Fan blade fault diagnosis and early warning method based on multi-source data fusion

Abstract

The invention relates to the field of fan fault diagnosis, in particular to a fan blade fault diagnosis and early warning method based on multi-source data fusion, which comprises the steps of data acquisition, data preprocessing, feature extraction, data fusion and early warning strategy formulation. By fusing the vibration data, the strain data and the image data, the operation state of the fan blade can be reflected more comprehensively and accurately, the accuracy of fault diagnosis and the timeliness of early warning are remarkably improved, the false alarm rate and the missing report rate are reduced, unnecessary shutdown overhaul is reduced, and the availability rate and the power generation efficiency of a wind turbine generator are improved. Potential fault hidden dangers can be found in an early stage, powerful support is provided for preventive maintenance, and the maintenance cost and the operation and maintenance risk are reduced. Operation management of the wind power plant is optimized, and the overall economic benefit and market competitiveness are improved.

Description

Technical Field

[0001] The present invention relates to the field of fan fault diagnosis, and in particular to a fan blade fault diagnosis and early warning method based on multi-source data fusion. Background Art

[0002] With the rapid development of the wind power industry, the unit capacity of wind turbines has gradually increased. As one of the key components of wind turbines, the length of wind turbine blades has gradually increased from 15m in the early days to 148m. In recent years, blade breakage failures have occurred frequently, which affects the operation of the main engine and even causes the risk of tower collapse. As the core component of a wind turbine, the operating status of wind turbine blades is directly related to the efficiency and safety of the entire unit. However, due to factors such as harsh working environment and complex and variable loads, wind turbine blades are prone to various failures, such as cracks, wear, and imbalance.

[0003] Traditional wind turbine blade fault diagnosis methods usually rely on a single type of data, which results in low diagnostic accuracy and untimely warnings. For example, relying solely on vibration data may lead to misjudgment due to external interference, using strain data alone may make it difficult to capture early subtle damage, and image data alone may be limited by shooting angles and environmental factors. Summary of the invention

[0004] In view of the problems existing in the background technology, a wind turbine blade fault diagnosis and early warning method based on multi-source data fusion is proposed. By fusing vibration data, strain data and image data, it can more comprehensively and accurately reflect the operating status of wind turbine blades, significantly improve the accuracy of fault diagnosis and the timeliness of early warning, reduce the false alarm rate and missed alarm rate, reduce unnecessary shutdowns for maintenance, and improve the availability and power generation efficiency of wind turbines.

[0005] The present invention proposes a fan blade fault diagnosis and early warning method based on multi-source data fusion, the steps comprising:

[0006] S1. Data acquisition: collecting vibration data, strain data and image data;

[0007] S2, data preprocessing: preprocessing operations are performed on the collected multi-source data respectively;

[0008] S3, feature extraction: extract features from vibration data, strain data and image data respectively;

[0009] S4, data fusion: adopt fusion strategy to fuse multi-source features to form a comprehensive feature vector;

[0010] S5. Formulation of early warning strategy: Based on the output results of the fault diagnosis model, combined with the operating conditions of the wind turbine blades and the preset safety threshold, formulate a reasonable early warning strategy.

[0011] Preferably, during data preprocessing, all multi-source data acquisition devices are time-calibrated with the server, and time tags are marked during data acquisition and storage.

[0012] Preferably, during data preprocessing, the vibration data and strain data are subjected to denoising and smoothing to remove outliers; the image data are cropped, scaled, and grayscaled to unify the image size and format, and denoising is performed.

[0013] Preferably, the vibration data feature extraction method is as follows:

[0014] S31, time domain feature extraction: extract effective value and peak value;

[0015] S32, frequency domain feature extraction: perform fast Fourier transform to extract the main frequency and frequency bandwidth.

[0016] Preferably, the strain data feature extraction method is as follows:

[0017] S33, obtaining strain peak value and valley value: determining the maximum value and the minimum value in the strain data;

[0018] S34, obtaining the strain change rate: calculating the change rate of the strain over time, reflecting the dynamic change of the strain;

[0019] S35. Obtain strain distribution characteristics: analyze the distribution of strain at different positions of the blade.

[0020] Preferably, the image data feature extraction method is as follows:

[0021] S36, performing image segmentation and extraction;

[0022] S37, extracting shape features;

[0023] S38, extracting texture features;

[0024] S39. Extract edge features.

[0025] Preferably, a fusion strategy based on weighted average is selected during data fusion, and different weights are assigned according to the importance of different data types to fault diagnosis. The specific steps are as follows:

[0026] S41, determining weights;

[0027] S42, weighted calculation;

[0028] S43, use the validation set data to evaluate the fusion results;

[0029] S44, outputting the optimized fusion features for subsequent wind turbine blade fault diagnosis and early warning;

[0030] S45. Result analysis and application: analyze and process the fused eigenvalues.

[0031] Preferably, the specific steps of early warning strategy formulation are as follows:

[0032] S51, setting a warning threshold;

[0033] S52, weight allocation;

[0034] S53, fusion computing;

[0035] S54, warning level classification;

[0036] S55, real-time monitoring and updating;

[0037] S56. Modify the early warning strategy based on the working conditions;

[0038] S57, manual intervention and verification.

[0039] Compared with the prior art, the present invention has the following beneficial technical effects:

[0040] By integrating vibration data, strain data and image data, the present invention can more comprehensively and accurately reflect the operating status of wind turbine blades, significantly improve the accuracy of fault diagnosis and the timeliness of early warning, reduce the false alarm rate and missed alarm rate, reduce unnecessary shutdowns and maintenance, and improve the availability and power generation efficiency of wind turbines. It can detect potential fault hazards early, provide strong support for preventive maintenance, and reduce maintenance costs and operation and maintenance risks. It helps to optimize the operation and management of wind farms and improve overall economic benefits and market competitiveness. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a flow chart of the fan blade fault diagnosis and early warning method based on multi-source data fusion in the present invention;

[0042] Figure 2 It is a data preprocessing flow chart in the present invention;

[0043] Figure 3 This is a flow chart of vibration data feature extraction in the present invention;

[0044] Figure 4 It is a flow chart of strain data feature extraction in the present invention;

[0045] Figure 5 This is a flow chart of image feature extraction in the present invention. DETAILED DESCRIPTION

[0046] Embodiment 1

[0047] like Figure 1As shown, the present invention proposes a fan blade fault diagnosis and early warning method based on multi-source data fusion, the steps comprising:

[0048] S1. Data collection: by installing corresponding sensors on the wind turbine blades, vibration data and strain data are collected, and high-definition cameras are used to obtain blade image data;

[0049] S2. Data preprocessing: Perform preprocessing operations such as noise reduction and normalization on the collected multi-source data to improve data quality and availability; perform denoising and smoothing on vibration data and strain data to remove outliers; perform operations such as cropping and scaling on image data to unify image size and format, and perform denoising;

[0050] S3, feature extraction: extract vibration amplitude, frequency and other features from vibration data, extract strain distribution, change trend and other features from strain data, extract crack length, wear area and other features from image data;

[0051] S4, data fusion: adopt weighted average fusion strategy to fuse the multi-source features extracted above to form a comprehensive feature vector;

[0052] S5. Formulation of early warning strategy: Based on the output results of the fault diagnosis model, combined with the operating conditions of the wind turbine blades and the preset safety threshold, formulate a reasonable early warning strategy; when the model predicts that the blades may fail, a warning signal is issued in a timely manner.

[0053] like Figure 2 As shown in the figure, during data preprocessing, multi-source data acquisition devices are time-calibrated with the server, and time tags are marked during data acquisition and storage. De-noising and smoothing are performed on vibration data and strain data to remove outliers; image data is cropped, scaled, and grayed to unify image size and format, and denoising is performed.

[0054] like Figure 3 As shown, the vibration data feature extraction method is as follows:

[0055] S31, time domain feature extraction: extract the effective value (RMS) and peak value;

[0056] Effective value (RMS): indicates the energy level of the vibration signal; Peak value: the maximum value in the signal;

[0057] S32, frequency domain feature extraction: perform fast Fourier transform (FFT) to extract the main frequency and frequency bandwidth;

[0058] Fast Fourier Transform (FFT): Convert the time domain signal to the frequency domain to obtain the spectrum;

[0059] Main frequency: the frequency component with the largest energy in the spectrum;

[0060] Frequency bandwidth: The frequency range that contains a certain energy ratio (such as 90%).

[0061] like Figure 4 As shown in Figure 2, the strain data feature extraction method is as follows:

[0062] S33, obtaining strain peak value and valley value: determining the maximum value and the minimum value in the strain data;

[0063] S34, obtaining the strain change rate: calculating the change rate of the strain over time, reflecting the dynamic change of the strain;

[0064] S35. Obtaining strain distribution characteristics: analyzing the distribution of strain at different positions of the blade.

[0065] like Figure 5 As shown, the image data feature extraction method is as follows:

[0066] S36, performing image segmentation extraction: separating the leaf image from the background;

[0067] S37, extracting shape features: calculating geometric parameters such as the length, width, and area of ​​the blade;

[0068] S38, extracting texture features: using gray level co-occurrence matrix (GLCM) to extract texture features such as contrast, correlation, and energy;

[0069] S39. Extract edge features: Apply the Canny edge detection algorithm to extract edge features such as the number and length of edges.

[0070] When fusion data is used, a fusion strategy based on weighted average is selected, and different weights are assigned according to the importance of different data types to fault diagnosis. The specific steps are as follows:

[0071] S41, determining weights: determining weight values ​​for vibration data, strain data, and image data respectively according to the importance, reliability, and contribution to fault diagnosis of the data; these weight values ​​are usually between 0 and 1, and their sum is 1;

[0072] S42, weighted calculation: multiply the extracted features by the corresponding weights respectively, sum the weighted features, and obtain the fused feature value;

[0073] S43. Use the validation set data to evaluate the fusion results: use indicators such as accuracy and F1 value for evaluation;

[0074] S44, outputting the optimized fusion features for subsequent wind turbine blade fault diagnosis and early warning;

[0075] For example, suppose the eigenvalue of vibration data is V and the weight is w 1 ; The characteristic value of the strain data is S, and the weight is w 2 ; The eigenvalue of the image data is I, and the weight is w 3 The fused eigenvalue F 1 It can be calculated as: F 1 =w 1 *V+w 2 *S+w 3 *I;

[0076] S45. Result analysis and application: analyzing and processing the fused eigenvalues, for example, inputting them into a fault diagnosis model to diagnose and predict wind turbine blade faults.

[0077] F 1 The value is a comprehensive indicator used to evaluate the performance of a classification model, which combines precision and recall. Precision refers to the proportion of samples predicted to be positive that are actually positive; recall refers to the proportion of samples predicted to be positive that are actually positive.

[0078] F 1 The value is calculated as: 1 =2*(precision*recall) / (precision+recall);

[0079] F 1 The value range is between 0 and 1. 1 The higher the value, the better the performance of the model.

[0080] Suppose we want to determine whether a blade image contains a crack (positive example), and the model predicts 100 cracks, of which 80 actually have cracks (true positive examples, TP), and 20 actually do not have cracks (false positive examples, FP). In fact, there are 90 images with cracks (true positive examples + false negative examples, TP+FN).

[0081] Precision = TP / (TP+FP) = 80 / (80+20) = 0.8

[0082] Recall = TP / (TP+FN) = 80 / 90≈0.889

[0083] F 1 Value = 2*(0.8*0.889) / (0.8+0.889)≈0.842.

[0084] The specific steps for developing an early warning strategy are as follows:

[0085] S51. Set warning thresholds: Based on historical data and experimental analysis, determine the normal range and abnormal thresholds for each data source.

[0086] For example, the effective value of vibration data exceeds a certain value, the strain value of strain data exceeds a specific range, or the length of cracks detected in image data reaches a certain extent;

[0087] S52. Weight allocation: different weights are allocated to vibration data, strain data, and image data according to the sensitivity and importance of the data source to the fault; for example, if vibration data is more sensitive to certain specific faults, it may be given a higher weight;

[0088] S53, fusion calculation: Use weighted average or other fusion algorithms to fuse multi-source data to obtain comprehensive evaluation indicators;

[0089] S54, Warning level classification: According to the numerical range of the fusion results, four warning levels are divided, for example, mild warning, moderate warning and severe warning;

[0090] S55. Real-time monitoring and updating: Acquire new data in real time and continuously update fusion calculations and early warning judgments. As data accumulates, regularly re-evaluate and adjust early warning thresholds and weights;

[0091] S56. Combine the working conditions and consider the operating conditions of the fan, such as speed, power, ambient temperature and other factors, and modify the early warning strategy. The same data may have different meanings under different working conditions;

[0092] S57, Manual intervention and verification: For the issued warning, manual intervention and verification are allowed to ensure the accuracy of the warning. At the same time, the warning strategy is further optimized based on the results of manual verification.

[0093] For example, when the fused indicator exceeds the mild warning threshold, the system sends a notification to the operation and maintenance personnel, prompting them to pay attention to the blade status; when it reaches the moderate warning threshold, more frequent monitoring is arranged; when it exceeds the severe warning threshold, the machine is immediately shut down for maintenance.

[0094] Embodiment 2

[0095] This embodiment selects the wind turbine blade No. 1 in the wind farm as the research object. Vibration sensors, strain sensors and cameras are installed at the key parts of the blades. Then, the collected multi-source data is subjected to pre-processing operations such as denoising and normalization to eliminate noise and data deviation. Fast Fourier transform (FFT) is used to extract vibration spectrum characteristics, strain change trends, sound energy distribution and other features that can effectively characterize the operating status of the wind turbine blades from the pre-processed data. The multi-source features extracted above are fused to form a comprehensive feature vector by using a data fusion algorithm based on weighted average. The algorithm achieves effective integration of multi-source information by calculating the trust and uncertainty of each feature. After multiple iterative training, the training is completed when the accuracy and recall rate of the model on the validation set reach a stable and better level. Finally, the trained model is evaluated using the test set data to ensure that the model can accurately identify the crack fault of the blade. After actual operation verification, the method of the present invention can detect the crack fault of the blade 3 days in advance with an accuracy rate of 92%. Through continuous iterative training, the performance indicators of the model (such as accuracy, F1 value, etc.) are monitored on the validation set. When the performance of the model on the validation set no longer improves, the training is stopped to obtain the optimal model parameters. Finally, the trained model is evaluated on the test set. In practical applications, accurate warning of blade wear failures was successfully achieved 5 days in advance, with a diagnostic accuracy rate of up to 96%.

[0096] The present invention fuses vibration data, strain data and image data, wherein:

[0097] Vibration data: The vibration characteristics of the blades can sensitively reflect structural changes and faults. For example, problems such as imbalance, looseness or breakage will cause changes in the vibration mode. By analyzing parameters such as vibration frequency, amplitude and phase, blade faults can be effectively diagnosed.

[0098] Strain data: Strain measurement can directly reflect the stress and deformation of the blade during operation. When the blade has structural damage, the strain distribution will change significantly, which is very valuable for early detection of fatigue cracks and structural strength problems of the blade.

[0099] Image data: By acquiring the surface image of the blade through a high-definition camera, visible damage such as cracks, corrosion, and wear on the blade surface can be intuitively detected. Image data can also be used to monitor the shape changes and attachments of the blade.

[0100] Therefore, the present invention can more comprehensively and accurately reflect the operating status of wind turbine blades, significantly improving the accuracy of fault diagnosis and the timeliness of early warning. Compared with traditional methods, it has the following advantages: it reduces the false alarm rate and missed alarm rate, reduces unnecessary shutdowns and maintenance, and improves the availability and power generation efficiency of wind turbines. It can detect potential fault hazards early, provide strong support for preventive maintenance, and reduce maintenance costs and operation and maintenance risks. It helps to optimize the operation and management of wind farms and improve overall economic benefits and market competitiveness.

[0101] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto, and various changes can be made within the knowledge scope of technicians in the relevant technical field without departing from the purpose of the present invention.

Claims

1. A wind turbine blade fault diagnosis and early warning method based on multi-source data fusion, characterized in that the steps include: S1. Data acquisition: collecting vibration data, strain data and image data; S2, data preprocessing: preprocessing operations are performed on the collected multi-source data respectively; S3, feature extraction: extract features from vibration data, strain data and image data respectively; S4, data fusion: adopt fusion strategy to fuse multi-source features to form a comprehensive feature vector; S5. Formulation of early warning strategy: Based on the output results of the fault diagnosis model, combined with the operating conditions of the wind turbine blades and the preset safety threshold, formulate a reasonable early warning strategy.

2. The wind turbine blade fault diagnosis and early warning method based on multi-source data fusion according to claim 1 is characterized in that: During data preprocessing, multi-source data acquisition devices are time-calibrated with the server, and time tags are marked during data acquisition and storage.

3. The wind turbine blade fault diagnosis and early warning method based on multi-source data fusion according to claim 1 is characterized in that: During data preprocessing, the vibration data and strain data are denoised and smoothed to remove outliers; the image data is cropped, scaled, and grayscaled to unify the image size and format, and denoised.

4. The wind turbine blade fault diagnosis and early warning method based on multi-source data fusion according to claim 1 is characterized in that: The vibration data feature extraction method is as follows: S31, time domain feature extraction: extract effective value and peak value; S32. Frequency domain feature extraction: Perform fast Fourier transform to extract the main frequency and frequency bandwidth.

5. The wind turbine blade fault diagnosis and early warning method based on multi-source data fusion according to claim 1 is characterized in that: The strain data feature extraction method is as follows: S33, obtaining strain peak value and valley value: determining the maximum value and the minimum value in the strain data; S34, obtaining the strain change rate: calculating the change rate of the strain over time, reflecting the dynamic change of the strain; S35. Obtaining strain distribution characteristics: analyzing the distribution of strain at different positions of the blade.

6. The wind turbine blade fault diagnosis and early warning method based on multi-source data fusion according to claim 1 is characterized in that: The image data feature extraction method is as follows: S36, performing image segmentation and extraction; S37, extracting shape features; S38, extracting texture features; S39. Extract edge features.

7. The wind turbine blade fault diagnosis and early warning method based on multi-source data fusion according to claim 1 is characterized in that: When fusion data is used, a fusion strategy based on weighted average is selected, and different weights are assigned according to the importance of different data types to fault diagnosis. The specific steps are as follows: S41, determining weights; S42, weighted calculation; S43, use the validation set data to evaluate the fusion results; S44, outputting the optimized fusion features for subsequent wind turbine blade fault diagnosis and early warning; S45. Result analysis and application: analyze and process the fused eigenvalues.

8. The wind turbine blade fault diagnosis and early warning method based on multi-source data fusion according to claim 1 is characterized in that: The specific steps for developing an early warning strategy are as follows: S51, setting a warning threshold; S52, weight allocation; S53, fusion computing; S54, warning level classification; S55, real-time monitoring and updating; S56. Modify the early warning strategy based on the working conditions; S57, manual intervention and verification.

Images