A wind power rolling bearing residual life prediction method based on transfer learning

Abstract

The application discloses a wind power rolling bearing residual life prediction method based on transfer learning. The method first uses an envelope demodulation algorithm to process a vibration acceleration signal, screens a demodulation feature with good monotonicity, and constructs a bearing health index. Then, through a minimum distance sample pairing algorithm, target domain degradation samples are paired with source domain full life samples. Then, a deep convolution transfer neural network is constructed, paired sample learning feature extraction and mapping are used, and bearing health state estimation is realized. Finally, the state estimation result is fitted and combined with a failure threshold to predict the residual life. The application has the beneficial effects that an effective health index construction method is proposed, a precise sample pairing algorithm is designed, a high-performance state estimation model is established, and the prediction accuracy of the wind power bearing residual life is significantly improved.

Description

Technical Field

[0001] This invention relates to the field of power generation equipment condition monitoring and remaining life prediction, and in particular to a method for predicting the remaining life of wind turbine rolling bearings based on transfer learning. Background Technology

[0002] Rolling bearings, as supporting components of rotating machinery, are widely used in the drive chains and other rotating parts of wind turbine generators. The operational performance of rolling bearings has a significant impact on the safe operation of wind turbine generators. Failure to predict rolling bearing failure in a timely manner and to perform appropriate maintenance can lead to unexpected shutdowns of wind turbine generators, or even catastrophic accidents. Therefore, conducting health status assessments and predicting the remaining life of rolling bearings in wind turbine generators is of great importance. However, existing research methods have limitations in transfer learning prediction. Transfer sample pairing does not adequately consider sample similarity and does not incorporate cross-domain sample similarity into transfer learning, which is detrimental to the accurate prediction of remaining life. Summary of the Invention

[0003] Based on the above problems and the shortcomings of existing technologies, this invention proposes a method for predicting the remaining life of wind turbine rolling bearings based on transfer learning.

[0004] This invention includes the following steps:

[0005] Step 1: Divide the signal into several filtering frequency bands based on the wavelet packet algorithm, and demodulate the demodulation features of the vibration signal in each filtering frequency band based on the envelope algorithm to obtain demodulation features at different scales.

[0006] Step 2: Select demodulation features with good monotonicity as monotonic features, and use a monotonicity metric to measure the monotonicity of the demodulation features. The calculation expression is:

[0007]

[0008] In the formula, Mon(n,α) is a measure of the monotonicity of the envelope spectrum energy characteristics at sampling time n and scale α, and F Y (m,α) represents the envelope spectrum energy characteristics at sampling time m and scale α, and ε(.) is a piecewise function.

[0009] Step 3: Integrate and normalize the degradation increments of monotonic features to construct a bearing health index with good monotonicity. Establish a degradation increment accumulation algorithm, with the calculation expression as follows:

[0010]

[0011] In the formula, D i (n) is the degradation increment of the i-th monotonic feature at sampling time n, F Z(n,i) is the envelope spectrum energy of the i-th monotonic feature at sampling time n, D(n) is the degradation increment of all monotonic features at sampling time n, FDS is the degradation starting point, and I is the number of monotonic features.

[0012] Step 4: Normalize the fused degradation increments to construct a health indicator with good monotonicity.

[0013] Step 5: Calculate the cross-domain Euclidean distance between each degraded sample in the target domain and the full-lifetime samples in the source domain. The calculation expression is as follows:

[0014]

[0015] In the formula, K is the number of sample dimensions, (x1, x2, ..., x...). K ) and (y1,y2,…,y K ) are the elements of the sample matrices of the target domain and the source domain, respectively.

[0016] Step 6: Search for the source domain sample that has the smallest distance to each degenerate sample in the target domain, and form a sample pair.

[0017] Step 7: Construct a deep transfer neural network, train it using paired samples, learn feature extraction and mapping, and achieve bearing health status estimation.

[0018] Step 8: Fit the state estimation results and combine them with the failure threshold to predict the remaining lifetime. The calculation expression is as follows:

[0019] RUL(t n =inf{t:HI(t) n +t)≥FT|t n}

[0020] In the formula, RUL(t) n ) is the sampling time t n The remaining lifespan, t n At the current moment, HI(t) n +t) is the sampling time t n +t is a health indicator, and FT is the failure threshold.

[0021] The beneficial effects of this invention are as follows: it proposes a method for constructing bearing health indicators with good monotonicity, designs a sample pairing algorithm based on minimum distance, establishes a state estimation method based on deep convolutional transfer neural network, and improves the prediction accuracy of wind power bearing remaining life. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating a method for predicting the remaining life of wind turbine rolling bearings based on transfer learning.

[0023] Figure 2 This indicates the results of the health indicator construction.

[0024] Figure 3 This represents the structure of a deep transfer neural network.

[0025] Figure 4 This is a graph showing the predicted remaining lifetime of the instance. Detailed Implementation

[0026] A preferred embodiment of the present invention will now be described in detail with reference to 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 providing a clearer and more explicit definition of the scope of protection of the present invention.

[0027] The figure shows a flowchart of a method for predicting the remaining life of wind turbine rolling bearings based on transfer learning. The data for this example comes from vibration acceleration signals from high-speed shaft measuring points of a 2.0MW wind turbine generator set in a wind farm. The data sampling frequency is 25600Hz, the sampling duration is 1.28s, the sampling interval is 1 day, and the recorded duration from healthy to degradation to failure is 212 days.

[0028] According to the embodiments, the implementation process of the present invention includes the following steps:

[0029] S1: Demodulation features at different scales are obtained by demodulating vibration signals based on wavelet packet algorithm and envelope algorithm.

[0030] S2: Select demodulation features with good monotonicity as monotonic features.

[0031] S3: Integrate and normalize the degradation increments of monotonic features to construct a bearing health index with good monotonicity, such as... Figure 2 As shown.

[0032] S4: Calculate the cross-domain Euclidean distance between each degraded sample in the target domain and the full-lifetime sample in the source domain.

[0033] S5: Search for the source domain sample that has the smallest distance to each degraded sample in the target domain, and form a sample pair.

[0034] S6: Construct a deep transfer neural network, such as Figure 3 As shown, the bearing health status is estimated by using paired samples for training to learn feature extraction and mapping.

[0035] S7: Fit the state estimation results and combine them with the failure threshold to predict the remaining lifetime, such as... Figure 4 As shown.

[0036] The following is a detailed explanation of each step:

[0037] S1: Demodulation features at different scales are obtained by demodulating vibration signals based on wavelet packet algorithm and envelope algorithm.

[0038] In wind turbine generators, the vibration signals of rolling bearings contain rich state information, and extracting this information is crucial for predicting the remaining service life of the bearings. To effectively extract this information from the vibration signals, a demodulation method combining wavelet packet algorithm and envelope algorithm can be used.

[0039] 1.1 Wavelet packet algorithm.

[0040] The wavelet packet algorithm is an extension of the wavelet transform, enabling more detailed analysis of the frequency components of a signal. It divides the frequency band into multiple levels, thus capturing the characteristics of different frequency bands more accurately. In the vibration signal analysis of rolling bearings, the wavelet packet algorithm can divide the signal into multiple filter bands, with the signal within each band reflecting the vibration characteristics of the bearing under different conditions.

[0041] The wavelet packet algorithm first selects a suitable wavelet basis function, and then decomposes the original signal into multiple sub-bands by performing multiple convolution and downsampling operations. The signal in each sub-band contains information within a specific frequency range, which is of great significance for subsequent feature extraction and state monitoring.

[0042] 1.2 Envelope Algorithm.

[0043] Envelope algorithms are a commonly used signal demodulation technique that extracts the envelope of a signal to reflect changes in its amplitude. In the vibration signals of rolling bearings, the envelope signal can more intuitively reflect the bearing's vibration state.

[0044] To obtain the envelope signal, a Hilbert transform is typically performed on the original signal. The Hilbert transform is a linear transform that converts a real-valued signal into an analytic signal. The real part of the analytic signal is the original signal, and the imaginary part is the result of the Hilbert transform. By calculating the magnitude of the analytic signal, the envelope of the original signal can be obtained.

[0045] In the vibration signal analysis of rolling bearings, envelope demodulation of the signal within each filtered frequency band yields demodulation characteristics at different scales. These demodulation characteristics reflect the vibration properties of the bearing under different conditions, providing a foundation for subsequent health index construction and remaining life prediction.

[0046] S2: Select demodulation features with good monotonicity as monotonic features.

[0047] After obtaining demodulation features at different scales, these features need to be screened to identify monotonic features that can reflect the bearing degradation process. Monotonic features refer to features whose values ​​exhibit monotonical changes (such as monotonically increasing or monotonically decreasing) as the bearing degradation process progresses.

[0048] 2.1 Monotonicity measurement.

[0049] To quantify the monotonicity of a feature, a monotonicity measure can be introduced. A monotonicity measure is a mathematical expression used to calculate the degree of monotonicity of a feature value as it changes over time. The monotonicity measure of the feature value at each sampling time can be calculated, and then these values ​​can be statistically analyzed to assess the monotonicity of the feature.

[0050] The formula for calculating monotonicity can be constructed based on the relative rate of change or the absolute amount of change of eigenvalues. For example, the difference between eigenvalues ​​at adjacent sampling times can be calculated, and the sign of these differences can be statistically analyzed to assess the monotonicity of the feature. Furthermore, smoothing algorithms can be introduced to reduce the impact of noise on the monotonicity measure.

[0051] 2.2 Feature Filtering.

[0052] After obtaining the monotonicity measure of the demodulation features, monotonic features can be filtered based on the magnitude of the monotonicity measure value. A threshold can be set, and features with a monotonicity measure value greater than this threshold can be considered monotonic features. These monotonic features can more accurately reflect the degradation process of bearings, providing a reliable basis for subsequent health indicator construction and remaining life prediction.

[0053] It's important to note that feature selection is a complex process that requires consideration of multiple factors. Besides monotonicity metrics, factors such as feature stability, repeatability, and sensitivity can also be considered to comprehensively evaluate feature quality. Through comprehensive selection, a set of high-quality monotonic features can be obtained for subsequent health indicator construction and remaining life expectancy prediction.

[0054] S3: Integrate and normalize the degradation increments of monotonic features to construct a bearing health index with good monotonicity.

[0055] After obtaining the monotonic features, these features need to be further processed to construct a bearing health index with good monotonicity. The health index is a numerical indicator that reflects the health status of a bearing; it changes monotonically as the bearing degrades.

[0056] 3.1 Calculation of degradation increment.

[0057] To construct health indicators, it is first necessary to calculate the degradation increment of monotonic features. The degradation increment refers to the change in feature value between adjacent sampling times. By calculating the degradation increment, minute changes in feature value over time can be captured, thus more accurately reflecting the degradation process of the bearing.

[0058] The degradation increment of each monotonic feature at each sampling time can be calculated, and these degradation increments can be accumulated to obtain the total degradation increment at each sampling time. The total degradation increment reflects the degree of degradation of the bearing at the current sampling time, providing a basis for the construction of subsequent health indicators.

[0059] 3.2 Construction of health indicators.

[0060] After obtaining the total degradation increment, it can be normalized to construct a bearing health index with good monotonicity. Normalization is a commonly used data preprocessing technique that can scale data to a specific range (such as between 0 and 1), thereby eliminating dimensional and numerical differences between different features.

[0061] The total degradation increment can be divided by a baseline value (such as the maximum degradation increment or the initial degradation increment) to obtain a normalized health index. This health index changes monotonically as the bearing degradation process progresses, providing a more intuitive reflection of the bearing's health status.

[0062] S4: Calculate the cross-domain Euclidean distance between each degraded sample in the target domain and the full-lifetime sample in the source domain.

[0063] In transfer learning, to transfer knowledge from the source domain to the target domain, it is necessary to find the similarity between samples in the target domain and samples in the source domain. This similarity can be measured by calculating the distance between samples. In this invention, Euclidean distance is used to calculate the cross-domain distance between each degraded sample in the target domain and the full-lifetime sample in the source domain.

[0064] 4.1 Euclidean distance calculation.

[0065] Euclidean distance is a commonly used distance metric that calculates the straight-line distance between two samples in a multidimensional space. In this invention, the feature vectors of the target domain sample and the source domain sample are used as coordinate points in a multidimensional space, and then the Euclidean distance between these coordinate points is calculated.

[0066] 4.2 The significance of cross-domain distance.

[0067] Calculating the cross-domain distance between target domain samples and source domain samples aims to find the source domain sample most similar to the target domain sample. This similarity metric is fundamental to transfer learning because it allows knowledge from the source domain to be transferred to the target domain. In this invention, by calculating the cross-domain distance, the most similar full-lifetime source domain sample can be found for each degraded sample in the target domain, thus forming sample pairings.

[0068] S5: Search for the source domain sample that has the smallest distance to each degraded sample in the target domain, and form a sample pair.

[0069] After obtaining the cross-domain distance between the target domain samples and the source domain samples, it is necessary to search for the source domain sample with the smallest distance to each degenerate sample in the target domain to form a sample pair. This sample pairing is a key step in transfer learning because it allows the model to be trained using full-lifetime data in the source domain and then applied to the degenerate samples in the target domain.

[0070] 5.1 Search for the sample with the minimum distance.

[0071] To find the source domain sample that is closest to the target domain sample, a nearest neighbor search algorithm can be used. This algorithm searches for the nearest neighbor of the target sample in a multidimensional space, i.e., the sample with the smallest distance. In this invention, this step can be implemented using existing nearest neighbor search libraries, such as the KDTree or BallTree algorithms in scikit-learn.

[0072] 5.2 The significance of sample pairing.

[0073] By pairing samples, full-lifetime data from the source domain can be combined with degradation data from the target domain, resulting in a richer dataset. This dataset contains data on the entire process from health to degradation to failure, which can be used to train more accurate predictive models. Furthermore, sample pairing can reduce domain disparities in transfer learning and improve the model's generalization ability.

[0074] S6: Construct a deep transfer neural network, train it using paired samples, learn feature extraction and mapping, and achieve bearing health status estimation.

[0075] After obtaining the paired samples, a deep transfer neural network needs to be constructed to learn feature extraction and mapping, and to achieve bearing health status estimation. A deep transfer neural network is a neural network model that combines deep learning and transfer learning; it can utilize knowledge from the source domain to improve prediction performance in the target domain.

[0076] 6.1 Deep transfer neural network structure.

[0077] The structure of a deep transfer neural network typically includes a feature extraction layer, a transfer layer, and a prediction layer. The feature extraction layer is used to extract useful feature representations from the original input data; the transfer layer is used to transfer knowledge from the source domain to the target domain; and the prediction layer is used to predict the output in the target domain based on the extracted features and the transferred knowledge.

[0078] In this invention, the structure of the deep transfer neural network can be designed according to the specific task and dataset. For example, a convolutional neural network (CNN) can be used as a feature extraction layer to extract time-frequency features from vibration signals; a fully connected network can be used as a transfer layer to transfer knowledge from the source domain; and a regression network can be used as a prediction layer to predict the health status or remaining service life of a bearing.

[0079] 6.2 Model Training and Optimization.

[0080] After constructing the deep transfer neural network, it needs to be trained using paired samples. During training, the backpropagation algorithm can be used to optimize the network's weights and bias parameters to minimize prediction error. Furthermore, regularization techniques such as L1 / L2 regularization and dropout can be employed to prevent overfitting.

[0081] By training and optimizing a deep transfer neural network, a model capable of accurately estimating the health status of bearings can be obtained. This model can be used to monitor the operating status of wind turbine rolling bearings in real time and promptly identify potential faults.

[0082] S7: Fit the state estimation results and combine them with the failure threshold to predict the remaining lifetime.

[0083] After obtaining the estimated bearing health status, it needs to be fitted and combined with a failure threshold to predict the remaining service life. Remaining service life refers to the time or number of cycles that a bearing will take from its current state to failure. Predicting the remaining service life can provide important reference for maintenance decisions regarding wind turbine rolling bearings.

[0084] 7.1 Fitting of state estimation results.

[0085] To fit the state estimation results, several common fitting methods can be used, such as linear regression, multinomial regression, and exponential regression. These methods can construct a fitting model based on the changing trend of the health state estimate over time. In this invention, the fitting method most suitable for the data characteristics and prediction needs can be selected.

[0086] The purpose of fitting the model is to smooth the health status estimate, reduce the influence of noise and random fluctuations, and thus more accurately reflect the degradation trend of the bearing. Through fitting, a continuous health status curve can be obtained, which describes the entire process of the bearing from health to degradation and then to failure.

[0087] 7.2 Determination of failure threshold.

[0088] The failure threshold refers to the threshold at which a bearing is considered to be about to fail or requires maintenance when its health condition reaches a certain critical value. This threshold can be determined based on actual engineering needs and experience. In this invention, the failure threshold can be set by analyzing historical data and expert experience.

[0089] Once the failure threshold is determined, it can be combined with the fitted health state curve to predict the remaining service life of the bearing. The expected time point when the bearing will reach failure can be determined by finding the intersection of the health state curve and the failure threshold.

[0090] 7.3 Calculation of remaining useful life.

[0091] Once the expected failure time is determined, the remaining service life of the bearing can be calculated. The formula for calculating the remaining service life can be designed based on actual needs and the characteristics of the prediction model.

[0092] Calculating the remaining service life can provide important time references for maintenance decisions regarding wind turbine rolling bearings. For example, when the remaining service life is short, maintenance plans can be scheduled in advance to avoid unexpected downtime or accidents caused by bearing failure.

[0093] In this invention, Figure 1 The flowchart illustrates the overall process of the wind turbine rolling bearing remaining life prediction method proposed in this invention, including key steps such as signal demodulation, feature screening, health index construction, sample pairing, neural network construction, state estimation, and remaining life prediction.

[0094] Figure 2 The figure shows the results of the health index construction. It illustrates a bearing health index with good monotonicity constructed by fusing degradation increments with monotonic features and normalizing them. The monotonic trend of the health index over time can intuitively reflect the health status of the bearing, providing a reliable basis for subsequent remaining service life prediction. By observing the changes in the health index, the degree of bearing degradation and remaining service life can be preliminarily determined, providing an important reference for wind turbine maintenance decisions.

[0095] Figure 3: Deep Transfer Neural Network Structure. This figure depicts the structure of a deep transfer neural network, including key components such as feature extraction layers, transfer layers, and prediction layers. By illustrating the hierarchical structure and connections of the neural network, a deeper understanding can be gained of how this invention utilizes deep learning and transfer learning techniques to extract features, transfer knowledge, and predict states. The design of this neural network structure fully considers the characteristics of wind turbine rolling bearing vibration signals and the needs of remaining life prediction, contributing to improved prediction accuracy and generalization ability.

[0096] Figure 4 Example Remaining Life Prediction Results. This figure shows the results of predicting the remaining life of actual wind turbine rolling bearings using the method proposed in this invention. By comparing the predicted results with the actual lifespan, the effectiveness and accuracy of this invention can be verified. The example prediction results figure helps to intuitively understand the performance of this invention in practical applications, enhancing trust and understanding of the technical effects of this invention.

[0097] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. For those skilled in the art, the present invention can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting the remaining life of a wind power rolling bearing based on transfer learning, characterized in that, The method specifically comprises the following steps: Step 1: based on the wavelet packet algorithm and envelope algorithm demodulation vibration signal to get different scale demodulation characteristics; Step 2: screening demodulation characteristics with good monotonicity as monotonic characteristics, using monotonicity measure to quantify the monotonicity of demodulation characteristics, the expression is: where Mon(n, a) is a monotonicity measure of the envelope spectrum energy feature at sampling time n and scale a, F Y (m, a) is the envelope spectrum energy feature at sampling time m and scale a, and ε(u) is a piecewise function; u is the argument of the piecewise function. Step 3: fusion monotonicity of degradation increment and normalization, build bearing health index with good monotonicity, establish a degradation increment cumulative algorithm, the expression is: where D i (n) is the degradation increment of the i-th monotonic feature at sampling time n, F Z (n, i) is the envelope spectrum energy of the i-th monotonic feature at sampling time n, j is the sampling time, D(n) is the degradation increment of all monotonic features at sampling time n, FDS is the degradation starting point, and I is the number of monotonic features; Step 4: calculate the cross-domain Euclidean distance between each degradation sample in the target domain and the full life sample in the source domain; Step 5: search for the source domain sample with the minimum distance to each degradation sample in the target domain, form sample pairing; Step 6: build deep migration neural network, train using paired samples, learn feature extraction and mapping, realize bearing health state estimation; Step 7: fitting the state estimation results and combining the failure threshold to predict the remaining life.

2. The method of claim 1, wherein the method is characterized by, The normalized fusion degradation increment is used to build a health index with good monotonicity.

3. The wind power rolling bearing remaining useful life prediction method based on transfer learning according to claim 1, characterized in that, The Euclidean distance between each sample in the target domain and the full life sample in the source domain is calculated, and the source domain sample with the minimum distance to each sample in the target domain is searched to obtain sample pairing; the closer the distance between two sample features, the greater the similarity between them, and the more suitable for cross-domain migration.

4. The wind power rolling bearing remaining useful life prediction method based on transfer learning according to claim 1, characterized in that, A deep migration neural network is built to learn feature extraction and mapping, and the algorithm process is as follows: Firstly, the feature extraction and index mapping of the network are trained using the full life data of the source domain; Secondly, the hidden layer feature migration of the network is trained using the paired data of the health stage of the source domain and the target domain; then, the hidden layer feature migration of the network is trained using the paired data of the degradation stage of the source domain and the target domain; Finally, the trained neural network is used for health state estimation.

Images