multiband envelope demodulation two-dimensional spectrogram by converting a vibration acceleration signal, calculating and normalizing the variation of the spectrogram, and obtaining a health index and a normalized degradation increment; secondly, extracting features from two dimensions of the two-dimensional spectrogram by using a channel attention module, and designing nonlinear mapping of a two-dimensional convolutional neural network learning spectrogram and a degradation increment; and then, establishing a recursive deep learning model, and calculating a posterior probability of a future health index in combination with model weight Gaussian sampling to realize rolling prediction. And finally, predicting the residual life according to the failure threshold. The method has the advantages that the health index based on the change increment of the multiband envelope demodulation two-dimensional spectrogram is constructed, the two-dimensional convolutional neural network containing the channel attention module is designed, recursive deep learning and weight Gaussian sampling are fused, and the prediction precision of the residual life of the wind power bearing is improved.

Description

Technical Field

[0001] The invention relates to the field of power generation equipment state monitoring and remaining life prediction, and in particular to a method for predicting the remaining life of a wind power rolling bearing by using incremental mapping and recursive estimation. Background Art

[0002] Rolling bearings are widely used in the main shaft, gearbox, generator and other systems of wind turbine transmission systems. They are in harsh operating environments for a long time and are affected by factors such as alternating loads and random excitations. They are prone to failure and continuous degradation, resulting in a gradual decline in performance. If bearing failure is not predicted in time and appropriate maintenance is not performed, it will lead to unexpected shutdown of wind turbines or even catastrophic accidents. Deep learning algorithms have powerful feature extraction and nonlinear expression capabilities and are widely used in the field of remaining life prediction of bearings and other electromechanical equipment. However, existing research has shortcomings in incremental mapping and recursive estimation. Most normalized health indicators do not reflect the dynamic degradation process of bearings. Traditional recursive deep learning cannot calculate the posterior probability of the degradation state, which affects the accuracy and robustness of wind turbine bearing remaining life prediction. Summary of the invention

[0003] Based on the above problems and the defects of the prior art, the present invention proposes a method for predicting the remaining life of wind power rolling bearings using incremental mapping and recursive estimation, which mainly predicts the remaining life by estimating the development trend of health indicators.

[0004] The present invention comprises the following steps:

[0005] Step 1: Based on the wavelet packet algorithm, several filter bands with different center frequencies and bandwidths are divided. The vibration signal is demodulated by narrowband envelope for different frequency bands to obtain envelope spectra. The multi-band envelope demodulation two-dimensional spectra are formed by splicing them in order from low to high according to the center frequencies of each filter for degradation information extraction.

[0006] Step 2: Use the change in the two-dimensional spectrum at the previous and next moments to construct the dynamic degradation increment. The calculation expression is:

[0007] X={x1,x2,...,x K}

[0008]

[0009] Where X represents the spectrum of the bearing degradation process over its entire life, K is the number of sampling moments of the bearing degradation process over its entire life, and Y is (a) represents absolute increment, |.| represents absolute value, and sum(.) represents the sum of all elements of the matrix.

[0010] Step 3: Reduce the influence of each bearing's inherent characteristics on the construction of the health index by subtracting the average value of the healthy stage from the absolute increment. The calculation expression is:

[0011]

[0012] In the formula, FDT represents the degradation starting point, Y (f) represents the baseline increment, and mean(.) represents the average value.

[0013] Step 4: Normalize the health index according to the degradation stage, and construct the normalized health index and the corresponding normalized increment.

[0014] Step 5: The convolutional neural network first uses the channel attention module to perform squeeze and excitation attention operations on the frequency band and frequency dimensions of the two-dimensional spectrum, respectively, to obtain weight vectors, which are multiplied with the original spectrum in turn to obtain the weighted spectrum, and then convolution is performed to extract deep degradation features, learn the nonlinear mapping relationship between features and degradation increments, and realize incremental mapping estimation.

[0015] Step 6: The recursive neural network first uses the training set data to train the network weights, and then calculates the likelihood estimate of the health indicator at future moments through Gaussian sampling of the network weights to obtain the posterior probability of the health indicator.

[0016] Step 7: Obtain the posterior estimate of the health indicator at the future moment through recursive rolling prediction, and predict the remaining life by combining the failure threshold. Its calculation expression is:

[0017]

[0018] Where, t now represents the current sampling time, FT represents the failure threshold, x t represents the system state at sampling time t.

[0019] The beneficial effects of the present invention are as follows: a method for constructing a health indicator based on the incremental change of a two-dimensional spectrum of multi-band envelope demodulation is proposed, a two-dimensional convolutional neural network including a channel attention module is designed to learn the nonlinear mapping relationship of features, and a recursive deep learning model and weighted Gaussian sampling are integrated to enhance the rolling prediction capability of the model, thereby improving the remaining life prediction accuracy of wind turbine bearings. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A flowchart showing the remaining life prediction method for wind turbine rolling bearings based on incremental mapping and recursive estimation.

[0021] Figure 2 It represents the degradation dynamic normalized health index and the corresponding degradation increment constructed by this method.

[0022] Figure 3 Represents the attention convolutional neural network structure constructed by this method.

[0023] Figure 4 A graph showing the remaining life prediction results of an instance. DETAILED DESCRIPTION

[0024] A preferred embodiment of the present invention is described in detail below in conjunction with the accompanying drawings so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby making a clearer and more explicit definition of the protection scope of the present invention.

[0025] like Figure 1 As shown, it is a flow chart of the remaining life prediction method of wind power rolling bearing based on incremental mapping and recursive estimation. The data of the embodiment comes from the vibration acceleration signal of the high-speed shaft measuring point of a 2.0MW wind turbine in a wind farm. The data sampling frequency is 25600Hz, the sampling time is 1.28s, the sampling interval is 1 day, and the duration from health to degradation and then to failure is 187 days.

[0026] According to the embodiment, the implementation process of the present invention includes the following steps:

[0027] S1: Divide the filter frequency bands into several ones with different center frequencies and bandwidths, perform narrowband envelope demodulation on the vibration signal for different frequency bands to obtain envelope spectra, and splice them in order from low to high according to the center frequencies of the filters to form a multi-band envelope demodulation two-dimensional spectrum.

[0028] S2: Use the changes in the two-dimensional spectrum at the previous and next moments to construct the dynamic degradation increment, and accumulate the dynamic health index.

[0029] S3: Normalize the health index according to the degradation stage, and construct the normalized health index and the corresponding normalized increment, such as Figure 2 shown.

[0030] S4: Design a two-dimensional convolutional neural network to extract the features of the two-dimensional spectrum and learn the nonlinear mapping relationship between the features and the degradation increment. The network structure is as follows Figure 3 shown.

[0031] S5: Build a recursive deep learning model and use the training set data to train the network weights.

[0032] S6: Fusion network weight Gaussian sampling is used to calculate the posterior probability of health indicators at future moments and implement rolling prediction of degradation states.

[0033] S7: Combine the recursive posterior estimation result of the degradation state at the future moment and the preset failure threshold to predict the remaining life.

[0034] The detailed description of the above steps S1-S7 is as follows:

[0035] Step 1: Construct a multi-band envelope demodulation two-dimensional spectrum:

[0036] Divide the filter band: First, divide several filter bands with different center frequencies and bandwidths based on the wavelet packet algorithm. The setting of these bands should cover the possible fault frequency range of rolling bearings.

[0037] Narrowband envelope demodulation: For each filter frequency band, the vibration signal is demodulated by narrowband envelope to obtain the envelope spectrum. The envelope spectrum can reflect the vibration characteristics of the rolling bearing in the corresponding frequency band.

[0038] Splicing two-dimensional spectra: According to the order of the center frequencies of each filter from low to high, the envelope spectra obtained are spliced ​​to form a multi-band envelope demodulation two-dimensional spectrum. This two-dimensional spectrum can fully reflect the vibration characteristics of the rolling bearing and provide a basis for the subsequent degradation increment calculation.

[0039] Step 2: Build dynamic health indicators:

[0040] Calculate degradation increment: Use the change of the two-dimensional spectrum at the previous and next time to construct the dynamic degradation increment. Specifically, calculate the difference of the two-dimensional spectrum at adjacent time, and take the absolute value and sum it to get the absolute increment.

[0041] Cumulative health index: The absolute increments are accumulated to obtain the dynamic health index. This index can reflect the degree of degradation of the rolling bearing and gradually increases as the degradation process proceeds.

[0042] Step 3: Normalize health indicators:

[0043] Determine the degradation starting point: During the degradation process of the rolling bearing, determine a degradation starting point, which is usually selected at the moment when the performance of the rolling bearing begins to decline significantly.

[0044] Calculate the baseline increment: Calculate the average absolute increment in the healthy stage before the degradation starting point as the baseline increment.

[0045] Normalization: The dynamic health index is subtracted from the baseline increment and normalized to obtain the normalized health index. At the same time, the normalized increment is calculated to reflect the degradation rate of the rolling bearing.

[0046] Step 4: Design a 2D convolutional neural network:

[0047] Feature extraction: A two-dimensional convolutional neural network is used to extract features from the multi-band envelope demodulation two-dimensional spectrogram. The convolutional layer in the network can learn the deep degradation features in the spectrogram.

[0048] Incremental mapping estimation: Design the network structure so that it can learn the nonlinear mapping relationship between features and degraded increments. By training the network, incremental mapping estimation is achieved, that is, predicting the normalized increment based on the extracted features.

[0049] Step 5: Build a recursive deep learning model:

[0050] Build recursive networks: Build recursive deep learning models, such as long short-term memory networks (LSTM) or gated recurrent units (GRU), to capture the temporal dependencies in the rolling element bearing degradation process.

[0051] Training network weights: Using the training set data, the network weights are trained through the back-propagation algorithm, so that the recursive network can accurately predict health indicators at future moments.

[0052] Step 6: Implement rolling prediction of degradation state:

[0053] Fused Gaussian sampling: During the training of the recurrent network, the network weights are fused with Gaussian sampling to simulate the network output under different parameter combinations. This helps to enhance the robustness and generalization ability of the model.

[0054] Calculate the posterior probability: Use network weight Gaussian sampling to calculate the posterior probability distribution of health indicators at future moments. This can reflect the uncertainty of the prediction results and provide a reliable basis for rolling prediction.

[0055] Rolling Forecast: Rolling forecast is achieved based on the health indicators at the current moment and the predicted health indicators at the future moment. In the rolling forecast process, the forecast model is continuously updated with new observation data to improve the accuracy of the forecast.

[0056] Step 7: Predict the remaining life:

[0057] Set failure threshold: According to the failure mechanism and performance degradation law of rolling bearings, a failure threshold is set. When the predicted health index is lower than the threshold, the rolling bearing is considered to fail.

[0058] Calculate the remaining life: Combine the recursive posterior estimation results of the degradation state at the future moment and the preset failure threshold to calculate the remaining life of the rolling bearing. Specifically, the remaining life is determined based on the time difference between the predicted health indicator change trend and the failure threshold.

[0059] The following is a detailed description of each step:

[0060] S1: Divide the filter band and generate a two-dimensional spectrum

[0061] First, it is necessary to divide several filter bands with different center frequencies and bandwidths based on the wavelet packet algorithm. In this embodiment, the vibration signal is divided into 8 different frequency bands, each with a different center frequency and bandwidth. The setting of these frequency bands is determined based on the analysis of the characteristic frequency of rolling bearing faults and prior knowledge.

[0062] Next, narrowband envelope demodulation is performed on the vibration signal for different frequency bands. Narrowband envelope demodulation is a signal processing technology used to extract the envelope components in the rolling bearing vibration signal, which contain the fault information of the rolling bearing. By performing narrowband envelope demodulation on the vibration signal of each frequency band, the corresponding envelope spectrum can be obtained.

[0063] Finally, the envelope spectra of each frequency band are spliced ​​together in the order of the filter center frequency from low to high to form a multi-band envelope demodulation two-dimensional spectrum. This two-dimensional spectrum contains rich information about the rolling bearing during the full life degradation process, which can be used for subsequent degradation analysis and remaining life prediction.

[0064] S2: Building dynamic degradation increments and health indicators

[0065] After obtaining the two-dimensional spectrum, it is necessary to use the change in the two-dimensional spectrum at the previous and next moments to construct the dynamic degradation increment. Specifically, the absolute increment between the two-dimensional spectra at two adjacent moments can be calculated, that is, the sum of the absolute values ​​of the differences between the corresponding elements of the two-dimensional spectra at the two moments. This absolute increment reflects the degree of degradation of the rolling bearing at the current moment relative to the previous moment.

[0066] The dynamic degradation increment is constructed by using the change of the two-dimensional spectrum at the previous and next moments. The calculation expression is:

[0067] X={x1,x2,...,x K}

[0068]

[0069] Where X represents the spectrum of the bearing degradation process over its entire life, K is the number of sampling moments of the bearing degradation process over its entire life, and Y is (a) represents absolute increment, |.| represents absolute value, and sum(.) represents the sum of all elements of the matrix.

[0070] Then, these absolute increments are accumulated to obtain the dynamic health index. This dynamic health index reflects the cumulative degradation of the rolling bearing from the healthy state to the current moment. It should be noted that during the accumulation process, the absolute increments need to be normalized to eliminate the dimension differences and value range differences between different frequency bands.

[0071] S3: Normalized health indicators and increments

[0072] After obtaining the dynamic health index, the health index needs to be normalized for the degradation stage. The purpose of normalization is to map the value range of the health index to a fixed range for subsequent prediction and analysis. In this embodiment, the value range of the health index is selected to be mapped to the range of [0,1].

[0073] By subtracting the average value of the healthy stage from the absolute increment, the influence of the inherent characteristics of each bearing on the construction of the health index is reduced, and the calculation expression is:

[0074]

[0075] In the formula, FDT represents the degradation starting point, Y (f) represents the baseline increment, and mean(.) represents the average value.

[0076] Specifically, the average health index value of the rolling bearing in the healthy stage can be calculated as the benchmark value. Then, the health index value at each moment is compared with the benchmark value, and the relative value is calculated. Finally, these relative values ​​are normalized to obtain the normalized health index.

[0077] At the same time, the increment of the normalized health index can also be calculated, that is, the difference between the normalized health indexes at two adjacent moments. This increment reflects the degradation rate of the rolling bearing at the current moment relative to the previous moment.

[0078] S4: Designing a 2D Convolutional Neural Network

[0079] In order to learn the nonlinear mapping relationship between the two-dimensional spectrogram features and the degradation increment, a two-dimensional convolutional neural network (CNN) needs to be designed. This network can automatically extract the deep features in the two-dimensional spectrogram and learn the mapping relationship between these features and the degradation increment.

[0080] In this embodiment, the designed two-dimensional CNN structure is as follows Figure 3 As shown in Figure 1, the network consists of multiple convolutional layers, pooling layers, fully connected layers, and output layers. The convolutional layer is used to extract local features in the two-dimensional spectrogram; the pooling layer is used to reduce the dimension of the feature map and reduce the amount of calculation; the fully connected layer is used to map the extracted features to the output space of the degraded increment; and the output layer is used to output the prediction results.

[0081] In addition, a channel attention module is introduced to enhance the network's attention to important features in the two-dimensional spectrogram. The channel attention module can perform squeeze and excitation attention operations on the two dimensions of the band and frequency of the two-dimensional spectrogram, obtain weight vectors respectively, and multiply them with the original spectrogram in turn to obtain the weighted spectrogram. In this way, the network can pay more attention to those features that have an important impact on the degradation increment prediction.

[0082] S5: Building a recursive deep learning model

[0083] After obtaining the two-dimensional CNN, a recursive deep learning model needs to be established to predict health indicators at future moments. Recursive neural network (RNN) is a neural network structure suitable for processing sequence data, which can capture the temporal dependencies in sequence data.

[0084] In this embodiment, a long short-term memory network (LSTM) is selected as the basic structure of the recursive deep learning model. LSTM is a special RNN structure that overcomes the shortcomings of traditional RNN in the long-term dependency problem by introducing a gating mechanism and a memory unit.

[0085] The training set data is used to train the weights of the LSTM network. During the training process, the features extracted by the two-dimensional CNN are used as the input of the LSTM network, and the normalized health indicators are used as the output. By optimizing the network weights, the LSTM network can learn the mapping relationship between features and health indicators.

[0086] In order to improve the generalization ability and prediction accuracy of the LSTM network, the weighted Gaussian sampling method is also used. Specifically, during the training process, the weights of the LSTM network are sampled with a Gaussian distribution to simulate the network output under different parameter combinations. In this way, multiple prediction results can be obtained, and the mean and variance of these results can be calculated to obtain the posterior probability distribution of health indicators at future moments.

[0087] S6: Implementing rolling prediction of degradation status

[0088] After obtaining the posterior probability distribution of the health index at the future moment, the rolling prediction of the degradation state can be realized. Specifically, the rolling prediction error can be calculated based on the health index at the current moment and the predicted health index at the future moment, and the prediction model can be adjusted according to the error.

[0089] The rolling prediction process is iterative. In each iteration, the latest observation data and prediction results are used to update the prediction model and calculate the predicted value of the health indicator at the next moment. In this way, the real degradation process can be gradually approached and the accuracy and reliability of the prediction can be improved.

[0090] In addition, the uncertainty of the rolling forecast can be calculated based on the results of the rolling forecast. Uncertainty reflects the reliability and confidence of the forecast results. By comparing the forecast uncertainty at different times, the stability and robustness of the forecast model can be evaluated.

[0091] S7: Predicting remaining life

[0092] Finally, the remaining life of the rolling bearing can be predicted by combining the recursive posterior estimation result of the degradation state at the future moment and the preset failure threshold. Specifically, the predicted health index at the future moment can be compared with the failure threshold. When the predicted health index is lower than the failure threshold, it is considered that the rolling bearing will fail. At this time, the remaining life can be calculated based on the time difference between the current moment and the failure moment.

[0093] It should be noted that the determination of the failure threshold is based on the failure mechanism and performance degradation law of the rolling bearing. In practical applications, it is necessary to select an appropriate failure threshold based on the specific type and working environment of the rolling bearing.

[0094] In addition, the results of rolling prediction can be used to update the failure threshold. As the rolling bearing degradation process continues, its failure mechanism may change. Therefore, it is necessary to dynamically adjust the failure threshold based on the latest observation data and prediction results to ensure the accuracy and reliability of the prediction results.

[0095] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for predicting the remaining life of a wind turbine rolling bearing using incremental mapping and recursive estimation, characterized in that: The method specifically comprises the following steps: Step 1: Divide the filter bands with different center frequencies and bandwidths, perform narrowband envelope demodulation on the vibration signal for different frequency bands to obtain envelope spectra, and splice them in order from low to high according to the center frequencies of the filters to form a multi-band envelope demodulation two-dimensional spectrum; Step 2: Use the changes in the two-dimensional spectrum at the previous and next moments to construct the dynamic degradation increment, and accumulate the dynamic health index; Step 3: Normalize the health index according to the degradation stage, and construct the normalized health index and the corresponding normalized increment; Step 4: Design a two-dimensional convolutional neural network to extract the features of the two-dimensional spectrum and learn the nonlinear mapping relationship between the features and the degradation increment; Step 5: Establish a recursive deep learning model and use the training set data to train the network weights; Step 6: Fusion network weight Gaussian sampling, calculate the posterior probability of health indicators at future moments, and achieve rolling prediction of degradation status; Step 7: Combine the recursive posterior estimation results of the degradation state at the future time and the preset failure threshold to predict the remaining life.

2. A method for predicting the remaining life of a wind turbine rolling bearing by using incremental mapping and recursive estimation according to claim 1, characterized in that: In step 1, a number of filter bands with different center frequencies and bandwidths are divided based on the wavelet packet algorithm, and the vibration signal is demodulated by narrowband envelope demodulation for different frequency bands to obtain envelope spectra. The multi-band envelope demodulation two-dimensional spectrogram is formed by splicing them in order from low to high according to the center frequencies of each filter, which is used for degradation information extraction.

3. The method for predicting the remaining life of a wind turbine rolling bearing by using incremental mapping and recursive estimation according to claim 1, characterized in that: In step 2, the dynamic degradation increment is constructed using the change in the two-dimensional spectrum at the previous and next moments. The calculation expression is: X={x1,x2,...,x K } Where X represents the spectrum of the bearing degradation process over its entire life, K is the number of sampling moments of the bearing degradation process over its entire life, and Y is (a) represents absolute increment, |.| represents absolute value, and sum(.) represents the sum of all elements of the matrix.

4. The method for predicting the remaining life of a wind turbine rolling bearing by using incremental mapping and recursive estimation according to claim 1, characterized in that: In step 3, the influence of the inherent characteristics of each bearing on the construction of the health index is reduced by subtracting the average value of the healthy stage from the absolute increment. The calculation expression is: In the formula, FDT represents the degradation starting point, Y (f) represents the baseline increment, and mean(.) represents the average value.

5. The method for predicting the remaining life of a wind turbine rolling bearing by using incremental mapping and recursive estimation according to claim 1, characterized in that: The health indicators are normalized according to the degradation stage, and the normalized health indicators and the corresponding normalized increments are constructed.

6. The method for predicting the remaining life of a wind turbine rolling bearing by using incremental mapping and recursive estimation according to claim 1, characterized in that: The convolutional neural network first uses the channel attention module to perform squeeze and excitation attention operations on the frequency band and frequency dimensions of the two-dimensional spectrum, respectively, to obtain weight vectors, which are multiplied with the original spectrum in turn to obtain the weighted spectrum, and then convolution is performed to extract deep degradation features, learn the nonlinear mapping relationship between features and degradation increments, and realize incremental mapping estimation.

7. The method for predicting the remaining life of a wind turbine rolling bearing by using incremental mapping and recursive estimation according to claim 1, characterized in that: The recursive neural network first uses the training set data to train the network weights, and then calculates the likelihood estimate of the health indicator at future moments through Gaussian sampling of the network weights to obtain the posterior probability of the health indicator.

Images