A method and system for acoustic print detection of a wind turbine gearbox

Abstract

This invention relates to the field of voiceprint detection technology, specifically to a method and system for voiceprint detection of wind turbine gearboxes. The method constructs a baseline voiceprint feature library covering various health states and operating conditions, and trains an adaptive classification model and a transfer learning baseline model based on this library. This enables adaptive intelligent detection of gearbox voiceprints under varying operating conditions. Furthermore, by simply acquiring the real-time voiceprint signal of the gearbox to be detected and extracting its real-time features, the trained classification model for the corresponding operating condition can be used for rapid matching and diagnosis, outputting abnormal voiceprint features and fault identification results. This eliminates the need for retraining or manual intervention, significantly improving the operating condition adaptability and real-time response capability of the voiceprint detection technology.

Description

Technical Field

[0001] This invention belongs to the field of voiceprint detection technology, specifically relating to a voiceprint detection method and system for wind turbine gearboxes. Background Technology

[0002] The gearbox of a wind turbine is the core component that transmits wind energy and converts rotational speed; its operating status directly determines the safety and stability of the wind turbine. Because wind turbines operate in complex outdoor environments for extended periods, facing strong wind noise, temperature variations, vibration, and shocks, gearboxes are prone to problems such as tooth wear, loose bearings, and insufficient lubrication. Failure to detect and address these issues promptly can lead to gearbox damage, resulting in generator shutdowns and significant economic losses.

[0003] Currently, acoustic signature detection methods for wind turbine gearboxes have been applied to some extent, but existing technologies still have many shortcomings: acoustic signature feature extraction is greatly affected by environmental noise, and traditional single feature extraction methods are difficult to effectively highlight the acoustic signature features of early minor faults in the gearbox, resulting in low accuracy in early fault identification; existing detection methods mostly use fixed acoustic signature feature libraries, which cannot adapt to the characteristics of wind turbine generators operating under varying conditions (different wind speeds, loads, and speeds), and are prone to false alarms and missed alarms under varying conditions, and have poor adaptability to different gearbox models; existing methods can only achieve fault identification, but cannot provide early warning of early faults, nor can they trace the cause of the fault, making it difficult to provide accurate decision-making basis for operation and maintenance work, resulting in low operation and maintenance efficiency and failing to effectively prevent the fault from escalating. Summary of the Invention

[0004] In view of the above-mentioned problems existing in the gearboxes of existing wind turbine generator sets, the purpose of this invention is to provide a method and system for soundprint detection of wind turbine generator set gearboxes.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: On one hand, the present invention provides a method for acoustic fingerprint detection of a wind turbine gearbox, comprising: Step S1: Collect historical acoustic signature signals of the wind turbine gearbox under various known health conditions and various operating conditions, and preprocess the historical acoustic signature signals to obtain historical acoustic signature operation signals; Step S2: Extract multi-dimensional voiceprint features from historical voiceprint operation signals using dual channels independently, and classify them according to operating conditions and fault categories based on the multi-dimensional voiceprint features to construct a benchmark voiceprint feature library. Step S3: Based on the labeled samples in the benchmark voiceprint feature library, train the working condition adaptive classification model and pre-train the transfer learning baseline model based on historical voiceprint operation signals. Step S4: Collect the real-time acoustic signature signal of the gearbox of the wind turbine generator set to be tested, perform preprocessing and feature extraction to obtain the real-time feature vector; Step S5: Combine the real-time operating conditions corresponding to the real-time feature vectors, call the operating condition adaptive classification model corresponding to the real-time operating conditions, match the real-time feature vectors with the benchmark acoustic signature feature library, and output the acoustic signature abnormal features and fault identification results of the gearbox of the wind turbine generator set to be detected.

[0006] Preferably, the voiceprint detection method further includes: if the model of the gearbox of the wind turbine generator to be detected is different from the gearbox model corresponding to the historical voiceprint signal, then the transfer learning baseline model is enabled, and the real-time feature vector is mapped into an embedded vector and matched with the prototype of a small number of labeled samples in the target domain to achieve cross-model generalization recognition.

[0007] Preferably, in step S1, historical acoustic signature signals of the wind turbine gearbox under various known health conditions and various operating conditions are collected, and the historical acoustic signature signals are preprocessed, including: pre-emphasis processing of the historical acoustic signature signals in sequence to compensate for high frequency attenuation, frame-by-frame windowing processing to obtain short-time stable signal frames, and noise suppression using the least mean square adaptive filtering algorithm to eliminate background noise interference from wind noise, mechanical vibration and tower echo.

[0008] Preferably, the multiple known health conditions include at least the normal operating condition, tooth surface wear condition, tooth root crack condition, tooth surface pitting condition, and bearing failure condition; the multiple operating conditions include different speed ranges and different load ranges.

[0009] Preferably, in step S2, multi-dimensional voiceprint features are extracted from historical voiceprint operation signals using dual channels independently, including: constructing a first filter bank and a second filter bank that operate in parallel; the first filter bank uses a gamma-pass filter bank to extract gamma-pass cepstral coefficients and their first and second order difference spectra; the second filter bank uses a filter bank based on power spectrum normalization to extract power regularized cepstral coefficients; and the feature vectors output by the first filter bank and the second filter bank are spliced ​​and fused to form multi-dimensional voiceprint features.

[0010] Preferably, a benchmark voiceprint feature library is constructed by classifying multidimensional voiceprint features according to operating conditions and fault categories. This includes: performing clustering processing on multidimensional voiceprint features to determine the feature centers and distribution parameters corresponding to various operating conditions and faults, thereby constructing the benchmark voiceprint feature library.

[0011] Preferably, in step S3, a working condition adaptive classification model is trained based on the labeled samples in the benchmark voiceprint feature library, including: optimizing the parameters of the K-nearest neighbor classification model using the planetary optimization algorithm. The planetary optimization algorithm uses the classification accuracy as the fitness function to automatically search for the optimal K value and distance metric parameters to obtain an optimized KNN classifier that is adaptive to different working conditions.

[0012] Preferably, a transfer learning baseline model is pre-trained based on historical voiceprint running signals, including: training a Siamese network with a triplet loss function, the Siamese network outputting an embedding vector to characterize voiceprint similarity.

[0013] Secondly, the present invention also provides an acoustic signature detection system for a wind turbine gearbox, comprising: The historical data acquisition module is used to collect historical acoustic fingerprint signals of the wind turbine gearbox under various known health conditions and various operating conditions, and to preprocess the historical acoustic fingerprint signals to obtain historical acoustic fingerprint operation signals. The benchmark feature library construction module is used to extract multi-dimensional voiceprint features from historical voiceprint operation signals using dual channels independently, and classify them according to operating conditions and fault categories based on the multi-dimensional voiceprint features to construct the benchmark voiceprint feature library. The model building and training module is used to train an adaptive classification model based on labeled samples in the benchmark voiceprint feature library, and to pre-train a transfer learning baseline model based on historical voiceprint operation signals. The real-time data acquisition module is used to acquire the real-time acoustic signature signal of the gearbox of the wind turbine generator set under test, perform preprocessing and feature extraction to obtain the real-time feature vector; The intelligent diagnostic matching module is used to combine the real-time feature vector with the real-time operating conditions, call the operating condition adaptive classification model corresponding to the real-time operating conditions, match the real-time feature vector with the benchmark acoustic signature feature library, and output the acoustic signature abnormal features and fault identification results of the gearbox of the wind turbine generator set to be tested.

[0014] Thirdly, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the acoustic fingerprint detection method for the gearbox of a wind turbine generator set as described in the first aspect.

[0015] Compared with the prior art, the present invention has at least the following beneficial technical effects: This invention discloses a method and system for acoustic signature detection of wind turbine gearboxes. By constructing a benchmark acoustic signature feature library covering various health states and operating conditions, and training an adaptive classification model and a transfer learning baseline model based on this library, the invention achieves adaptive intelligent detection of gearbox acoustic signatures under varying operating conditions. Furthermore, by simply acquiring the real-time acoustic signature signal of the gearbox to be detected and extracting its real-time features, the trained classification model for the corresponding operating condition can be used for rapid matching and diagnosis, outputting abnormal acoustic signature features and fault identification results without retraining or manual intervention. This significantly improves the operating condition adaptability and real-time response capability of the acoustic signature detection technology. Simultaneously, through the pre-trained transfer learning baseline model, this acoustic signature detection method can effectively support generalized recognition across models and turbine units, solving the practical deployment difficulties of traditional acoustic signature detection methods in wind power sites due to scarce samples, variable operating conditions, and large differences in turbine models. This provides a highly robust and versatile technical solution for early fault warning and intelligent operation and maintenance of wind turbine gearboxes. Attached Figure Description

[0016] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a schematic flowchart of the voiceprint detection method according to Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the voiceprint detection system according to Embodiment 2 of the present invention; Figure 3 This is a schematic diagram of the network-side server structure in Embodiment 3 of the present invention. Detailed Implementation

[0018] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0019] In the description of this invention, it should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0020] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0021] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0022] The accompanying drawings illustrate various structural schematic diagrams according to embodiments disclosed in this invention. These drawings are not to scale, and some details have been enlarged for clarity, and some details may have been omitted. The shapes of the various regions and layers shown in the drawings, as well as their relative sizes and positional relationships, are merely exemplary and may deviate from reality due to manufacturing tolerances or technical limitations. Furthermore, those skilled in the art can design regions / layers with different shapes, sizes, and relative positions as needed.

[0023] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0024] Example 1 like Figure 1 As shown, this embodiment provides a method for acoustic fingerprint detection of a wind turbine gearbox. The method includes: Step S1: Collect historical acoustic signature signals of the wind turbine gearbox under various known health conditions and various operating conditions, and preprocess the historical acoustic signature signals to obtain historical acoustic signature operation signals.

[0025] In this embodiment, the various known health states include at least the normal operating state, tooth surface wear state, tooth root crack state, tooth surface pitting state, and bearing failure state. Specifically, the normal state refers to a newly manufactured or recently overhauled gearbox with operating parameters that meet design standards. The tooth surface wear state refers to data collected from units with confirmed wear through accelerated wear testing or on-site collection. The tooth root crack state refers to data collected from gears installed on a testing platform to simulate crack failures through wire cutting or electrical discharge machining. The tooth surface pitting state refers to data collected from gears installed on a testing platform to simulate pitting failures through electrochemical corrosion or mechanical impact. The bearing failure state refers to data collected from three sub-states: inner ring crack, outer ring crack, and cage fracture.

[0026] As an optional embodiment, multiple operating conditions include different speed ranges and different load ranges; under each healthy state, data can be collected under the following three speed ranges and three load range combinations: speed range: 30%, 70%, and 100% of rated speed; corresponding load ranges: 0% (no load), 50% (half load), and 100% (full load) of rated load; a total of 5 healthy states * 3 speeds * 3 loads can be collected, resulting in a total of 45 operating condition combinations. Each combination should collect no less than 50 samples, for a total of no less than 2250 samples.

[0027] As an optional embodiment, the preprocessing steps specifically include: performing pre-emphasis processing on the historical acoustic signature signal in sequence to compensate for high frequency attenuation, performing frame-by-frame windowing processing to obtain short-time stable signal frames, and using the least mean square adaptive filtering algorithm to suppress noise and eliminate background noise interference from wind noise, mechanical vibration and tower echo.

[0028] In this embodiment, specifically, pre-emphasis refers to using a first-order high-pass filter to compensate for the high-frequency attenuation of sound waves propagating in the air, thereby increasing the energy proportion of high-frequency fault characteristics such as tooth surface impact. The first-order high-pass filter can be represented as: ; In this embodiment, frame segmentation and windowing refers to using a frame segmentation strategy with a frame length of 25ms and a frame shift of 10ms, and applying a Hamming window to each frame to reduce spectral leakage. The Hamming window can be represented as: ; In this embodiment, LMS adaptive filtering noise suppression refers to using the background noise collected by the cabin far-field microphone as a reference signal. The noisy signal acquired by the contact sensor is used as the desired signal. Using step size factor The LMS algorithm with filter order L=128 iteratively updates the weight coefficients and outputs the denoised signal. In this embodiment, after the above preprocessing, historical voiceprint operation signals are obtained for subsequent feature extraction.

[0029] Step S2: Extract multi-dimensional voiceprint features from historical voiceprint operation signals using dual channels independently, and classify them according to operating conditions and fault categories based on the multi-dimensional voiceprint features to construct a benchmark voiceprint feature library.

[0030] In this embodiment, the steps for independently extracting multidimensional voiceprint features through dual channels are as follows: a first filter bank and a second filter bank are constructed to run in parallel. The first filter bank uses a gamma-pass filter bank to extract gamma-pass cepstral coefficients and their first and second order difference spectra. The second filter bank uses a filter bank based on power spectrum normalization to extract power regularized cepstral coefficients. The feature vectors output by the first filter bank and the second filter bank are spliced ​​and fused to form multidimensional voiceprint features.

[0031] As an optional embodiment, the Gamma-Ton cepstral coefficients (GTCC) enhance the sensitivity to early, weak fault characteristics of the gearbox by simulating the filtering characteristics of the basilar membrane of the human ear; the Power Regularized Cepstral Coefficients (PNCC) enhance the anti-interference capability of the voiceprint features in a strong wind noise environment by normalizing the power spectrum and processing noise robustness; the dual-channel hybrid feature formed by the fusion of the two has both noise robustness and sensitivity to weak faults.

[0032] As an optional embodiment, the steps for constructing a benchmark voiceprint feature library include: performing clustering processing on multidimensional voiceprint features to determine the feature centers and distribution parameters corresponding to various operating conditions and various faults, thereby constructing a benchmark voiceprint feature library.

[0033] Step S3: Based on the labeled samples in the benchmark voiceprint feature library, train the working condition adaptive classification model, and pre-train the transfer learning baseline model based on historical voiceprint operation signals.

[0034] In this embodiment, the steps of training the working condition adaptive classification model specifically include: optimizing the parameters of the K-nearest neighbor classification model using the planetary optimization algorithm. The planetary optimization algorithm uses classification accuracy as the fitness function to automatically search for the optimal K value and distance metric parameters to obtain an optimized KNN classifier that is adaptive to different working conditions.

[0035] In this embodiment, the steps of pre-training the transfer learning baseline model specifically include: training a Siamese network with a triplet loss function, the Siamese network outputting an embedding vector to characterize voiceprint similarity.

[0036] Step S4: Collect the real-time acoustic signature signal of the gearbox of the wind turbine generator set to be tested, perform preprocessing and feature extraction to obtain the real-time feature vector.

[0037] In this embodiment, the same sensor arrangement, sampling parameters, and preprocessing procedure as in step S1 are used on the gearbox of the wind turbine generator under test to collect real-time acoustic signature signals. Feature extraction also adopts the dual-channel parallel extraction method in step S2: GTCC and its differential spectrum (72-dimensional) are extracted through a gamma-pass filter bank, and PNCC and its differential spectrum (72-dimensional) are extracted through a power spectrum normalization filter bank, splicing them to form a 144-dimensional real-time hybrid feature vector. Simultaneously, the current gearbox speed (in rpm) and load (in kW) are read in real-time by the on-site PLC system as real-time operating parameters.

[0038] Step S5: Combine the real-time operating conditions corresponding to the real-time feature vectors, call the operating condition adaptive classification model corresponding to the real-time operating conditions, match the real-time feature vectors with the benchmark acoustic signature feature library, and output the acoustic signature abnormal features and fault identification results of the gearbox of the wind turbine generator set to be detected.

[0039] In this embodiment, the operating condition range is calculated based on the real-time collected speed value (e.g., 1180 rpm) and load value (e.g., 1.2 MW). Specifically, the speed range is divided into: ≤600 rpm is classified as low speed, 601-1000 rpm as medium speed, and ≥1001 rpm as high speed. The load range is divided into: ≤30% is light load, 31%-70% is medium load, and ≥71% is heavy load. The real-time operating condition (1180 rpm, 1.2 MW) is determined to be in the high-speed-heavy load range.

[0040] In this embodiment, the specific steps for matching the real-time feature vector with the benchmark voiceprint feature library include: retrieving the classifier corresponding to the high-speed-heavy-load operating condition from the set of adaptive KNN classifiers trained in step S3; and matching the real-time hybrid feature vector obtained in step S4. The Mahalanobis distance is calculated between the samples and the centers of all categories within the high-speed-heavy-load condition group in the baseline acoustic signature feature library constructed in step S2. The categories of the K samples with the smallest distances are then selected for majority voting to obtain the fault classification results. The calculation is as follows: .

[0041] in, Characterization Mahalanobis distance to class c Characterizing real-time blended feature vectors, The mean vector representing category c.

[0042] Then, based on the voting results, the following outputs are provided: Fault type: for example, "tooth surface wear"; Fault level: classified according to the degree of deviation of the distance value: mild warning (distance > 2σ), moderate alarm (distance > 3σ), severe fault (distance > 4σ); Fault location: if the fault type is bearing fault, further locate to the inner ring, outer ring or cage based on characteristic frequency matching.

[0043] As an optional embodiment, if the model of the gearbox of the wind turbine generator to be detected is different from the gearbox model corresponding to the historical voiceprint signal, the transfer learning baseline model is enabled. The real-time feature vector is mapped to the embedded vector and then matched with the prototype of a small number of labeled samples in the target domain to achieve cross-model generalization recognition.

[0044] As an optional embodiment, after step S5, the voiceprint detection method may further include performing time alignment and correlation analysis on the abnormal voiceprint features output in step S5 with the speed fluctuations, gearbox temperature, and load fluctuations obtained from the SCADA system; dynamically calculating the warning threshold under the current operating condition using a multivariate Gaussian regression model trained on historical normal data; and triggering a warning when the deviation of the voiceprint features exceeds the dynamic threshold. Simultaneously, the abnormal features are input into a fault tracing database (which stores historical fault cases and their corresponding voiceprint feature patterns and wear curves), and the most similar historical cases are retrieved using a cosine similarity matching algorithm. The method then outputs fault cause inferences, suggested maintenance measures, and estimated remaining lifespan.

[0045] In this embodiment, to further verify the feasibility and superiority of this embodiment, on-site testing was conducted at a wind farm. The planetary gearbox of a 1.5MW doubly fed wind turbine generator set was selected as the test object. The gearbox had been running continuously for 3 years and had potential early failures.

[0046] I. On-site Implementation Process First, three acoustic signature sensors are arranged in the gearbox housing, high-speed shaft bearing housing, and low-speed shaft position to synchronously collect speed and load SCADA data. Following step S1 in this embodiment, pre-emphasis, frame-by-frame windowing, and LMS adaptive filtering are performed to effectively suppress wind noise and tower echo noise, achieving a noise suppression ratio of 28dB. Second, dual-channel feature extraction is used, fusing 144-dimensional features from GTCC and PNCC, and a high-speed-heavy-load optimized KNN classifier is retrieved and matched with the benchmark acoustic signature feature library using Mahalanobis distance. Finally, the real-time detection output shows a fault type of slight tooth surface wear and a fault level of mild warning. Upon inspection after shutdown and opening the cover, slight wear marks were found on the gear teeth, completely consistent with the diagnostic results.

[0047] II. Comparison of Implementation Results Data

[0048] III. Application Conclusions In practical applications in wind farms, this embodiment accurately identifies early gear wear faults, solving the problems of false alarms under varying operating conditions and missed detection of early faults in traditional methods. It significantly improves detection efficiency and accuracy, and can directly support intelligent operation and maintenance of wind turbine gearboxes, demonstrating strong field feasibility and technical superiority.

[0049] This embodiment provides a method for acoustic signature detection of wind turbine gearboxes. This method constructs a baseline acoustic signature feature library covering various health states and operating conditions, and trains an adaptive classification model and a transfer learning baseline model based on this library. This enables adaptive intelligent detection of gearbox acoustic signatures under varying operating conditions. Furthermore, by simply acquiring the real-time acoustic signature signal of the gearbox to be detected and extracting its real-time features, the trained classification model for the corresponding operating condition can be used for rapid matching and diagnosis, outputting abnormal acoustic signature features and fault identification results without retraining or manual intervention. This significantly improves the operating condition adaptability and real-time response capability of the acoustic signature detection technology. Simultaneously, through the pre-trained transfer learning baseline model, this acoustic signature detection method can effectively support generalized recognition across models and turbine units, solving the practical deployment difficulties of traditional acoustic signature detection methods in wind power sites due to scarce samples, variable operating conditions, and significant turbine model differences. This provides a highly robust and versatile technical solution for early fault warning and intelligent operation and maintenance of wind turbine gearboxes.

[0050] Example 2 like Figure 3 As shown, this embodiment provides a soundprint detection system 100 for a wind turbine gearbox, including a historical data acquisition module 11, a benchmark feature library construction module 12, a model construction and training module 13, a real-time data acquisition module 14, and an intelligent diagnosis and matching module 15.

[0051] The historical data acquisition module 11 is used to acquire historical acoustic fingerprint signals of the wind turbine gearbox under various known health conditions and various operating conditions, and to preprocess the historical acoustic fingerprint signals to obtain historical acoustic fingerprint operation signals.

[0052] The benchmark feature library construction module 12 is used to extract multi-dimensional voiceprint features from historical voiceprint operation signals using dual channels independently, and classify them according to operating conditions and fault categories based on the multi-dimensional voiceprint features to construct a benchmark voiceprint feature library.

[0053] The model building training module 13 is used to train an adaptive classification model based on labeled samples in the benchmark voiceprint feature library, and to pre-train a transfer learning baseline model based on historical voiceprint operation signals.

[0054] The real-time data acquisition module 14 is used to acquire the real-time acoustic signature signal of the gearbox of the wind turbine generator set to be tested, perform preprocessing and feature extraction to obtain the real-time feature vector.

[0055] The intelligent diagnostic matching module 15 is used to combine the real-time feature vector with the real-time operating conditions, call the operating condition adaptive classification model corresponding to the real-time operating conditions, match the real-time feature vector with the benchmark acoustic signature feature library, and output the acoustic signature abnormal features and fault identification results of the gearbox of the wind turbine generator set to be detected.

[0056] This embodiment provides an acoustic signature detection system for wind turbine gearboxes. This system is based on the acoustic signature detection method for wind turbine gearboxes provided in Embodiment 1 above; therefore, specific details will not be repeated. The acoustic signature detection system for wind turbine gearboxes in this embodiment constructs a baseline acoustic signature feature library covering multiple health states and operating conditions. Based on this library, it trains an adaptive classification model and a transfer learning baseline model, achieving adaptive intelligent detection of gearbox acoustic signatures under varying operating conditions. Furthermore, it only requires acquiring the real-time acoustic signature signal of the gearbox to be detected and extracting features; that is, it uses the trained classification model corresponding to the operating condition for rapid matching and diagnosis, without the need for retraining or manual intervention, significantly improving the operating condition adaptability and real-time response capability of the acoustic signature detection system.

[0057] Example 3 The third embodiment of the present invention relates to a network-side server, such as... Figure 3 As shown, it includes at least one processor 302; and a memory 301 communicatively connected to at least one processor 302; wherein the memory 301 stores instructions executable by at least one processor 302, the instructions being executed by at least one processor 302 to enable at least one processor 302 to execute the above-described acoustic fingerprint detection method for wind turbine gearbox.

[0058] The memory 301 and processor 302 are connected via a bus, which may include any number of interconnecting buses and bridges. The bus connects various circuits of one or more processors 302 and memory 301 together. The bus can also connect various other circuits, such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. A bus interface provides an interface between the bus and the transceiver. The transceiver can be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by processor 302 is transmitted over a wireless medium via an antenna, which further receives data and transmits it to processor 302.

[0059] Processor 302 is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory 301 can be used to store data used by processor 302 during operation.

[0060] Example 4 This embodiment relates to a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the acoustic fingerprint detection method for the gearbox of a wind turbine generator set described in Embodiment 1 above.

[0061] That is, those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0062] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the scope of the invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0063] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can be appropriately combined to form other embodiments that can be understood by those skilled in the art. The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.

Claims

1. A method of acoustic print detection for a wind turbine gearbox, characterized by, include: Step S1: Collect historical acoustic signature signals of the wind turbine generator gearbox under various known health conditions and various operating conditions, and preprocess the historical acoustic signature signals to obtain historical acoustic signature operation signals; Step S2: Extract multi-dimensional voiceprint features from historical voiceprint operation signals using dual channels independently, and classify them according to operating conditions and fault categories based on the multi-dimensional voiceprint features to construct a benchmark voiceprint feature library. Step S3: Based on the labeled samples in the benchmark voiceprint feature library, train the working condition adaptive classification model and pre-train the transfer learning baseline model based on historical voiceprint operation signals. Step S4: Collect the real-time acoustic signature signal of the gearbox of the wind turbine generator set to be tested, perform preprocessing and feature extraction to obtain the real-time feature vector; Step S5: Combine the real-time operating conditions corresponding to the real-time feature vectors, call the operating condition adaptive classification model corresponding to the real-time operating conditions, match the real-time feature vectors with the benchmark acoustic signature feature library, and output the acoustic signature abnormal features and fault identification results of the gearbox of the wind turbine generator set to be detected.

2. The acoustic fingerprinting method of a wind turbine gearbox according to claim 1, characterized in that, The voiceprint detection method also includes: if the model of the gearbox of the wind turbine generator to be detected is different from the gearbox model corresponding to the historical voiceprint signal, then the transfer learning baseline model is enabled, and the real-time feature vector is mapped into an embedded vector and matched with the prototype of a small number of labeled samples in the target domain to achieve cross-model generalization recognition.

3. The acoustic fingerprinting method of a wind turbine gearbox according to claim 1, characterized in that, In step S1, historical acoustic signature signals of the wind turbine gearbox under various known health conditions and various operating conditions are collected. The historical acoustic signature signals are preprocessed, including: pre-emphasis processing of the historical acoustic signature signals to compensate for high frequency attenuation, frame-by-frame windowing processing to obtain short-time stable signal frames, and noise suppression using the least mean square adaptive filtering algorithm to eliminate background noise interference from wind noise, mechanical vibration and tower echo.

4. The acoustic fingerprinting method of a wind turbine gearbox according to claim 3, characterized in that, The various known health conditions include at least normal operation, tooth surface wear, tooth root crack, tooth surface pitting, and bearing failure; the various operating conditions include different speed ranges and different load ranges.

5. The acoustic fingerprinting method of a wind turbine gearbox according to claim 1, characterized in that, In step S2, multi-dimensional voiceprint features are extracted from historical voiceprint operation signals using dual channels independently. This includes: constructing a first filter bank and a second filter bank that operate in parallel. The first filter bank uses a gamma-pass filter bank to extract gamma-pass cepstral coefficients and their first and second order difference spectra. The second filter bank uses a filter bank based on power spectrum normalization to extract power regularized cepstral coefficients. The feature vectors output by the first filter bank and the second filter bank are spliced ​​and fused to form multi-dimensional voiceprint features.

6. The acoustic fingerprinting method of a wind turbine gearbox according to claim 5, characterized in that, Based on multidimensional voiceprint features, a benchmark voiceprint feature library is constructed by classifying them according to operating conditions and fault categories. This includes: performing clustering processing on multidimensional voiceprint features to determine the feature centers and distribution parameters corresponding to various operating conditions and faults, thereby constructing the benchmark voiceprint feature library.

7. The acoustic fingerprinting method of a wind turbine gearbox according to claim 1, characterized in that, In step S3, an adaptive classification model for working conditions is trained based on the labeled samples in the benchmark voiceprint feature library. This includes: optimizing the parameters of the K-nearest neighbor classification model using the planetary optimization algorithm. The planetary optimization algorithm uses the classification accuracy as the fitness function to automatically search for the optimal K value and distance metric parameters to obtain an optimized KNN classifier that is adaptive to different working conditions.

8. The acoustic fingerprinting method of a wind turbine gearbox according to claim 7, characterized in that, The transfer learning baseline model is pre-trained based on historical voiceprint operation signals, including: training a Siamese network with a triplet loss function, which outputs an embedding vector to characterize voiceprint similarity.

9. A soundprint detection system for a wind turbine gearbox, characterized by, include: The historical data acquisition module is used to collect historical acoustic fingerprint signals of the wind turbine gearbox under various known health conditions and various operating conditions, and to preprocess the historical acoustic fingerprint signals to obtain historical acoustic fingerprint operation signals. The benchmark feature library construction module is used to extract multi-dimensional voiceprint features from historical voiceprint operation signals using dual channels independently, and classify them according to operating conditions and fault categories based on the multi-dimensional voiceprint features to construct the benchmark voiceprint feature library. The model building and training module is used to train an adaptive classification model based on labeled samples in the benchmark voiceprint feature library, and to pre-train a transfer learning baseline model based on historical voiceprint operation signals. The real-time data acquisition module is used to acquire the real-time acoustic signature signal of the gearbox of the wind turbine generator set under test, perform preprocessing and feature extraction to obtain the real-time feature vector; The intelligent diagnostic matching module is used to combine the real-time feature vector with the real-time operating conditions, call the operating condition adaptive classification model corresponding to the real-time operating conditions, match the real-time feature vector with the benchmark acoustic signature feature library, and output the acoustic signature abnormal features and fault identification results of the gearbox of the wind turbine generator set to be tested.

10. A computer readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the acoustic fingerprint detection method for the gearbox of a wind turbine generator set as described in any one of claims 1 to 8.

Images