Fan abnormity detection method, electronic equipment, medium and product

Abstract

The invention discloses a fan abnormity detection method, electronic equipment, a medium and a product, and relates to the technical field of machine learning, and the fan abnormity detection method comprises the steps that a fan is provided with an audio sensor array; acquiring original audio data generated by a target detection part in the fan in a running state through an audio sensor array; extracting acoustic features from the original audio data, and associating the acoustic features with the identifier of the target detection component and the working condition parameters of the fan to obtain to-be-detected acoustic features; and inputting the to-be-detected acoustic features into a preset abnormal audio detection model, and outputting an abnormal detection result of the target detection component in the running state. According to the method and the device, a new way of carrying out anomaly detection from the audio dimension is realized, and automation, precision, comprehensiveness and early warning capability of fan monitoring are improved.

Description

Technical Field

[0001] This application relates to the field of machine learning technology, and in particular to a method, electronic device, medium and product for detecting wind turbine anomalies. Background Technology

[0002] Ventilation fan condition monitoring primarily relies on sensing specific signals directly coupled to the equipment, such as vibration, temperature, or current. However, this method has limitations in deployment, monitoring dimensions, and information acquisition: effective implementation usually requires establishing a direct physical or electrical connection with the equipment under test, resulting in insufficient deployment flexibility. Furthermore, these traditional detection dimensions (such as mechanical vibration, thermal, or electromagnetic characteristics) mostly reflect only partial information about the equipment's condition, and the correlation between the obtained data and complex fault mechanisms is limited, making it difficult to comprehensively and early reveal operational anomalies. Therefore, this traditional monitoring method has limited detection effectiveness and cannot meet practical application needs. Summary of the Invention

[0003] The main purpose of this application is to provide a method, electronic device, medium and product for detecting fan malfunctions, which aims to solve the technical problem that traditional detection methods have limited detection effects and cannot meet the needs of practical applications.

[0004] To achieve the above objectives, this application proposes a method for detecting fan malfunctions, wherein the fan is equipped with an audio sensor array; The method for detecting fan malfunctions includes: The audio sensor array is used to collect raw audio data generated by the target detection component in the wind turbine during operation. Acoustic features are extracted from the original audio data, and the acoustic features are associated with the identifier of the target detection component and the operating parameters of the fan to obtain the acoustic features to be detected. The acoustic features to be detected are input into a preset abnormal audio detection model, and the abnormal detection results of the target detection component in the running state are output.

[0005] Optionally, the audio sensor array includes a first type of microphone and a second type of microphone, wherein the first type of microphone and the second type of microphone are located in different positions; The step of acquiring raw audio data generated by the target detection component in the wind turbine during operation through the audio sensor array includes: First audio data is acquired using the first type of microphone, and second audio data is acquired using the second type of microphone. Based on the delay compensation parameter corresponding to the target detection component, the first audio data and the second audio data are fused to obtain the original audio data generated by the target detection component in operation. The delay compensation parameter is determined based on the spatial positional relationship between the target detection component and the first type of microphone and the second type of microphone.

[0006] Optionally, the step of fusing the first audio data and the second audio data based on the delay compensation parameters corresponding to the target detection component to obtain the original audio data generated by the target detection component in its operating state includes: From the first audio data and the second audio data, audio data to be delayed and non-delayed audio data are determined, wherein the microphone corresponding to the audio data to be delayed is closer to the target detection component than the microphone corresponding to the non-delayed audio data; The delayed audio data is obtained by delaying the audio data to be delayed using the delay compensation parameters. The delayed audio data is superimposed with the non-delayed audio data to obtain the original audio data.

[0007] Optionally, the step of extracting acoustic features from the original audio data includes: The original audio signal is preprocessed to obtain preprocessed audio data. The preprocessing includes at least noise reduction and time-frequency conversion. Acoustic features are extracted from the preprocessed audio data, wherein the types of acoustic features include time-domain features, frequency-domain features, and time-frequency-domain features.

[0008] Optionally, the step of preprocessing the original audio signal to obtain preprocessed audio data includes: filtering the original audio data to obtain first audio data; performing frame segmentation and windowing on the first audio data to obtain second audio data; and performing time-frequency conversion on the second audio data to generate a corresponding spectrogram as the preprocessed audio data.

[0009] Optionally, the method for detecting fan malfunctions further includes: Based on the historical operating data of the observed wind turbines, a training sample set is constructed, wherein the observed wind turbines include the wind turbine and / or related wind turbines; The initial detection model is iteratively trained using the training sample set until the preset training conditions are met, thus obtaining the preset abnormal audio detection model.

[0010] Optionally, the step of constructing a training sample set based on the historical operating data of the observed wind turbines includes: For any abnormal record in the historical operation data, the abnormal component is determined based on the cause of the abnormality in the abnormal record; The target audio segment is extracted from historical audio data pointing to the abnormal component based on a preset anomaly observation window; Based on the target audio segment, normal audio segments are filtered from the historical audio data, wherein the difference between the normal audio segment and the target audio segment is greater than a preset threshold, and the wind turbine operating condition corresponding to the normal audio segment is consistent with the wind turbine operating condition corresponding to the target audio segment. In the case of obtaining the normal audio segment, based on the target audio segment, the wind turbine operating condition corresponding to the target audio segment, and the abnormal component, an abnormal training sample is constructed in the training sample set, and based on the normal audio segment, the wind turbine operating condition corresponding to the normal audio segment, and the abnormal component, a normal training sample is constructed in the training sample set. If the normal audio segment is not obtained through filtering, the abnormal record is discarded.

[0011] In addition, to achieve the above objectives, this application also proposes an electronic device, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the fan anomaly detection method as described above.

[0012] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the fan anomaly detection method described above.

[0013] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the fan anomaly detection method described above.

[0014] One or more technical solutions proposed in this application have at least the following technical effects: In this embodiment, the audio sensor array collects raw audio data generated by the target detection component in the wind turbine during operation. Acoustic features are extracted from the raw audio data and associated with the identifier of the target detection component and the operating parameters of the wind turbine to obtain the acoustic features to be detected. The acoustic features to be detected are input into a preset abnormal audio detection model, and the abnormal detection result of the target detection component during operation is output. In other words, this application constructs a complete closed-loop wind turbine audio detection and diagnosis process. It improves signal quality through array-based acquisition, achieves context awareness through deep fusion of features with operating conditions and components, and finally outputs accurate diagnoses through the model. Compared with traditional methods that rely on fixed thresholds and empirical formulas, this application realizes a new approach to abnormal detection from the audio dimension, enabling the detection system to identify hidden fault modes, improving the comprehensiveness of detection, and providing quantitative status assessment. This provides a direct basis for predictive maintenance and significantly improves the automation, accuracy, and early warning capabilities of wind turbine monitoring. Attached Figure Description

[0015] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating the first embodiment of the fan malfunction detection method of this application; Figure 2 This is a flowchart illustrating the second embodiment of the fan malfunction detection method of this application; Figure 3 This is a flowchart illustrating the third embodiment of the fan malfunction detection method of this application; Figure 4 This is a flowchart illustrating the fourth embodiment of the fan malfunction detection method of this application; Figure 5 This is a schematic diagram of the equipment structure of the hardware operating environment involved in the fan malfunction detection method in this application embodiment.

[0018] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0020] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0021] Ventilation fan condition monitoring primarily relies on sensing specific signals directly coupled to the equipment, such as vibration, temperature, or current. However, this method has limitations in deployment, monitoring dimensions, and information acquisition: effective implementation usually requires establishing a direct physical or electrical connection with the equipment under test, resulting in insufficient deployment flexibility. Furthermore, these traditional detection dimensions (such as mechanical vibration, thermal, or electromagnetic characteristics) mostly reflect only partial information about the equipment's condition, and the correlation between the obtained data and complex fault mechanisms is limited, making it difficult to comprehensively and early reveal operational anomalies. Therefore, this traditional monitoring method has limited detection effectiveness and cannot meet practical application needs.

[0022] The main solution of this application embodiment is: the fan is equipped with an audio sensor array; the original audio data generated by the target detection component in the fan during operation is collected through the audio sensor array; acoustic features are extracted from the original audio data, and the acoustic features are associated with the identifier of the target detection component and the operating parameters of the fan to obtain the acoustic features to be detected; the acoustic features to be detected are input into a preset abnormal audio detection model, and the abnormal detection results of the target detection component during operation are output.

[0023] This application constructs a complete closed-loop wind turbine audio detection and diagnosis process. It improves signal quality through array-based acquisition, achieves context awareness through deep fusion of features with operating conditions and components, and finally outputs accurate diagnoses through model output. Compared with traditional methods that rely on fixed thresholds and empirical formulas, this application realizes a new approach to anomaly detection from the audio dimension, enabling the detection system to identify hidden fault modes, improving the comprehensiveness of detection, and providing quantitative condition assessment. This provides a direct basis for predictive maintenance and significantly improves the automation, accuracy, and early warning capabilities of wind turbine monitoring.

[0024] It should be noted that the execution subject of this embodiment can be a computing service device with data processing, network communication and program running functions, such as a wind turbine generator set, an operation and maintenance platform associated with the wind turbine generator set, a cloud platform, a computer, a mobile phone, etc., or an electronic device capable of realizing the above functions.

[0025] Reference Figure 1This is a flowchart illustrating the first embodiment of the fan malfunction detection method of this application. In this embodiment, the fan is equipped with an audio sensor array. The fan malfunction detection method includes steps S10 to S30: Step S10: Collect raw audio data generated by the target detection component in the wind turbine during operation using an audio sensor array; It should be noted that, in this embodiment, the above-mentioned method for detecting wind turbine anomalies can be applied to wind turbines or to the operation and maintenance platform associated with wind turbines.

[0026] For example, an audio sensor array can be installed on the wind turbine. This array can target every component of the wind turbine, or components with a high failure rate (such as the front and rear bearings of the generator, or the cooling fan blades). For instance, for bearing monitoring, a multi-microphone linear array can be used, arranged at specific intervals (e.g., 5 cm) along the bearing housing axis, with the array's main axis aligned with the center of the bearing housing. Therefore, the specific target detection component can be set based on actual needs. During data acquisition, a synchronization acquisition card ensures that the sampling clocks of all channels are synchronized to maximize the capture of direct sound waves from the target component and suppress interference from other directions.

[0027] Alternatively, audio data can be collected from the wind turbine body using a ring or spherical microphone array. After acquiring multi-channel audio, beamforming algorithms (such as delay summation beamforming or minimum variance distortionless response beamforming) can be applied for post-processing.

[0028] Understandably, array-based acquisition makes it possible to use array signal processing technology for noise suppression, sound source enhancement, and preliminary separation, fundamentally improving the anti-interference capability of single-point audio monitoring and ensuring the audio quality of the acquired raw audio data.

[0029] Step S20: Extract acoustic features from the original audio data, and associate the acoustic features with the identifier of the target detection component and the operating parameters of the fan to obtain the acoustic features to be detected. For example, the original audio data can be preprocessed, and multi-domain feature calculations can be performed on the preprocessed audio data to extract acoustic features. Optionally, the acoustic features can be time-domain features or frequency features, etc. It should be noted that the above-mentioned operating parameters can be the wind turbine's power generation or the wind turbine's rotational speed. The noise generated by the wind turbine may vary under different operating conditions. Moreover, the noise generated during operation may also differ for different components. Therefore, in this embodiment, the extracted acoustic features, the corresponding target detection component identifier, and the wind turbine's operating parameters are associated to obtain the acoustic features to be detected. For example, the extracted acoustic features (such as a 128-dimensional MFCC feature vector), the normalized operating parameter vector (such as a 4-dimensional vector), and the target detection component identifier encoded using one-hot encoding (for example, encoding "front bearing" as [1,0,0]) are concatenated to form a comprehensive acoustic feature vector to be detected. This vector simultaneously contains three pieces of information: "what component", "under what conditions", and "what kind of sound was emitted", thus ensuring the comprehensiveness of the audio information.

[0030] Understandably, by strongly correlating and fusing acoustic features with precise operating context and component identity, the subsequent model can effectively distinguish acoustic changes caused by normal fluctuations in operating conditions from genuine fault symptoms. This greatly improves the model's robustness and diagnostic specificity under varying operating conditions, and solves the problem of high false alarm rates in traditional methods due to neglecting the influence of operating conditions.

[0031] Step S30: Input the acoustic features to be detected into the preset abnormal audio detection model and output the abnormal detection results of the target detection component in the running state.

[0032] For example, the associated acoustic features to be detected are input into a preset abnormal audio detection model. This preset abnormal audio detection model is pre-trained to have anomaly detection capabilities. The preset abnormal audio detection model then outputs the anomaly detection results of the target detection component in its operational state. The anomaly detection results may include whether an anomaly exists, and if an anomaly exists, the anomaly detection results may also include the cause of the anomaly.

[0033] Optionally, the aforementioned pre-defined abnormal audio detection model can be a deep fully connected network or a one-dimensional convolutional network. Its input layer dimension matches the feature vector to be detected. After training, for the input features to be detected, the output layer generates a probability distribution using the Softmax function. For example, the output vector might be [P(normal), P(bearing inner race fault), P(bearing outer race fault), P(blade dust accumulation)]. The category corresponding to the maximum probability, along with its confidence level (probability value), is output as the abnormal detection result. When the confidence level is below a certain threshold (e.g., 0.8), an "uncertain" status can be output to prompt manual review.

[0034] In this embodiment, an audio sensor array is used to collect raw audio data generated by the target detection component in the wind turbine during operation. Acoustic features are extracted from the raw audio data and correlated with the identifier of the target detection component and the operating parameters of the wind turbine to obtain the acoustic features to be detected. The acoustic features to be detected are input into a preset abnormal audio detection model, and the abnormal detection results of the target detection component during operation are output. That is, this application constructs a complete closed-loop wind turbine audio detection and diagnosis process. It improves signal quality through array-based acquisition, achieves context awareness through deep fusion of features with operating conditions and components, and finally outputs accurate diagnoses through the model. Compared with traditional methods that rely on fixed thresholds and empirical formulas, this application realizes a new approach to abnormal detection from the audio dimension, enabling the detection system to identify hidden fault modes, improving the comprehensiveness of detection, and providing quantitative status assessment. This provides a direct basis for predictive maintenance and significantly improves the automation, accuracy, and early warning capabilities of wind turbine monitoring.

[0035] Reference Figure 2 This is a flowchart illustrating the second embodiment of this example. Content in this embodiment that is the same as or similar to the above embodiments can be referred to the above description and will not be repeated hereafter. The audio sensor array includes a first type of microphone and a second type of microphone, the positions of which are different; the steps for collecting raw audio data generated by the target detection component in the wind turbine during operation using the audio sensor array include steps S11 to S12: Step S11: First audio data is acquired through the first type of microphone, and second audio data is acquired through the second type of microphone; Step S12: Based on the delay compensation parameter corresponding to the target detection component, the first audio data and the second audio data are fused to obtain the original audio data generated by the target detection component in the running state. The delay compensation parameter is determined based on the spatial positional relationship between the target detection component and the first type of microphone and the second type of microphone.

[0036] It should be noted that in this embodiment, the audio sensor array includes a first type of microphone and a second type of microphone. Among them, the number of the first type of microphones can be one or more. Similarly, the number of the second type of microphones can be one or more. And the installation positions of the first type of microphones and the second type of microphones are different.

[0037] Exemplarily, the first audio data is collected from the ambient audio by the first type of microphones, and the second audio data is collected from the ambient audio by the second type of microphones. Since the positions of the first type of microphones and the second type of microphones are different, there are also differences between the first audio data and the second audio data. For example, the time when the sound wave emitted by the same sound source arrives at the first type of microphones and the second type of microphones may be different. Then, through the delay compensation parameter of the target detection component, the first audio data and the second audio data are fused to obtain the original audio data generated by the target detection component in the running state. It should be noted that the delay compensation parameter is determined according to the spatial position relationship between the target detection component and the first type of microphones as well as the second type of microphones. For example, the greater the difference in the distances of the two types of microphones relative to the target detection component, the greater the value of the above delay compensation parameter.

[0038] Exemplarily, let the target detection component be a point sound source P, the distance of the first type of microphone M1 be d1, and the distance of the second type of microphone M2 be d2 (d1 < d2). The speed of sound is c. The delay compensation parameter τ = (d1 - d2) / c, which is a negative value (meaning the signal is advanced). In the process of audio fusion, a delay of a fractional sampling point is applied to the signal from M1 (which needs to be implemented using precise interpolation algorithms such as sinc interpolation) so that it is aligned in time with the sound component from P in the signal from M2, and then the two signals are added with equal gain. That is, after the first type of audio data and the second type of audio data are delayed and fused, the sound of P can be superimposed and enhanced, while the sound other than P will be superimposed and weakened.

[0039] In a feasible implementation manner, the steps of fusing the first audio data and the second audio data based on the delay compensation parameter corresponding to the target detection component to obtain the original audio data generated by the target detection component in the running state include steps S121 to S123: Step S121, determine the audio data to be delayed and the non-delayed audio data from the first audio data and the second audio data, where the microphone corresponding to the audio data to be delayed is closer to the target detection component than the microphone corresponding to the non-delayed audio data; Step S122, perform a delay operation on the audio data to be delayed through the delay compensation parameter to obtain the delayed audio data; Step S123, superimpose the delayed audio data and the non-delayed audio data to obtain the original audio data.

[0040] For example, based on a pre-stored database of microphone and component spatial locations, the system automatically selects the channel signal emitted by the microphone closer to the target component from the first and second audio data as the audio data to be delayed, while the other is the non-delayed audio data. Further, the audio data to be delayed is then delayed based on delay compensation parameters. Optionally, the short-time Fourier transform spectrum of the signal to be delayed is multiplied by a linear phase term exp(-jωτ), where ω is the angular frequency and τ is the delay compensation parameter. Then, an inverse transform is performed back to the time domain to obtain the delayed audio data. The delayed audio data is then superimposed with the other non-delayed audio data. For further optimization, before superposition, different weights can be assigned to the two signals based on their long-term average signal-to-noise ratio under fault-free conditions, and a weighted sum can be performed to maximize the overall signal-to-noise ratio of the output signal.

[0041] Understandably, this application enhances acoustic signals from a specific direction (target component) through coherent superposition, while incoherently superimposing noise from other directions (which cannot be perfectly aligned due to their different arrival time differences), thereby weakening noise from other directions. This constitutes a simple spatial filter that effectively improves the significance of fault feature components in the original audio data, providing a "cleaner" input for subsequent analysis.

[0042] Reference Figure 3 This is a flowchart illustrating the third embodiment of this example. Content in this embodiment that is the same as or similar to the above embodiments can be referred to the above description, and will not be repeated hereafter. The steps for extracting acoustic features from the raw audio data include steps S21 to S22: Step S21: Preprocess the original audio signal to obtain preprocessed audio data. The preprocessing includes at least noise reduction and time-frequency conversion. Step S22: Extract acoustic features from the preprocessed audio data. The types of acoustic features include time-domain features, frequency-domain features, and time-frequency-domain features.

[0043] It should be noted that in this embodiment, the acquired raw time-domain waveform is transformed into a standardized data form that is more suitable for analysis and machine learning through acoustic features, thereby improving the signal quality.

[0044] For example, preprocessing operations such as noise reduction and time-frequency conversion can be performed on the original audio signal to obtain preprocessed audio data. Optionally, for noise reduction, a zero-phase digital filter (such as a Butterworth bandpass filter implemented using the `filtfilt` function) can be applied to the original audio signal. The passband range of this filter is preset according to the type of wind turbine and its fault characteristic frequency range to filter out low-frequency mechanically conducted vibrations and high-frequency electromagnetic interference. Optionally, spectral subtraction can be used for further noise reduction: during a period when the wind turbine is running stably and without alarms, a pure background noise sample is collected and its average power spectral density (PSD) is estimated; when processing subsequent signals, this noise spectrum is subtracted from the short-time power spectrum of each frame of the signal.

[0045] Optionally, for time-frequency conversion processing, the denoised signal is framed and windowed to reduce spectral leakage. A short-time Fourier transform is applied to each frame to calculate its complex spectrum, and the square of its amplitude is taken to obtain the power spectrum. Finally, the power spectra of consecutive frames are arranged along the time axis to form a power spectrum diagram, which serves as the preprocessed audio data. Optionally, a constant Q-transform is applied to the denoised signal, and the generated CQT spectrum diagram serves as the preprocessed audio data.

[0046] For example, acoustic features are extracted from preprocessed audio data. Optionally, acoustic features can be time-domain features. For instance, calculating the root mean square value (reflecting energy), kurtosis (reflecting impact characteristics), waveform factor (ratio of peak value to RMS value), and impulse factor (ratio of peak value to absolute average value) of the corresponding signal. These features are highly sensitive to transient impacts caused by bearing pitting, gear tooth breakage, etc. Optionally, acoustic features can be frequency-domain features. For instance, calculating the spectral centroid (reflecting the location of concentrated spectral energy), spectral roll-off point, and spectral entropy (reflecting the flatness or complexity of the spectrum). These features can characterize changes in the overall spectral structure caused by wear, imbalance, etc. Optionally, acoustic features can be time-frequency domain features. For instance, the preprocessed audio data is fed into a convolutional recurrent neural network (RNN), where the CNN (Convolutional Neural Network) part extracts the spatial (frequency) features of the spectrogram, the RNN (Recurrent Neural Network) part captures the dynamic evolution pattern in the time dimension, and finally the hidden state of the last time step of the RNN is used as the comprehensive time-frequency domain feature.

[0047] Understandably, by extracting multiple features across domains (time, frequency, and time-frequency), a comprehensive feature set describing the audio "fingerprint" is constructed. Time-domain features excel at capturing transient events, frequency-domain features reflect the overall energy distribution, and time-frequency domain features (such as MFCC or depth features) characterize the texture and dynamic patterns of the sound. This multi-perspective feature fusion ensures that regardless of whether the fault manifests as an impulse, harmonic shift, or complex modulation pattern, a corresponding, separable representation can be found in the feature space, greatly enhancing the generalization ability and diagnostic accuracy of subsequent classification models.

[0048] In one feasible implementation, the step of preprocessing the original audio signal to obtain preprocessed audio data includes steps S211 to S213: Step S211: Filter the original audio data to obtain the first audio data; Step S212: Perform frame segmentation and windowing processing on the first audio data to obtain the second audio data; Step S213: Perform time-frequency conversion on the second audio data to generate the corresponding spectrogram as the preprocessed audio data.

[0049] For example, the original audio data can be filtered using a pre-set filter to obtain the first audio data. For instance, a finite-length unit impulse response filter can be pre-set, and this filter can be applied to convolve the original audio data to filter out low-frequency vibration noise and high-frequency howling outside the passband. For example, the first audio data can be framed and windowed to obtain the second audio data. For instance, the continuous first audio data stream can be divided into segments of fixed length, called a frame. To ensure smooth transitions between frames and capture events that may occur at frame boundaries, an overlapping framing method is used, where the starting point of the next frame is offset from the starting point of the previous frame. The windowing operation involves multiplying each frame of data point-by-point by a window function (such as a Hanning window). The Hanning window smoothly transitions to zero at both ends, effectively reducing spectral leakage caused by directly truncating the signal (rectangular window), making the spectral analysis results of each frame more accurate.

[0050] For example, time-frequency conversion is performed on the second audio data, such as calculating the discrete Fourier transform of the windowed signal in each frame of the second audio data. The resulting complex spectrum is then converted to its modulus (amplitude spectrum) or its square (power spectrum). The power spectra of all consecutive frames are arranged in chronological order into a matrix, with rows corresponding to frequencies and columns corresponding to time frames. This matrix is ​​then visualized as an image, resulting in a linearly scaled spectrum.

[0051] Reference Figure 4This is a flowchart illustrating the fourth embodiment in this example. Content in this embodiment that is the same as or similar to the above embodiments can be referred to the above description, and will not be repeated hereafter. The method for detecting fan malfunctions also includes steps S100 to S200: Step S100: Based on the historical operating data of the observed wind turbines, construct a training sample set, wherein the observed wind turbines include the wind turbine and / or related wind turbines; Step S200: Iteratively train the initial detection model using the training sample set until the preset training conditions are met, and obtain the preset abnormal audio detection model.

[0052] It should be noted that in this embodiment, a high-quality training sample set will be constructed for the machine learning model.

[0053] For example, this can be done on the wind turbine itself, or by building a wind turbine cluster related to the turbine, such as a group of wind turbines of the same model and operating scenario. Obtain historical operating data of the turbine (or wind turbine cluster) since its commissioning, including: ① continuously or periodically recorded audio data; ② synchronized operating condition data (speed, power, temperature, etc.); ③ complete maintenance work order records. Strictly align these three types of data with their timestamps. Using each maintenance event as a key node, extract audio clips from before the failure (as positive samples) and audio clips from earlier, confirmed healthy periods under the same operating conditions (as negative samples).

[0054] Understandably, by systematically integrating historical data from multiple sources, a substantial initial data repository covering various operating states and failure modes was constructed. The introduction of the concept of "correlated wind turbines" effectively addresses the scarcity of single-device failure samples, providing the necessary data breadth for training a model with strong generalization ability and high robustness.

[0055] For example, the initial detection model is then subjected to supervised iterative training using the constructed training sample set to obtain the preset abnormal audio detection model. Optionally, the sample set can be randomly divided into a training set, a validation set, and a test set in a ratio (e.g., 7:2:1). An initial detection model architecture can be selected, such as a multilayer perceptron (input is a handcrafted feature vector), a one-dimensional convolutional neural network (input is a time-domain waveform frame), or a two-dimensional convolutional neural network (input is a spectrogram). Using the training set data, a defined loss function (e.g., cross-entropy loss) is minimized through a backpropagation algorithm and an optimizer (e.g., Adam). After each training round, the model performance is evaluated using the validation set, monitoring metrics such as accuracy and F1 score. Training is stopped when the validation set metrics no longer improve over several consecutive cycles (early stopping) or when the preset maximum number of iterations is reached. The model parameters that perform best on the validation set are selected, and finally, a final evaluation is performed on an independent test set to confirm its generalization ability, after which it is solidified as the preset abnormal audio detection model.

[0056] In one feasible implementation, the step of constructing a training sample set based on historical operating data of observed wind turbines includes steps S110 to S150: Step S110: For any abnormal record in the historical operation data, determine the abnormal component based on the abnormal cause in the abnormal record; Step S120: Extract the target audio segment from the historical audio data pointing to the abnormal component based on the preset abnormal observation window; Step S130: Based on the target audio segment, normal audio segments are selected from historical audio data. The difference between the normal audio segment and the target audio segment is greater than a preset threshold, and the wind turbine operating conditions corresponding to the normal audio segment are consistent with the wind turbine operating conditions corresponding to the target audio segment. Step S140: After obtaining normal audio segments, construct abnormal training samples in the training sample set based on the target audio segment, the wind turbine operating conditions corresponding to the target audio segment, and abnormal components; construct normal training samples in the training sample set based on normal audio segments, the wind turbine operating conditions corresponding to the normal audio segments, and abnormal components. Step S150: If the normal audio segment is not obtained through filtering, the abnormal record is discarded.

[0057] It should be noted that historical operational data typically includes a large number of anomalous records. The process of generating training samples for each anomalous record is largely the same. Therefore, in this embodiment, we will use one anomalous record as an example for explanation.

[0058] For example, for any abnormal record in historical maintenance data, a pre-developed fault cause parser can be used to parse the abnormal record. Optionally, the parser can be configured with a standardized component name dictionary (such as "drive-end bearing", "impeller", "motor winding") and a fault keyword library (such as "stripping", "crack", "overheating"). The parser performs word segmentation and keyword matching on unstructured maintenance work order text, automatically extracting the "fault object" and "fault phenomenon" and mapping them to standardized abnormal component identifiers. For example, the text "Replaced the non-drive-end bearing of wind turbine #1, found pitting on the inner ring" is parsed as the component identifier: Fan01_NDE_Bearing.

[0059] For example, specific audio segments are extracted from the anomaly log. Optionally, a pre-defined anomaly observation window [t_failure - T, t_failure] can be defined for backward look-back. The length of T can be empirically set based on the fault type (e.g., a short window for sudden fracture, a long window for chronic wear). All audio within the aforementioned time window is extracted from the historical audio data stream corresponding to the component (or, if using an array, the audio channel focused on the component via beamforming). To increase sample diversity, a sliding window (e.g., with a step size of 1 / 10 of the window length) can be used to extract multiple shorter (e.g., 2-second) target audio segments within the window. These segments are considered "positive" acoustic samples when the component experiences a specific fault.

[0060] For example, for each target audio segment, a paired sample operating under the same conditions but in a healthy state is found to distinguish between "normal and abnormal under the same conditions". Optionally, the average operating condition vector V_abnormal (e.g., [speed, power generation]) at the corresponding time point of the target audio segment is calculated. In the entire historical audio database (excluding all known fault periods and the time before and after), time periods where the Euclidean distance between the average operating condition vector and V_abnormal is less than a set tolerance (e.g., speed ±2%, power generation ±5%) are searched. From these time periods meeting the operating condition conditions, multiple candidate audio segments of the same duration are extracted. The Bavarian distance or Jaccard distance between each candidate segment and the target audio segment in a robust acoustic feature space (e.g., statistics of the first 13-dimensional MFCC coefficients) is calculated. A preset threshold is set, which can be determined by statistically analyzing the distribution of distances between randomly paired segments over all historical healthy periods (e.g., taking the 95th quantile). Only candidate segments whose distance to the target segment is greater than this threshold are retained as normal audio segments. This ensures that the selected normal samples and abnormal samples have statistically significant acoustic differences.

[0061] Understandably, through the dual constraints of "consistent operating conditions" and "acoustic differences," the selected normal samples and abnormal samples form a comparison pair of "control variables." This forces the model to learn the acoustic pattern changes truly caused by faults after excluding the influence of operating conditions, greatly improving the model's feature discrimination ability and robustness in actual variable operating condition applications, and effectively preventing the model from misjudging normal sounds under certain specific operating conditions as faults.

[0062] Furthermore, after selecting normal audio segments, abnormal training samples are constructed in the training sample set based on the target audio segment, the corresponding wind turbine operating condition, and the abnormal component. Optionally, the abnormal training sample is: {"audio_data": target audio segment (or its feature vector), "operating_condition": V_abnormal, "component_id": abnormal component identifier, "label": specific fault type code}. Normal training samples are constructed in the training sample set based on normal audio segments, the corresponding wind turbine operating condition, and the abnormal component. Optionally, normal training samples are: {"audio_data": selected normal audio segment (or its feature vector), "operating_condition": V_normal (should be highly consistent with V_abnormal), "component_id": abnormal component identifier, "label": "normal" code}. All constructed samples are added to the overall training sample set.

[0063] It is understood that in this embodiment, each sample not only contains core audio data (or features), but also is bound to the operating context at the time of its occurrence and the identity of the component to which it belongs. This rich annotation information enables the model to perform conditional learning, and is a key data format for achieving high-precision, component-level, and operating condition-adaptive fault diagnosis.

[0064] Furthermore, if no normal audio segments are found during the screening process, abnormal records are discarded. Specifically, during the search and screening process in step S130, if no normal audio segment that simultaneously meets both the operating condition matching and acoustic difference thresholds can be found in the entire compliance history database for a given target audio segment, the matching is deemed a failure. The system automatically discards this abnormal record and all its corresponding target audio segments from the current training sample construction process, preventing them from being added to the training sample set. This avoids introducing abnormal samples lacking clear and comparable health baselines. Such samples typically originate from incomplete historical data records, unique operating conditions that are difficult to match, or fault characteristics that are too weak to be easily confused with normal conditions. Therefore, discarding such samples improves the quality of the entire training set, thereby ensuring that the final trained model has higher reliability and accuracy.

[0065] The following is for reference. Figure 5 It shows a schematic diagram of a structure suitable for implementing an electronic device according to the embodiments of this application. 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.

[0066] like Figure 5 As shown, the electronic device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the electronic device. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touchscreen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 1003 including, for example, magnetic tape, hard disk, etc.; and communication devices 1009. Communication device 1009 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although electronic devices with various systems are shown in the figures, it should be understood that implementation or possession of all the systems shown is not required. More or fewer systems may be implemented alternatively.

[0067] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0068] The electronic device provided in this application, employing the fan anomaly detection method described in the above embodiments, can solve the technical problem that traditional monitoring methods have limited detection effects and cannot meet the needs of practical applications. Compared with the prior art, the beneficial effects of the electronic device provided in this application are the same as those of the fan anomaly detection method provided in the above embodiments, and other technical features of this electronic device are the same as those disclosed in the method of the previous embodiment, and will not be repeated here.

[0069] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0070] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0071] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the fan anomaly detection method in the above embodiments.

[0072] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0073] The aforementioned computer-readable storage medium may be included in an electronic device or may exist independently without being assembled into an electronic device.

[0074] The aforementioned computer-readable storage medium carries one or more programs, which, when executed by an electronic device, cause the electronic device to: The raw audio data generated by the target detection component in the wind turbine during operation is collected using an audio sensor array. Acoustic features are extracted from the raw audio data and correlated with the identification of the target detection component and the operating parameters of the fan to obtain the acoustic features to be detected. The acoustic features to be detected are input into a preset abnormal audio detection model, and the abnormal detection results of the target detection component in the running state are output.

[0075] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0076] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0077] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0078] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described method for detecting wind turbine anomalies. This solves the technical problem that traditional monitoring methods have limited detection effectiveness and cannot meet practical application needs. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the wind turbine anomaly detection method provided in the above embodiments, and will not be elaborated upon here.

[0079] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the wind turbine anomaly detection method described above.

[0080] The computer program product provided in this application can solve the technical problem that traditional monitoring methods have limited detection effects and cannot meet the needs of practical applications. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the fan anomaly detection method provided in the above embodiments, and will not be repeated here.

[0081] The above are only some embodiments of this application and do not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A method for detecting fan malfunctions, characterized in that, The fan is equipped with an audio sensor array; The method for detecting fan malfunctions includes: The audio sensor array is used to collect raw audio data generated by the target detection component in the wind turbine during operation. Acoustic features are extracted from the original audio data, and the acoustic features are associated with the identifier of the target detection component and the operating parameters of the fan to obtain the acoustic features to be detected. The acoustic features to be detected are input into a preset abnormal audio detection model, and the abnormal detection results of the target detection component in the running state are output.

2. The method for detecting fan malfunctions as described in claim 1, characterized in that, The audio sensor array includes a first type of microphone and a second type of microphone, the first type of microphone and the second type of microphone being located in different positions; The step of acquiring raw audio data generated by the target detection component in the wind turbine during operation through the audio sensor array includes: First audio data is acquired using the first type of microphone, and second audio data is acquired using the second type of microphone. Based on the delay compensation parameter corresponding to the target detection component, the first audio data and the second audio data are fused to obtain the original audio data generated by the target detection component in operation. The delay compensation parameter is determined based on the spatial positional relationship between the target detection component and the first type of microphone and the second type of microphone.

3. The method for detecting fan malfunctions as described in claim 2, characterized in that, The step of fusing the first audio data and the second audio data based on the delay compensation parameters corresponding to the target detection component to obtain the original audio data generated by the target detection component in operation includes: From the first audio data and the second audio data, audio data to be delayed and non-delayed audio data are determined, wherein the microphone corresponding to the audio data to be delayed is closer to the target detection component than the microphone corresponding to the non-delayed audio data; The delayed audio data is obtained by delaying the audio data to be delayed using the delay compensation parameters. The delayed audio data is superimposed with the non-delayed audio data to obtain the original audio data.

4. The method for detecting fan malfunctions as described in claim 1, characterized in that, The step of extracting acoustic features from the original audio data includes: The original audio signal is preprocessed to obtain preprocessed audio data. The preprocessing includes at least noise reduction and time-frequency conversion. Acoustic features are extracted from the preprocessed audio data, wherein the types of acoustic features include time-domain features, frequency-domain features, and time-frequency-domain features.

5. The method for detecting fan malfunctions as described in claim 4, characterized in that, The step of preprocessing the original audio signal to obtain preprocessed audio data includes: filtering the original audio data to obtain first audio data; performing frame segmentation and windowing on the first audio data to obtain second audio data; and performing time-frequency conversion on the second audio data to generate a corresponding spectrogram as the preprocessed audio data.

6. The method for detecting fan malfunctions as described in claim 1, characterized in that, The method for detecting fan malfunctions also includes: Based on the historical operating data of the observed wind turbines, a training sample set is constructed, wherein the observed wind turbines include the wind turbine and / or related wind turbines; The initial detection model is iteratively trained using the training sample set until the preset training conditions are met, thus obtaining the preset abnormal audio detection model.

7. The method for detecting fan malfunctions as described in claim 6, characterized in that, The steps for constructing the training sample set based on the historical operating data of the observed wind turbines include: For any abnormal record in the historical operation data, the abnormal component is determined based on the cause of the abnormality in the abnormal record; The target audio segment is extracted from historical audio data pointing to the abnormal component based on a preset anomaly observation window; Based on the target audio segment, normal audio segments are filtered from the historical audio data, wherein the difference between the normal audio segment and the target audio segment is greater than a preset threshold, and the wind turbine operating condition corresponding to the normal audio segment is consistent with the wind turbine operating condition corresponding to the target audio segment. In the case of obtaining the normal audio segment, based on the target audio segment, the wind turbine operating condition corresponding to the target audio segment, and the abnormal component, an abnormal training sample is constructed in the training sample set, and based on the normal audio segment, the wind turbine operating condition corresponding to the normal audio segment, and the abnormal component, a normal training sample is constructed in the training sample set. If the normal audio segment is not obtained through filtering, the abnormal record is discarded.

8. An electronic device, characterized in that, The electronic device includes: a processor, a memory, and a fan malfunction detection program stored in the memory and executable on the processor, wherein the fan malfunction detection program, when executed, implements the steps of the fan malfunction detection method as described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a fan malfunction detection program, which, when executed, implements the steps of the fan malfunction detection method as described in any one of claims 1-7.

10. A computer program product, characterized in that, The computer program product includes a fan malfunction detection program, which, when executed by a processor, implements the steps of the fan malfunction detection method as described in any one of claims 1-7.

Images