Fan blade fault detection method and related device
By acquiring the sweeping sound signal generated by the rotation of wind turbine blades, performing feature processing, and combining it with a deep learning model, the problem of insufficient reliability of existing wind turbine blade fault detection technology is solved, and high-reliability, low-cost wind turbine blade fault detection is achieved.
Patent Information
- Application Number
- CN202512056786.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-31
- Publication Date
- 2026-03-17
AI Technical Summary
Existing wind turbine blade fault detection technologies are unreliable, making it difficult to accurately detect blade damage in complex environments. Furthermore, conventional methods are costly, complex to maintain, or rely on large amounts of labeled data.
By acquiring the sweeping sound signal generated by the rotation of wind turbine blades, feature processing is performed to obtain overall and local spectral features. These features are then combined with a deep learning model for wind turbine blade fault detection. The fusion of overall and local spectral features improves the reliability of the detection.
It enables highly reliable fault detection of wind turbine blades in complex environments, reduces detection costs, minimizes damage to the blade structure, and improves the real-time performance and accuracy of detection.
Smart Images
Figure CN121676286A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of equipment testing technology, and in particular to a method and related device for detecting wind turbine blade faults. Background Technology
[0002] Wind turbine blades, as a major component of wind turbines, account for approximately 15%-20% of the total cost of a wind turbine unit. Wind turbine units are often located in harsh environments, such as high mountains and oceans. The turbine blades rotate at high altitudes to drive the generator inside the nacelle to generate electricity. Due to long-term exposure to the high-altitude environment, wind turbine blades are easily damaged by airborne substances, atmospheric radiation, dust, lightning, heavy rain, and snow. Therefore, fault detection of wind turbine blades is particularly necessary.
[0003] Because detecting blade damage through regular manual inspections is inefficient, costly, and requires significant manpower and time, various wind turbine blade fault detection technologies have emerged, such as vibration detection, acoustic emission detection, fiber optic grating detection, ultrasonic detection, infrared thermal imaging detection, UAV-based image detection, audio-based detection, and neural network-based detection. However, all commonly used wind turbine blade fault detection technologies have shortcomings that can affect the reliability of fault detection.
[0004] Therefore, how to provide a method for detecting wind turbine blade faults to improve the reliability of wind turbine blade fault detection has become a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention
[0005] In view of the above problems, this application provides a method and related apparatus for detecting wind turbine blade faults, so as to improve the reliability of wind turbine blade fault detection. The specific solution is as follows:
[0006] The first aspect of this application provides a method for detecting faults in wind turbine blades, including:
[0007] Acquire the sweeping sound signal generated by the rotation of the fan blades;
[0008] The sweeping sound signal is subjected to feature processing to obtain overall spectral features and local spectral features; the overall spectral features are used to characterize the features of the complete frequency band of the preset acoustic signature spectral frequency range, and the local spectral features are used to characterize the cross-band fusion features of each frequency band corresponding to damage at different positions of the wind turbine blades in the preset acoustic signature spectral frequency range.
[0009] Based on the overall spectral characteristics and the local spectral characteristics, wind turbine blade fault detection is performed to obtain wind turbine blade fault detection results.
[0010] In one possible implementation, the determination of the local spectral features includes:
[0011] Determine the frequency bands corresponding to damage at different locations on the wind turbine blades within the preset acoustic signature spectrum frequency range;
[0012] Feature extraction processing is performed on each frequency band to obtain damage characteristics at different locations of the wind turbine blades;
[0013] The local spectral features are obtained by fusing damage characteristics at different locations on the wind turbine blades.
[0014] In one possible implementation, determining the frequency bands corresponding to damage at different locations on the wind turbine blades within a preset acoustic signature spectrum frequency range includes:
[0015] The preset acoustic spectrum frequency range is divided into multiple sub-bands;
[0016] Effective sub-frequency bands related to damage at different locations of the wind turbine blades are determined from the multiple sub-frequency bands, and these sub-frequency bands are used as the frequency bands corresponding to damage at different locations of the wind turbine blades within the preset acoustic spectrum frequency range.
[0017] In one possible implementation, the feature extraction process for each frequency band to obtain damage features at different locations on the wind turbine blades includes:
[0018] Differential features are extracted for each frequency band to obtain the differential features of each frequency band;
[0019] By mapping the differential features of each frequency band to the same dimension through linear projection, the damage features of different locations on the wind turbine blades can be obtained.
[0020] In one possible implementation, fusing the damage features at different locations on the wind turbine blades to obtain the local spectral features includes:
[0021] Calculate the attention weight by using the damage features of each wind turbine blade location as the query and the remaining wind turbine blade location damage features as the key and value;
[0022] The local spectral features are obtained by weighting and fusing the damage features at different locations of the wind turbine blades based on the attention weights.
[0023] In one possible implementation, the step of performing wind turbine blade fault detection based on the overall spectral features and the local spectral features to obtain wind turbine blade fault detection results includes:
[0024] The overall spectral features and the local spectral features are input into the wind turbine blade fault detection model. The blade fault detection model encodes the overall spectral features and the local spectral features respectively and then fuses them to obtain fused features. Based on the fused features, anomaly classification is performed to obtain the wind turbine blade fault detection result.
[0025] In one possible implementation, the training method for the wind turbine blade fault detection model includes:
[0026] Acquire training data, which includes swept air sound signal samples, the overall spectral features and local spectral features corresponding to the swept air sound signal samples, and the wind turbine blade fault detection result labels corresponding to the swept air sound signals;
[0027] Construct a training architecture for a wind turbine blade fault detection model;
[0028] The overall spectral features and the local spectral features are input together into the wind turbine blade fault detection model training architecture. The wind turbine blade fault detection model training architecture reconstructs the overall spectral features by performing directional masking based on the frequency bands corresponding to damage at different locations of the wind turbine blade, and is used for positive anomaly classification.
[0029] The wind turbine blade fault detection model training architecture is trained using the reconstruction loss of the overall spectral features and the positive anomaly classification loss as joint losses to obtain a trained wind turbine blade fault detection model.
[0030] A second aspect of this application provides a wind turbine blade fault detection device, comprising:
[0031] The acquisition unit is used to acquire the sweeping sound signal generated by the rotation of the wind turbine blades;
[0032] The feature processing unit is used to perform feature processing on the wind sweeping sound signal to obtain overall spectral features and local spectral features; the overall spectral features are used to characterize the features of the complete frequency band of the preset acoustic signature spectral frequency range, and the local spectral features are used to characterize the cross-band fusion features of each frequency band corresponding to damage at different positions of the wind turbine blades in the preset acoustic signature spectral frequency range.
[0033] The fault detection unit is used to perform wind turbine blade fault detection based on the overall spectrum characteristics and the local spectrum characteristics, and obtain the wind turbine blade fault detection result.
[0034] In one possible implementation, the feature processing unit includes:
[0035] The frequency band determination unit is used to determine the frequency bands corresponding to damage at different locations on the wind turbine blades within the preset acoustic spectrum frequency range.
[0036] The feature extraction and processing unit is used to perform feature extraction processing on each frequency band to obtain damage features at different locations of the wind turbine blades;
[0037] The feature fusion unit is used to fuse the damage features at different locations of the wind turbine blades to obtain the local spectral features.
[0038] In one possible implementation, the frequency band determination unit is specifically used for:
[0039] The preset acoustic spectrum frequency range is divided into multiple sub-bands;
[0040] Effective sub-frequency bands related to damage at different locations of the wind turbine blades are determined from the multiple sub-frequency bands, and these sub-frequency bands are used as the frequency bands corresponding to damage at different locations of the wind turbine blades within the preset acoustic spectrum frequency range.
[0041] In one possible implementation, the feature extraction processing unit is specifically used for:
[0042] Differential features are extracted for each frequency band to obtain the differential features of each frequency band;
[0043] By mapping the differential features of each frequency band to the same dimension through linear projection, the damage features of different locations on the wind turbine blades can be obtained.
[0044] In one possible implementation, the feature fusion unit is specifically used for:
[0045] Calculate the attention weight by using the damage features of each wind turbine blade location as the query and the remaining wind turbine blade location damage features as the key and value;
[0046] The local spectral features are obtained by weighting and fusing the damage features at different locations of the wind turbine blades based on the attention weights.
[0047] In one possible implementation, the fault detection unit is specifically used for:
[0048] The overall spectral features and the local spectral features are input into the wind turbine blade fault detection model. The blade fault detection model encodes the overall spectral features and the local spectral features respectively and then fuses them to obtain fused features. Based on the fused features, anomaly classification is performed to obtain the wind turbine blade fault detection result.
[0049] In one possible implementation, the device further includes: a wind turbine blade fault detection model training unit;
[0050] The wind turbine blade fault detection model training unit is specifically used for:
[0051] Acquire training data, which includes swept air sound signal samples, the overall spectral features and local spectral features corresponding to the swept air sound signal samples, and the wind turbine blade fault detection result labels corresponding to the swept air sound signals;
[0052] Construct a training architecture for a wind turbine blade fault detection model;
[0053] The overall spectral features and the local spectral features are input together into the wind turbine blade fault detection model training architecture. The wind turbine blade fault detection model training architecture reconstructs the overall spectral features by performing directional masking based on the frequency bands corresponding to damage at different locations of the wind turbine blade, and is used for positive anomaly classification.
[0054] The wind turbine blade fault detection model training architecture is trained using the reconstruction loss of the overall spectral features and the positive anomaly classification loss as joint losses to obtain a trained wind turbine blade fault detection model.
[0055] A third aspect of this application provides a computer program product including computer-readable instructions that, when executed on an electronic device, cause the electronic device to implement the wind turbine blade fault detection method described in the first aspect or any implementation thereof.
[0056] A fourth aspect of this application provides an electronic device, including at least one processor and a memory connected to the processor, wherein:
[0057] The memory is used to store computer programs;
[0058] The processor is used to execute the computer program so that the electronic device can implement the wind turbine blade fault detection method of the first aspect or any implementation thereof.
[0059] The fifth aspect of this application provides a computer-readable storage medium carrying one or more computer programs that, when executed by an electronic device, enable the electronic device to perform the wind turbine blade fault detection method described in the first aspect or any implementation thereof.
[0060] By employing the above technical solution, the wind turbine blade fault detection method and related apparatus provided in this application, after acquiring the sweeping sound signal generated by the rotation of the wind turbine blade, perform feature processing on it to obtain two dimensions of features: the overall spectral features of the preset acoustic signature frequency range and the local spectral features of the preset acoustic signature frequency range. Since the overall spectral features can characterize the features of the complete frequency band of the preset acoustic signature frequency range, and the local spectral features can characterize the cross-band fusion features of each frequency band corresponding to damage at different positions of the wind turbine blade within the preset acoustic signature frequency range, these two dimensions of features cover both the overall coarse-grained features and the fine-grained features of damage at different positions of the wind turbine blade. Therefore, by comprehensively utilizing these two dimensions of features for wind turbine blade fault detection, the reliability of wind turbine blade fault detection can be improved. Attached Figure Description
[0061] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.
[0062] Figure 1 A flowchart illustrating a wind turbine blade fault detection method provided in this application embodiment;
[0063] Figure 2 A schematic diagram of a wind turbine blade fault detection model architecture provided in this application embodiment;
[0064] Figure 3 This is a schematic diagram of a wind turbine blade fault detection model training architecture provided in an embodiment of this application;
[0065] Figure 4 This is a schematic diagram of the structure of a wind turbine blade fault detection device provided in an embodiment of this application;
[0066] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation
[0067] The embodiments of this application are described below with reference to the accompanying drawings. The terminology used in the implementation section of this application is for explaining specific embodiments only and is not intended to limit the scope of this application.
[0068] The embodiments of this application will now be described with reference to the accompanying drawings. Those skilled in the art will recognize that, with technological advancements and the emergence of new scenarios, the technical solutions provided in the embodiments of this application are equally applicable to similar technical problems.
[0069] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such terms are interchangeable where appropriate; this is merely a way of distinguishing objects with the same attributes in the embodiments of this application. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, so that a process, method, system, product, or apparatus that comprises a series of elements is not necessarily limited to those elements, but may include other elements not explicitly listed or inherent to those processes, methods, products, or apparatuses.
[0070] Wind turbine blades, as a major component of wind turbines, account for approximately 15%-20% of the total cost of a wind turbine unit. Wind turbine units are often located in harsh environments, such as high mountains and oceans. The turbine blades rotate at high altitudes to drive the generator inside the nacelle to generate electricity. Due to long-term exposure to the high-altitude environment, wind turbine blades are easily damaged by airborne substances, atmospheric radiation, dust, lightning, heavy rain, and snow. Therefore, fault detection of wind turbine blades is particularly necessary.
[0071] Because detecting blade damage through regular manual inspections is inefficient, costly, and requires a lot of manpower and time, various wind turbine blade fault detection technologies have emerged, such as vibration detection, acoustic emission detection, fiber optic grating detection, ultrasonic detection, infrared thermal imaging detection, UAV-based image detection, audio-based detection, and neural network-based detection.
[0072] Vibration and acoustic emission detection involve drilling holes in the blades to install sensors, which not only damages the original blade structure, affecting its integrity and safety, but also requires a large number of sensors, increasing hardware costs. Furthermore, the complex maintenance process, requiring regular inspection and replacement of sensors, further increases maintenance costs.
[0073] Fiber Bragg grating (FBG) inspection requires embedding FBG sensors inside the material during blade manufacturing or attaching sensors to the blade surface, which is inconvenient to maintain and has high environmental requirements.
[0074] Ultrasonic testing is mostly used for static testing before products leave the factory.
[0075] Infrared thermal imaging is not very sensitive for detecting deep damage and is easily affected by environmental factors (light, temperature, etc.), leading to inaccurate results.
[0076] Image detection based on drones can only be carried out periodically, and cannot obtain the real-time operating status of wind turbine blades, thus failing to detect potential faults in a timely manner. This may cause the faults to develop further during the inspection interval, resulting in greater losses.
[0077] Audio-based detection methods, such as comparing audio features and calculating the difference coefficient of the spectral envelope, are easily affected by environmental noise. It is difficult to unify the applicable threshold under different operating conditions (such as changes in wind turbine speed and power generation), making it difficult to obtain a stable and reliable model.
[0078] While neural network-based detection (such as backpropagation (BP) and core-return neural networks) can improve detection accuracy to some extent, they all require large amounts of labeled data for training. In practical applications, obtaining large amounts of labeled data is costly and difficult, and these models lack generalization ability under different operating conditions and environments, making it difficult to guarantee accurate detection of blade faults in complex environments. Moreover, existing neural network-based detection methods often lack interpretability in their results, making it difficult for maintenance personnel to understand the basis of the model's judgments.
[0079] It is evident that the commonly used wind turbine blade fault detection technologies all have defects, which can easily affect the reliability of wind turbine blade fault detection.
[0080] To address the aforementioned problems, this application provides a method for detecting wind turbine blade faults. The method for detecting wind turbine blade faults according to this application will be described in detail below with reference to the accompanying drawings.
[0081] Reference Figure 1 , Figure 1 This is a flowchart illustrating a wind turbine blade fault detection method provided in an embodiment of this application, as shown below. Figure 1 As shown in the embodiments of this application, a wind turbine blade fault detection method may include the following steps, which are described in detail below.
[0082] S101: Acquire the sweeping sound signal generated by the rotation of the fan blades;
[0083] In this application, a non-contact microphone can be deployed between the microphone and the wind turbine blades to collect the sweeping sound signal generated by the rotation of the wind turbine blades. The non-contact sensor avoids damage to the wind turbine blade structure and also reduces detection costs.
[0084] S102: Perform feature processing on the sweeping sound signal to obtain overall spectral features and local spectral features; the overall spectral features are used to characterize the features of the complete frequency band of the preset acoustic signature spectral frequency range, and the local spectral features are used to characterize the cross-band fusion features of each frequency band corresponding to damage at different positions of the wind turbine blades in the preset acoustic signature spectral frequency range.
[0085] In this application, the overall spectral characteristics correspond to the entire preset voiceprint spectrum range. The preset voiceprint spectrum frequency range can be set based on scenario requirements. In one possible implementation, the preset voiceprint spectrum frequency range is 0 to 8000 Hz. In another possible implementation, the sound signal can be subjected to Fbank feature extraction processing to obtain the overall spectral characteristics of the preset voiceprint spectrum frequency range.
[0086] In this application, considering that performing Fbank feature extraction processing on the acoustic signal to obtain the overall spectral features within the preset acoustic signature frequency range has two problems: First, the feature granularity is coarse, making it impossible to accurately capture the subtle differences in features of damage at different locations of the wind turbine blades in specific frequency bands (e.g., cracks at the tip of the wind turbine blade often generate resonance in the 2000-3500Hz frequency band, while cracks at the root of the wind turbine blade are concentrated in the 800-1500Hz frequency band); Second, the anti-interference capability is weak, as environmental noise (such as rainstorms and sandstorms) and interference from other components of the wind turbine (such as gear noise in the nacelle) often overlap with the frequency bands corresponding to damage at different locations of the wind turbine blades, resulting in the effective information in the overall spectral features being submerged.
[0087] Therefore, in this application, a local spectral feature of a preset acoustic signature frequency range is designed. The local spectral feature of the preset acoustic signature frequency range is used to characterize the cross-band fusion feature of each frequency band corresponding to damage at different positions of the wind turbine blade in the preset acoustic signature frequency range. This cross-band fusion feature can associate the features of each frequency band corresponding to damage at different positions of the wind turbine blade in the preset acoustic signature frequency range from the perspective of damage physics mechanism, accurately capture the subtle feature differences of damage at different positions of the wind turbine blade in a specific frequency band, and reduce feature interference of frequency bands unrelated to damage at different positions of the wind turbine blade.
[0088] In one possible implementation, the local spectral features of the preset acoustic signature frequency range are obtained by extracting and fusing the local spectral features of each effective sub-band corresponding to different damage locations of the wind turbine blades within the preset acoustic signature frequency range.
[0089] S103: Based on the overall spectral characteristics and the local spectral characteristics, wind turbine blade fault detection is performed to obtain wind turbine blade fault detection results.
[0090] In this application, wind turbine blade fault detection can be implemented based on rules, using the overall spectral features and the local spectral features, to obtain wind turbine blade fault detection results. Alternatively, it can be implemented based on a deep learning network model, using the overall spectral features and the local spectral features, to obtain wind turbine blade fault detection results. This application does not limit the specific method used.
[0091] One method involves implementing wind turbine blade fault detection based on the overall and local spectral features using rules. This can be achieved by pre-setting a correspondence between different overall and local spectral features and different wind turbine blade fault detection results, and then using this correspondence. Another method involves implementing wind turbine blade fault detection based on the overall and local spectral features using a deep learning network model. This can be achieved by pre-training a wind turbine blade fault detection model capable of detecting faults based on the overall and local spectral features, and then inputting the overall and local spectral features into the model to obtain the wind turbine blade fault detection results.
[0092] In one possible implementation, the wind turbine blade fault detection results are used to indicate whether the wind turbine blade is in a normal or abnormal state.
[0093] The wind turbine blade fault detection method disclosed in this embodiment, after acquiring the sweeping sound signal generated by the rotation of the wind turbine blade, performs feature processing to obtain two dimensions of features: the overall spectral features of the preset acoustic signature frequency range and the local spectral features of the preset acoustic signature frequency range. Since the overall spectral features can characterize the features of the complete frequency band of the preset acoustic signature frequency range, and the local spectral features can characterize the cross-band fusion features of each frequency band corresponding to damage at different positions of the wind turbine blade in the preset acoustic signature frequency range, these two dimensions of features cover both the overall coarse-grained features and the fine-grained features of damage at different positions of the wind turbine blade. Therefore, by comprehensively utilizing these two dimensions of features for wind turbine blade fault detection, the reliability of wind turbine blade fault detection can be improved.
[0094] In one possible implementation, the determination of the local spectral features includes:
[0095] S201: Determine the frequency bands corresponding to damage at different locations on the wind turbine blades within the preset acoustic signature spectrum frequency range;
[0096] In this application, the preset acoustic spectrum frequency range can be divided into frequency bands to obtain multiple frequency bands, and then each frequency band corresponding to damage at different positions of the wind turbine blade can be selected from the multiple frequency bands.
[0097] S202: Perform feature extraction processing on each frequency band to obtain damage characteristics at different locations of the wind turbine blades;
[0098] In this application, feature extraction is performed on different frequency bands corresponding to damage at different locations of the wind turbine blades to obtain damage features at different locations of the wind turbine blades.
[0099] S203: The damage characteristics at different locations of the wind turbine blades are fused to obtain the local spectral characteristics.
[0100] In this application, the local spectral features can be obtained by fusing damage features at different locations of the wind turbine blades based on an attention mechanism.
[0101] In one possible implementation, determining the frequency bands corresponding to damage at different locations on the wind turbine blades within a preset acoustic signature spectrum frequency range includes:
[0102] S301: Divide the preset acoustic spectrum frequency range into multiple sub-bands;
[0103] In this application, through experimental testing and physical mechanism analysis, the exclusive characteristic frequency bands corresponding to different fault types and different damage locations of wind turbine blades are determined, and the preset acoustic spectrum frequency range is divided into multiple sub-frequency bands accordingly.
[0104] In one possible implementation, assuming the preset voiceprint spectrum frequency range is 0 to 8000 Hz, the preset voiceprint spectrum frequency range can be divided into 5 sub-bands. The division method of each sub-band and its corresponding physical meaning are shown in the table below:
[0105]
[0106] S302: Determine the effective sub-frequency bands related to damage at different positions of the wind turbine blades from the multiple sub-frequency bands, and use them as the frequency bands corresponding to damage at different positions of the wind turbine blades in the preset acoustic spectrum frequency range.
[0107] In this application, some frequency bands among the multiple sub-frequency bands are unrelated to damage at different positions of the wind turbine blades. Therefore, in this application, effective sub-frequency bands related to damage at different positions of the wind turbine blades can be determined from the multiple sub-frequency bands and used as each frequency band corresponding to damage at different positions of the wind turbine blades in the preset acoustic spectrum frequency range. Invalid sub-frequency bands that are determined to be unrelated to damage at different positions of the wind turbine blades are filtered out, reducing interference from invalid information in wind turbine blade fault detection, so as to ensure the reliability of wind turbine blade fault detection.
[0108] For ease of understanding, exemplarily, assuming the preset acoustic spectrum frequency range is 0 to 8000Hz, the preset acoustic spectrum frequency range can be divided into the above 5 sub-bands. Then, only sub-band 2 (root crack), sub-band 3 (middle damage), and sub-band 4 (tip crack) can be retained as "effective sub-bands", while sub-band 1 and sub-band 5 can be directly filtered as "interference sub-bands".
[0109] In one possible implementation, the feature extraction process for each frequency band to obtain damage features at different locations on the wind turbine blades includes:
[0110] S401: Extract the differential features for each frequency band to obtain the differential features for each frequency band;
[0111] To avoid information loss caused by using general features, this application extracts differentiated features for each frequency band corresponding to damage at different locations on the wind turbine blades within a preset acoustic signature spectrum frequency range, thereby obtaining differentiated features for each frequency band.
[0112] For ease of understanding, exemplarily assuming the preset acoustic signature spectrum frequency range is 0 to 8000 Hz, the preset acoustic signature spectrum frequency range can be divided into 5 sub-bands, and sub-band 2 (root crack), sub-band 3 (middle damage), and sub-band 4 (tip crack) are retained as "effective sub-bands". The specific feature extraction scheme can then be as follows:
[0113]
[0114] S402: By using linear projection, the differential features of each frequency band are mapped to the same dimension to obtain the damage features of different positions of the wind turbine blades.
[0115] For ease of understanding, let's assume that the 10-second voiceprint segment is divided into 20 time frames in 500ms windows, and the feature frames of each sub-band are expanded into a matrix of "time length × sub-band dimension". Then, according to the feature extraction scheme described above, the differentiated features extracted from each sub-band are as follows:
[0116] Sub-band 2: 20×32 matrix, Sub-band 3: 20×64 matrix, Sub-band 4: 20×48 matrix;
[0117] By mapping the differentiated features of each sub-band to the same dimension (64-dimensional) through linear projection, three 20×64 feature matrices (with the same time series length of 20 and dimension of 64) are obtained, which are the damage features of different positions of the wind turbine blades.
[0118] In one possible implementation, fusing the damage features at different locations on the wind turbine blades to obtain the local spectral features includes:
[0119] Each wind turbine blade position damage feature is used as the query, and the remaining wind turbine blade position damage features are used as the key and value to calculate the attention weight; the damage features at different positions of the wind turbine blades are weighted and fused based on the attention weight to obtain the local spectral features.
[0120] For ease of understanding, exemplarily, assuming the preset acoustic signature spectrum frequency range is 0 to 8000Hz, the preset acoustic signature spectrum frequency range can be divided into 5 sub-bands, and sub-band 2 (root crack), sub-band 3 (middle damage), and sub-band 4 (tip crack) are retained as "effective sub-bands". The damage features at different locations of the wind turbine blade are root features, middle features, and tip features, respectively. Then, the root features, middle features, and tip features can be used as "queries (Q)" and the remaining two as "keys (K)" and "values (V)", respectively. Attention weights are calculated, and finally, weighted fusion is performed to output a 20×64 "global fusion feature".
[0121] For ease of understanding, the formula is as follows:
[0122]
[0123] Where α, β, and γ are the attention weights for the root feature, middle feature, and tip feature, respectively. For local spectral characteristics, For root characteristics, For the characteristics of the central region, It is a tip-shaped feature.
[0124] In one possible implementation, the step of performing wind turbine blade fault detection based on the overall spectral features and the local spectral features to obtain wind turbine blade fault detection results includes:
[0125] The overall spectral features and the local spectral features are input into the wind turbine blade fault detection model. The blade fault detection model encodes the overall spectral features and the local spectral features respectively and then fuses them to obtain fused features. Based on the fused features, anomaly classification is performed to obtain the wind turbine blade fault detection result.
[0126] In this application, the wind turbine blade fault detection model can be a self-supervised model based on the ViT architecture. In one possible implementation, please refer to... Figure 2 , Figure 2 This is a schematic diagram of a wind turbine blade fault detection model architecture provided in an embodiment of this application. Figure 2 As shown, the wind turbine blade fault detection model includes a first encoder, a second encoder, a fusion layer, and an anomaly classification layer. The first encoder is used to encode the overall spectral features to obtain an embedded representation of the overall spectral features. The second encoder is used to encode the local spectral features to obtain an embedded representation of the local spectral features. The fusion layer is used to fuse the embedded representations of the overall spectral features and the embedded representations of the local spectral features to obtain fused features. The anomaly classification layer is used to obtain the wind turbine blade fault detection result based on the fused features.
[0127] In one possible implementation, the training method for the wind turbine blade fault detection model includes:
[0128] S501: Acquire training data, the training data including swept air sound signal samples, the overall spectrum features and local spectrum features corresponding to the swept air sound signal samples, and the wind turbine blade fault detection result label corresponding to the swept air sound signals;
[0129] S502: Constructing a training architecture for a wind turbine blade fault detection model;
[0130] In this application, the training architecture for the wind turbine blade fault detection model can be a self-supervised model based on the ViT architecture. In one possible implementation, please refer to... Figure 3 , Figure 3 This is a schematic diagram of a wind turbine blade fault detection model training architecture provided in an embodiment of this application. Figure 3 As shown, the training architecture of the wind turbine blade fault detection model includes a first encoder, a second encoder, a decoder, a fusion layer, and an anomaly classification layer. The first encoder is used to encode the overall spectral features to obtain the embedded representation of the overall spectral features. The second encoder is used to encode the local spectral features to obtain the embedded representation of the local spectral features. The decoder is used to perform mask reconstruction on the embedded representation of the overall spectral features. The fusion layer is used to fuse the embedded representation of the overall spectral features and the embedded representation of the local spectral features to obtain the fused features. The anomaly classification layer is used to obtain the wind turbine blade fault detection result based on the fused features.
[0131] S503: The overall spectral features and the local spectral features are input together into the wind turbine blade fault detection model training architecture. The wind turbine blade fault detection model training architecture reconstructs the overall spectral features by performing directional masking based on the frequency bands corresponding to damage at different locations of the wind turbine blade, and is used for positive anomaly classification.
[0132] To address the lack of domain specificity in random masks in existing ViT architecture self-supervised models such as Audio-MAE, this application proposes a directional masking strategy based on the frequency bands corresponding to damage at different locations on wind turbine blades, which forces the model to focus on key feature frequency bands, thereby improving pre-training efficiency and downstream detection task performance.
[0133] For ease of understanding, let's assume that the frequency bands corresponding to damage at different locations on the wind turbine blades are the root (800-1500Hz), the middle (1500-200Hz), and the tip (2000-3500Hz). In this application, a mask quota is allocated during masking (total masking rate 75%, with each of the three frequency bands accounting for 25%), allowing the model to focus on learning and reconstructing the features of the frequency bands corresponding to damage at different locations on the wind turbine blades.
[0134] Considering that abnormal acoustic signatures caused by wind turbine blade damage will exhibit a sudden increase in energy in the corresponding frequency band (such as high-energy pulses generated by tip cracks in the 2000-3500Hz frequency band), this application calculates the energy distribution probability on the time axis for the frequency band corresponding to damage at different locations on the wind turbine blade. Based on the energy distribution probability, the probability of high-energy time points is selected, and a mask block is generated centered on this point.
[0135] For example, assuming that damage at different locations on a wind turbine blade corresponds to the following frequency bands: root (800-1500Hz), middle (1500-200Hz), and tip (2000-3500Hz), and the overall spectral characteristics are F-bank features, then for each frequency band, the energy distribution probability on the time axis is calculated as follows:
[0136] ;
[0137] in, , indicating three frequency band regions (root, middle, and tip);
[0138] express The corresponding Fbank frequency index set;
[0139] This represents the energy value of the Fbank feature at time step t and frequency index f;
[0140] This represents the normalized energy distribution (probability value) of region r at time step t.
[0141] Based on the probability of selecting high-energy time points according to energy distribution, a mask block is generated centered on these points.
[0142] .
[0143] S504: The wind turbine blade fault detection model training architecture is trained using the reconstruction loss of the overall spectral features and the positive anomaly classification loss as joint losses to obtain the trained wind turbine blade fault detection model.
[0144] In this application, the reconstruction loss of the overall spectral features is used to indicate the difference between the reconstruction result of the overall spectral features of the wind turbine blade fault detection model training architecture and the overall spectral features corresponding to the swept sound signal sample; the positive anomaly classification loss is used to indicate the difference between the positive anomaly classification result of the wind turbine blade fault detection model training architecture and the wind turbine blade fault detection result label corresponding to the swept sound signal.
[0145] For example, the final total loss is: +
[0146] in, For positive anomaly classification loss, For reconstruction loss.
[0147] The trained wind turbine blade fault detection model is the wind turbine blade fault detection model architecture excluding the decoder.
[0148] The above describes a wind turbine blade fault detection method provided by the embodiments of this application. The following will describe the apparatus for performing the above wind turbine blade fault detection method.
[0149] Please see Figure 4 , Figure 4 This is a schematic diagram of a wind turbine blade fault detection device provided in an embodiment of this application. Figure 4 As shown, the wind turbine blade fault detection device includes:
[0150] Acquisition unit 11 is used to acquire the sweeping sound signal generated by the rotation of the fan blades;
[0151] Feature processing unit 12 is used to perform feature processing on the wind sweeping sound signal to obtain overall spectral features and local spectral features; the overall spectral features are used to characterize the features of the complete frequency band of the preset acoustic signature spectral frequency range, and the local spectral features are used to characterize the cross-band fusion features of each frequency band corresponding to damage at different positions of the wind turbine blades in the preset acoustic signature spectral frequency range.
[0152] The fault detection unit 13 is used to perform wind turbine blade fault detection based on the overall spectrum characteristics and the local spectrum characteristics, and obtain the wind turbine blade fault detection result.
[0153] In one possible implementation, the feature processing unit includes:
[0154] The frequency band determination unit is used to determine the frequency bands corresponding to damage at different locations on the wind turbine blades within the preset acoustic spectrum frequency range.
[0155] The feature extraction and processing unit is used to perform feature extraction processing on each frequency band to obtain damage features at different locations of the wind turbine blades;
[0156] The feature fusion unit is used to fuse the damage features at different locations of the wind turbine blades to obtain the local spectral features.
[0157] In one possible implementation, the frequency band determination unit is specifically used for:
[0158] The preset acoustic spectrum frequency range is divided into multiple sub-bands;
[0159] Effective sub-frequency bands related to damage at different locations of the wind turbine blades are determined from the multiple sub-frequency bands, and these sub-frequency bands are used as the frequency bands corresponding to damage at different locations of the wind turbine blades within the preset acoustic spectrum frequency range.
[0160] In one possible implementation, the feature extraction processing unit is specifically used for:
[0161] Differential features are extracted for each frequency band to obtain the differential features of each frequency band;
[0162] By mapping the differential features of each frequency band to the same dimension through linear projection, the damage features of different locations on the wind turbine blades can be obtained.
[0163] In one possible implementation, the feature fusion unit is specifically used for:
[0164] Calculate the attention weight by using the damage features of each wind turbine blade location as the query and the remaining wind turbine blade location damage features as the key and value;
[0165] The local spectral features are obtained by weighting and fusing the damage features at different locations of the wind turbine blades based on the attention weights.
[0166] In one possible implementation, the fault detection unit is specifically used for:
[0167] The overall spectral features and the local spectral features are input into the wind turbine blade fault detection model. The blade fault detection model encodes the overall spectral features and the local spectral features respectively and then fuses them to obtain fused features. Based on the fused features, anomaly classification is performed to obtain the wind turbine blade fault detection result.
[0168] In one possible implementation, the device further includes: a wind turbine blade fault detection model training unit;
[0169] The wind turbine blade fault detection model training unit is specifically used for:
[0170] Acquire training data, which includes swept air sound signal samples, the overall spectral features and local spectral features corresponding to the swept air sound signal samples, and the wind turbine blade fault detection result labels corresponding to the swept air sound signals;
[0171] Construct a training architecture for a wind turbine blade fault detection model;
[0172] The overall spectral features and the local spectral features are input together into the wind turbine blade fault detection model training architecture. The wind turbine blade fault detection model training architecture reconstructs the overall spectral features by performing directional masking based on the frequency bands corresponding to damage at different locations of the wind turbine blade, and is used for positive anomaly classification.
[0173] The wind turbine blade fault detection model training architecture is trained using the reconstruction loss of the overall spectral features and the positive anomaly classification loss as joint losses to obtain a trained wind turbine blade fault detection model.
[0174] Each unit in the aforementioned wind turbine blade fault detection device can be implemented entirely or partially through software, hardware, or a combination thereof. These units can be embedded in or independent of the processor in a computer device, or stored in the computer device's memory as software, so that the processor can call and execute the corresponding operations of each unit.
[0175] This application also provides an electronic device in its embodiments. (See reference...) Figure 5 The diagram illustrates a structural schematic suitable for implementing the electronic device in the embodiments of this application. The electronic device in the embodiments of this application may include, but is not limited to, fixed terminals such as mobile phones, laptops, PDAs (personal digital assistants), PADs (tablet computers), desktop computers, etc. Figure 5 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.
[0176] like Figure 5 As shown, the electronic device may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage device 608 into a random access memory (RAM) 603. When the electronic device is powered on, the RAM 603 also stores various programs and data required for the operation of the electronic device. The processing unit 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.
[0177] Typically, the following devices can be connected to I / O interface 605: input devices 606 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 607 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 608 including, for example, memory cards, hard drives, etc.; and communication devices 609. Communication device 609 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 5 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown. More or fewer devices may be implemented or have alternatively.
[0178] This application also provides a computer program product including computer-readable instructions, which, when executed on an electronic device, cause the electronic device to implement any of the wind turbine blade fault detection methods provided in this application.
[0179] This application also provides a computer-readable storage medium that carries one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement any of the wind turbine blade fault detection methods provided in this application.
[0180] It should also be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. In addition, in the device embodiment drawings provided in this application, the connection relationship between modules indicates that they have a communication connection, which can be implemented as one or more communication buses or signal lines.
[0181] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware, or it can be implemented by special-purpose hardware including application-specific integrated circuits, special-purpose CPUs, special-purpose memory, special-purpose components, etc. Generally, any function performed by a computer program can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be diverse, such as analog circuits, digital circuits, or special-purpose circuits. However, for this application, software program implementation is more often the preferred implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a readable storage medium, such as a computer floppy disk, USB flash drive, mobile hard disk, ROM, RAM, magnetic disk, or optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, training equipment, or network device, etc.) to execute the methods described in the various embodiments of this application.
[0182] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product.
[0183] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, training device, or data center to another website, computer, training device, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a training device or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).
Claims
1. A wind turbine blade fault detection method, characterized by, The method comprises: acquiring a sweeping sound signal generated by rotation of a fan blade; performing feature processing on the sweeping sound signal to obtain an overall frequency spectrum feature and a local frequency spectrum feature; the overall frequency spectrum feature is used to represent features of complete frequency bands in a preset voiceprint spectrum frequency range, and the local frequency spectrum feature is used to represent cross-frequency band fusion features of each frequency band corresponding to damage at different positions of the fan blade in the preset voiceprint spectrum frequency range; performing fan blade fault detection based on the overall frequency spectrum feature and the local frequency spectrum feature to obtain a fan blade fault detection result.
2. The method of claim 1, wherein, The determination manner of the local frequency spectrum feature comprises: determining each frequency band corresponding to damage at different positions of the fan blade in the preset voiceprint spectrum frequency range; performing feature extraction processing on each frequency band to obtain fan blade damage features at different positions; fusing the fan blade damage features at different positions to obtain the local frequency spectrum feature.
3. The method of claim 2, wherein, The determination of each frequency band corresponding to damage at different positions of the fan blade in the preset voiceprint spectrum frequency range comprises: dividing the preset voiceprint spectrum frequency range into a plurality of sub-bands; determining effective sub-bands related to damage at different positions of the fan blade from the plurality of sub-bands as each frequency band corresponding to damage at different positions of the fan blade in the preset voiceprint spectrum frequency range.
4. The method of claim 2, wherein, The feature extraction processing on each frequency band to obtain fan blade damage features at different positions comprises: respectively performing differential feature extraction on each frequency band to obtain differential features of each frequency band; mapping the differential features of each frequency band to the same dimension through linear projection to obtain the fan blade damage features at different positions.
5. The method of claim 2, wherein, The fusion of the fan blade damage features at different positions to obtain the local frequency spectrum feature comprises: respectively taking each fan blade position damage feature as a query, and taking remaining fan blade position damage features as keys and values to calculate attention weights; performing weighted fusion on the fan blade damage features at different positions based on the attention weights to obtain the local frequency spectrum feature.
6. The method of claim 1, wherein, The fan blade fault detection based on the overall frequency spectrum feature and the local frequency spectrum feature to obtain a fan blade fault detection result comprises: inputting the overall frequency spectrum feature and the local frequency spectrum feature into a fan blade fault detection model, the blade fault detection model respectively encoding the overall frequency spectrum feature and the local frequency spectrum feature and then fusing to obtain fusion features, and performing abnormal classification based on the fusion features to obtain a fan blade fault detection result.
7. The method of claim 6, wherein, The training manner of the fan blade fault detection model comprises: obtaining training data, the training data comprising a sweeping sound signal sample, overall frequency spectrum features and local frequency spectrum features corresponding to the sweeping sound signal sample, and fan blade fault detection result labels corresponding to the sweeping sound signal; constructing a fan blade fault detection model training architecture; inputting the overall frequency spectrum feature and the local frequency spectrum feature into the fan blade fault detection model training architecture, the fan blade fault detection model training architecture being used for directional masking of the overall frequency spectrum feature based on frequency bands corresponding to damage at different positions of the fan blade and then reconstructing, and being used for positive abnormal classification. The fan blade fault detection model training architecture is trained by taking a reconstruction loss of overall spectral features and a positive anomaly classification loss as a joint loss to obtain a trained fan blade fault detection model.
8. A computer program product, characterised in that, The computer readable instructions, when executed on an electronic device, cause the electronic device to implement the fan blade fault detection method of any one of claims 1 to 7.
9. An electronic device, comprising: The electronic device comprises at least one processor and a memory connected to the processor, wherein: The memory is configured to store computer programs; The processor is configured to execute the computer programs to enable the electronic device to implement the fan blade fault detection method of any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, The storage medium carries one or more computer programs, which, when executed by an electronic device, enable the electronic device to implement the fan blade fault detection method of any one of claims 1 to 7.