A fan blade defect acoustic print recognition and crack diagnosis method and system based on deep learning

Abstract

The application relates to the technical field of fan blades, and provides a fan blade defect acoustic print recognition and crack diagnosis method and system based on deep learning, which comprises the following steps: S1, collecting original acoustic print data generated by a fan blade during operation in real time through a non-contact acoustic print collecting unit; S2, preprocessing the original acoustic print data; S3, inputting the digital acoustic print signal into a blade defect acoustic print recognition model based on deep learning which is constructed in advance, extracting acoustic print features, and judging whether an abnormality exists; S4, when it is judged that an abnormality exists, inputting abnormal acoustic print data into a blade crack diagnosis model based on normal abnormal acoustic print data fusion modeling to perform crack fault recognition, and generating a crack diagnosis result; and S5, visualizing the crack diagnosis result through a fan blade acoustic print monitoring and early warning platform and sending an alarm information. The application significantly improves the accuracy, reliability and intelligent level of fan blade defect monitoring.

Description

Technical Field

[0001] This invention relates to the field of wind turbine blade technology, and more specifically, to a method and system for wind turbine blade defect acoustic signature recognition and crack diagnosis based on deep learning. Background Technology

[0002] As the core component for capturing wind energy, wind turbine blades operate under harsh conditions such as wind and sandstorms, lightning strikes, icing, and alternating loads, making them highly susceptible to defects such as cracks, pinholes, and lightning damage. If these blade defects are not detected and addressed promptly, they can lead to blade breakage or even turbine collapse, causing significant safety accidents and economic losses. Therefore, real-time and accurate monitoring of wind turbine blade condition has significant engineering importance and economic benefits.

[0003] Currently, wind turbine blade condition monitoring mainly relies on two methods: manual inspection and vibration monitoring. Manual inspection typically involves observation through binoculars or close-up inspection by climbing the tower. This method suffers from problems such as long inspection cycles, low efficiency, strong subjectivity, and significant safety hazards, making it difficult to achieve real-time continuous monitoring of blade condition. Vibration monitoring methods involve placing accelerometers inside the blades or nacelles to collect vibration signals during blade operation and analyzing vibration characteristics to determine blade condition. However, vibration monitoring is limited by sensor installation locations, lacks sensitivity to early micro-cracks, and the sensors themselves are prone to failure due to long-term exposure to alternating loads, resulting in high maintenance costs.

[0004] In recent years, voiceprint monitoring technology has gradually gained attention as a non-contact method for monitoring equipment status. Existing voiceprint monitoring solutions typically use a single microphone to collect the operating sound of the fan and then use time-domain, frequency-domain analysis, or traditional machine learning methods for fault identification. However, these solutions suffer from the following technical problems in practical applications: First, the operating environment of wind turbine blades is complex, with multiple strong interference sources such as wind noise, gearbox noise, generator noise, and tower resonance. The signal-to-noise ratio of the acoustic fingerprint signal collected by a single microphone is low, making it difficult to effectively extract the weak acoustic fingerprint features of the blade itself, resulting in low monitoring accuracy and high false alarm rate.

[0005] Second, the existing acoustic signature acquisition devices have a fixed directionality and cannot be adaptively configured according to differences in wind turbine model, blade length, installation height, and site environment. The acquired acoustic signature signals contain a large number of interference components that are not in the target direction, which affects the subsequent analysis results.

[0006] Third, existing voiceprint recognition algorithms mostly use a single model to extract features and classify faults from voiceprint signals, which is highly dependent on positive and negative sample data. In actual engineering, blade crack samples are scarce, making it difficult for the model to learn effective crack features, resulting in poor generalization ability and a tendency to miss or false alarms.

[0007] Fourth, existing technologies lack the ability to make detailed diagnoses of abnormal sound patterns. They can only determine whether there is an abnormality, but cannot further identify whether it is a crack fault, let alone quantify the severity of the crack, making it difficult to provide maintenance personnel with accurate maintenance decision-making basis.

[0008] Fifth, existing voiceprint monitoring systems generally lack visualization and sample management functions, and the correlation between voiceprint data and equipment status is not intuitive, which is not conducive to maintenance personnel quickly understanding and mastering the equipment operating status.

[0009] Therefore, there is an urgent need in this field for a method and system for wind turbine blade defect acoustic signature recognition and crack diagnosis that can overcome the above-mentioned defects. Through multi-microphone array noise reduction, directional adaptive configuration, deep learning algorithm based on positive abnormal sample fusion modeling, and a visual monitoring and early warning platform, high-quality acquisition, accurate recognition, and intelligent crack diagnosis of wind turbine blade acoustic signatures can be achieved, effectively improving the accuracy and reliability of wind turbine blade defect monitoring. Summary of the Invention

[0010] The present invention provides a method and system for identifying acoustic signatures and diagnosing cracks in wind turbine blades based on deep learning, which can overcome some or all of the shortcomings of the prior art.

[0011] According to the present invention, a method for wind turbine blade defect acoustic signature recognition and crack diagnosis based on deep learning includes the following steps: Step S1: Collect raw acoustic data generated by the wind turbine blades in real time by using a non-contact acoustic data acquisition unit deployed at a preset position on or around the outer wall of the wind turbine tower. Step S2: Preprocess the raw voiceprint data, including noise reduction based on a multi-microphone array, directional configuration, and analog-to-digital conversion to generate a digital voiceprint signal; Step S3: Input the digital voiceprint signal into the pre-built deep learning-based blade defect voiceprint recognition model, extract voiceprint features, and determine whether there is an anomaly; Step S4: When an anomaly is detected, the abnormal acoustic signature data is input into the blade crack diagnosis model based on the fusion model of positive and abnormal acoustic signature data to identify crack faults and generate crack diagnosis results. Step S5: Visualize the crack diagnosis results through the wind turbine blade acoustic signature monitoring and early warning platform and issue an alarm message.

[0012] Preferably, in step S1, the non-contact voiceprint acquisition unit is a digital intelligent sensor that does not require a dedicated acquisition device. Its frequency range is 20Hz to 96kHz, the signal-to-noise ratio is greater than 65dB, and it is installed using bolts or a single independent rod.

[0013] Preferably, in step S2, the noise reduction process is represented as follows: ; In the formula, Number of microphone arrays The discrete signal acquired by the i-th microphone. These are the corresponding noise reduction weight coefficients. It is a digital voiceprint signal.

[0014] Preferably, in step S2, the noise reduction processing based on the multi-microphone array includes: using a circular array composed of no less than 8 microphones to collect sound, and improving the signal-to-noise ratio through a multi-microphone array noise reduction algorithm; The directional configuration includes configuring the directional of the acquisition unit to any one of omnidirectional, 90°, or 120°, depending on the needs of the on-site scenario.

[0015] Preferably, in step S3, an anomaly detection function is used to determine whether an anomaly exists: ; In the formula, For deep feature extraction networks, To reconstruct the network, Given the signal length, if Then it is judged as abnormal. This is a preset abnormal threshold.

[0016] Preferably, in step S3, the method for constructing the deep learning-based blade defect acoustic signature recognition model and the blade crack diagnosis model includes: Collect positive sample acoustic fingerprint data of wind turbine blades under normal operating conditions and negative sample acoustic fingerprint data under abnormal operating conditions; The positive and negative sample voiceprint data are labeled to form a device voiceprint sample library; based on the device voiceprint sample library, a deep neural network is used for training, wherein the loss function during the training process is... Represented as: ; In the formula, The reconstruction loss is used to minimize the reconstruction error of positive sample speakerprints. The classification loss is used to measure the classification accuracy of negative samples. This is the balance coefficient.

[0017] Preferably, in step S4, the diagnostic model outputs the crack probability. : ; In the formula, These are deep features of abnormal voiceprints extracted by a deep network. This is the weight matrix. For bias terms, For the Sigmoid activation function, when It was determined at that time that a crack existed. The threshold for crack detection.

[0018] This invention provides a deep learning-based acoustic signature recognition and crack diagnosis system for wind turbine blade defects. It employs the aforementioned deep learning-based acoustic signature recognition and crack diagnosis method for wind turbine blade defects and includes: The non-contact acoustic fingerprint acquisition unit is deployed on the outer wall of the wind turbine tower or at a preset position around the tower to collect raw acoustic fingerprint data of the wind turbine blades in real time during operation. The data processing terminal includes a switch and an application server, which are used to receive the raw voiceprint data and perform data processing, feature extraction, model recognition and diagnostic analysis steps. The monitoring and early warning platform is connected to the data computing terminal and is used to realize real-time monitoring of wind turbine blade acoustic data, abnormal alarms, display of crack diagnosis results, and annotation and management of sound data through visualization technology, spatial information flow display technology, and historical flow display technology.

[0019] Preferably, the non-contact voiceprint acquisition unit includes: A circular array consisting of at least eight digital microphones; The processor is a high-performance processor with four or more cores and a clock speed of ≥1.5GHz; The memory includes ≥64GB of ROM and ≥1GB of RAM; The communication module supports Ethernet, Bluetooth, or WiFi communication and is used for data transmission with the data computing terminal.

[0020] Preferably, the monitoring and early warning platform includes: The data dashboard module is used to display wind turbine models, statistical information, alarm information, and sound information. The statistical information includes the total number of wind turbines and the total number of alarmed wind turbines. The alarm center module is used to record the alarm records of the wind turbines and supports filtering by wind turbine name, time, and alarm type. The equipment details page module is used to display the health status of the wind turbine, as well as the model judgment results for each audio track, and provides viewing functions for voiceprint playback, spectrogram, amplitude graph, and frequency domain analysis spectrum; The data analysis module is used to filter target equipment from the wind turbine list and display its real-time data, microphone measurement data, model results and waveforms, and spectrum analysis results.

[0021] The beneficial effects of this invention are as follows: This invention employs a non-contact acoustic signature acquisition unit deployed at a predetermined location on or around the outer wall of the wind turbine tower. This acquisition unit is a digital intelligent sensor with high-performance specifications, including a frequency range of 20Hz to 96kHz and a signal-to-noise ratio (SNR) > 65dB. Building upon this, step S2 introduces noise reduction processing based on a multi-microphone array. The signals from a circular array consisting of at least eight microphones are weighted and fused, and combined with configurable directivity (omnidirectional, 90°, or 120°), interference from wind noise, mechanical noise, and other factors is effectively suppressed, significantly improving the SNR of the acquired acoustic signature signal. Compared to existing single-microphone solutions, this invention can extract high-resolution blade acoustic signature features in high-noise environments, laying a reliable data foundation for subsequent diagnostics.

[0022] In step S3 of this invention, a deep learning-based blade defect acoustic signature recognition model is employed, which accurately identifies abnormal acoustic signatures through an anomaly discrimination function. During the model training phase, a loss function is constructed to fuse positive and abnormal acoustic signature data. The reconstruction loss utilizes a large number of normal samples to learn the inherent patterns of acoustic signatures, while the classification loss utilizes a small number of abnormal samples to enhance discrimination ability. This fusion modeling approach effectively solves the problem of poor model generalization ability under conditions of scarce crack samples, ensuring a high recognition rate for normal states while improving the detection capabilities for anomalies and cracks.

[0023] In step S4 of this invention, when an anomaly is detected, the abnormal acoustic signature is input into the blade crack diagnosis model based on positive anomaly fusion modeling, the crack probability is output, and a qualitative judgment of the crack is achieved through a preset threshold. This probability output can not only determine whether a crack exists, but also reflect the confidence level of the crack, providing maintenance personnel with a quantitative diagnostic basis, facilitating the formulation of graded maintenance strategies, and avoiding blind maintenance or missed detection.

[0024] This invention includes a monitoring and early warning platform that integrates a data dashboard, alarm center, equipment details page, and data analysis module. Through visualization, spatial information flow, and historical flow display technologies, it comprehensively showcases wind turbine models, acoustic signature data, anomaly alarms, and diagnostic results. Users can monitor acoustic signatures in real time, listen to abnormal sounds, view spectrograms and frequency domain analysis graphs, and annotate and manage acoustic signature data samples. This platform organically combines data acquisition, model analysis, result presentation, and operation and maintenance management, solving the problems of weak correlation between data and equipment status and fragmented operations in existing technologies, thus improving operation and maintenance efficiency and user experience.

[0025] The non-contact voiceprint acquisition unit in this invention adopts a domestically designed architecture. Its processor is quad-core or higher with a clock speed ≥1.5GHz, ROM ≥64GB, and RAM ≥1GB. It supports Ethernet, Bluetooth, or WiFi communication, has PoE power supply capability, and is adaptable to various environments such as high-altitude cold, dry, and humid conditions. The data processing server is equipped with a high-performance CPU and large-capacity storage, capable of supporting real-time processing and long-term storage of voiceprint data from more than 72 wind turbines. The entire system meets industrial-grade protection standards (IP66, IP40), has a wide operating temperature range (-25℃~+70℃), and achieves secure data transmission through a fiber optic ring network, ensuring long-term stable operation of the system in the harsh environment of wind farms.

[0026] This invention supports remote configuration commands through a communication module, enabling remote adjustment of the sampling rate (48kHz / 192kHz selectable), bit depth (16bit / 32bit selectable), directivity, and gain parameters of the acquisition unit without on-site operation, significantly reducing maintenance costs and improving the maintainability and adaptability of the system. Attached Figure Description

[0027] Figure 1 This is a flowchart of a deep learning-based method for identifying acoustic signatures and diagnosing cracks in wind turbine blades, as described in this embodiment. Detailed Implementation

[0028] To further understand the content of this invention, a detailed description of the invention will be provided in conjunction with the accompanying drawings and embodiments. It should be understood that the embodiments are merely illustrative and not limiting of the invention. Example

[0029] like Figure 1 As shown in the figure, this embodiment provides a method for sound signature recognition and crack diagnosis of wind turbine blade defects based on deep learning, which includes the following steps: Step S1: Collect raw acoustic data generated by the wind turbine blades in real time by using a non-contact acoustic data acquisition unit deployed at a preset position on or around the outer wall of the wind turbine tower. Step S2: Preprocess the raw voiceprint data, including noise reduction based on a multi-microphone array, directional configuration, and analog-to-digital conversion to generate a digital voiceprint signal; Step S3: Input the digital voiceprint signal into the pre-built deep learning-based blade defect voiceprint recognition model, extract voiceprint features, and determine whether there is an anomaly; Step S4: When an anomaly is detected, the abnormal acoustic signature data is input into the blade crack diagnosis model based on the fusion model of positive and abnormal acoustic signature data to identify crack faults and generate crack diagnosis results. Step S5: Visualize the crack diagnosis results through the wind turbine blade acoustic signature monitoring and early warning platform and issue an alarm message.

[0030] In step S1, the non-contact voiceprint acquisition unit is a digital intelligent sensor that does not require a dedicated acquisition device. Its frequency range is 20Hz to 96kHz, the signal-to-noise ratio is greater than 65dB, and it is installed using bolts or a single independent rod.

[0031] In step S2, the noise reduction process is represented as follows: ; In the formula, Number of microphone arrays The discrete signal acquired by the i-th microphone. These are the corresponding noise reduction weight coefficients. It is a digital voiceprint signal.

[0032] In step S2, the noise reduction processing based on the multi-microphone array includes: using a circular array of no less than 8 microphones to collect sound, and improving the signal-to-noise ratio through a multi-microphone array noise reduction algorithm; The directional configuration includes configuring the directional of the acquisition unit to any one of omnidirectional, 90°, or 120°, depending on the needs of the on-site scenario.

[0033] In step S3, the presence of an anomaly is determined using an anomaly detection function: ; In the formula, For deep feature extraction networks, To reconstruct the network, Given the signal length, if Then it is judged as abnormal. This is a preset abnormal threshold.

[0034] In step S3, the method for constructing the deep learning-based blade defect acoustic signature recognition model and the blade crack diagnosis model includes: Collect positive sample acoustic fingerprint data of wind turbine blades under normal operating conditions and negative sample acoustic fingerprint data under abnormal operating conditions; The positive and negative sample voiceprint data are labeled to form a device voiceprint sample library; based on the device voiceprint sample library, a deep neural network is used for training, wherein the loss function during the training process is... Represented as: ; In the formula, The reconstruction loss is used to minimize the reconstruction error of positive sample speakerprints. The classification loss is used to measure the classification accuracy of negative samples. This is the balance coefficient.

[0035] In step S4, the diagnostic model outputs the crack probability. : ; In the formula, These are deep features of abnormal voiceprints extracted by a deep network. This is the weight matrix. For bias terms, For the Sigmoid activation function, when It was determined at that time that a crack existed. The threshold for crack detection.

[0036] This embodiment provides a deep learning-based acoustic signature recognition and crack diagnosis system for wind turbine blade defects. It employs the aforementioned deep learning-based acoustic signature recognition and crack diagnosis method for wind turbine blade defects and includes: The non-contact acoustic fingerprint acquisition unit is deployed on the outer wall of the wind turbine tower or at a preset position around the tower to collect raw acoustic fingerprint data of the wind turbine blades in real time during operation. The data processing terminal includes a switch and an application server, which are used to receive the raw voiceprint data and perform data processing, feature extraction, model recognition and diagnostic analysis steps. The monitoring and early warning platform is connected to the data computing terminal and is used to realize real-time monitoring of wind turbine blade acoustic data, abnormal alarms, display of crack diagnosis results, and annotation and management of sound data through visualization technology, spatial information flow display technology, and historical flow display technology.

[0037] The non-contact voiceprint acquisition unit includes: A circular array consisting of at least eight digital microphones; The processor is a high-performance processor with four or more cores and a clock speed of ≥1.5GHz; The memory includes ≥64GB of ROM and ≥1GB of RAM; The communication module supports Ethernet, Bluetooth, or WiFi communication and is used for data transmission with the data computing terminal.

[0038] The monitoring and early warning platform includes: The data dashboard module is used to display wind turbine models, statistical information, alarm information, and sound information. The statistical information includes the total number of wind turbines and the total number of alarmed wind turbines. The alarm center module is used to record the alarm records of the wind turbines and supports filtering by wind turbine name, time, and alarm type. The equipment details page module is used to display the health status of the wind turbine, as well as the model judgment results for each audio track, and provides viewing functions for voiceprint playback, spectrogram, amplitude graph, and frequency domain analysis spectrum; The data analysis module is used to filter target equipment from the wind turbine list and display its real-time data, microphone measurement data, model results and waveforms, and spectrum analysis results.

[0039] The data computing server adopts an x86 architecture, with a CPU of 64 cores or more, a main frequency of ≥2.2GHz, memory of ≥128GB, and hard disk of ≥15TB, used to store the collected voiceprint data and perform deep learning model operations. The non-contact voiceprint acquisition unit is connected to the data computing terminal via a POE optoelectronic switch. The switch is powered by POE, supports daisy-chain deployment, and has an IP40 or higher protection rating. Its operating temperature range is -40℃ to +85℃.

[0040] This embodiment systematically solves key problems in existing wind turbine blade acoustic signature monitoring technologies, such as low signal-to-noise ratio, scarce crack samples, insufficient diagnostic accuracy, and complex operation and maintenance, by organically combining multi-microphone array noise reduction and directivity adaptation, a deep learning model based on positive anomaly fusion, quantified crack diagnosis output, and an integrated monitoring and early warning platform. It significantly improves the accuracy, reliability, and intelligence level of wind turbine blade defect monitoring, and has important engineering application value and broad market prospects.

[0041] The present invention and its embodiments have been described above illustratively. This description is not restrictive, and the figures shown are only one embodiment of the present invention; the actual structure is not limited thereto. Therefore, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the present invention, such designs should fall within the protection scope of the present invention.

Claims

1. A method for acoustic signature recognition and crack diagnosis of wind turbine blade defects based on deep learning, characterized in that: Includes the following steps: Step S1: Collect raw acoustic data generated by the wind turbine blades in real time by using a non-contact acoustic data acquisition unit deployed at a preset position on or around the outer wall of the wind turbine tower. Step S2: Preprocess the raw voiceprint data, including noise reduction based on a multi-microphone array, directional configuration, and analog-to-digital conversion to generate a digital voiceprint signal; Step S3: Input the digital voiceprint signal into the pre-built deep learning-based blade defect voiceprint recognition model, extract voiceprint features, and determine whether there is an anomaly; Step S4: When an anomaly is detected, the abnormal acoustic signature data is input into the blade crack diagnosis model based on the fusion model of positive and abnormal acoustic signature data to identify crack faults and generate crack diagnosis results. Step S5: Visualize the crack diagnosis results through the wind turbine blade acoustic signature monitoring and early warning platform and issue an alarm message.

2. The method for soundprint recognition and crack diagnosis of wind turbine blade defects based on deep learning according to claim 1, characterized in that: In step S1, the non-contact voiceprint acquisition unit is a digital intelligent sensor that does not require a dedicated acquisition device. Its frequency range is 20Hz to 96kHz, the signal-to-noise ratio is greater than 65dB, and it is installed using bolts or a single independent rod.

3. The method for soundprint recognition and crack diagnosis of wind turbine blade defects based on deep learning according to claim 2, characterized in that: In step S2, the noise reduction process is represented as follows: ； In the formula, Number of microphone arrays The discrete signal acquired by the i-th microphone. These are the corresponding noise reduction weight coefficients. It is a digital voiceprint signal.

4. The method for soundprint recognition and crack diagnosis of wind turbine blade defects based on deep learning according to claim 3, characterized in that: In step S2, the noise reduction processing based on the multi-microphone array includes: using a circular array of no less than 8 microphones to collect sound, and improving the signal-to-noise ratio through a multi-microphone array noise reduction algorithm; The directional configuration includes configuring the directional of the acquisition unit to any one of omnidirectional, 90°, or 120°, depending on the needs of the on-site scenario.

5. The method for soundprint recognition and crack diagnosis of wind turbine blade defects based on deep learning according to claim 4, characterized in that: In step S3, the presence of an anomaly is determined using an anomaly detection function: ； In the formula, For deep feature extraction networks, To reconstruct the network, Given the signal length, if Then it is judged as abnormal. This is a preset abnormal threshold.

6. The method for soundprint recognition and crack diagnosis of wind turbine blade defects based on deep learning according to claim 5, characterized in that: In step S3, the method for constructing the deep learning-based blade defect acoustic signature recognition model and the blade crack diagnosis model includes: Collect positive sample acoustic fingerprint data of wind turbine blades under normal operating conditions and negative sample acoustic fingerprint data under abnormal operating conditions; The positive and negative sample voiceprint data are labeled to form a device voiceprint sample library; based on the device voiceprint sample library, a deep neural network is trained, wherein the loss function during the training process is... Represented as: ； In the formula, The reconstruction loss is used to minimize the reconstruction error of positive sample voiceprints. The classification loss is used to measure the classification accuracy of negative samples. This is the balance coefficient.

7. The method for soundprint recognition and crack diagnosis of wind turbine blade defects based on deep learning according to claim 6, characterized in that: In step S4, the diagnostic model outputs the crack probability: : ； In the formula, These are deep features of abnormal voiceprints extracted by a deep network. This is the weight matrix. For bias terms, For the Sigmoid activation function, when It was determined at that time that a crack existed. The threshold for crack detection.

8. A deep learning-based acoustic signature recognition and crack diagnosis system for wind turbine blade defects, characterized in that: It employs a deep learning-based acoustic signature recognition and crack diagnosis method for wind turbine blade defects as described in any one of claims 1-7, and includes: The non-contact acoustic fingerprint acquisition unit is deployed on the outer wall of the wind turbine tower or at a preset position around the tower to collect raw acoustic fingerprint data of the wind turbine blades in real time during operation. The data processing terminal includes a switch and an application server, which are used to receive the raw voiceprint data and perform data processing, feature extraction, model recognition and diagnostic analysis steps. The monitoring and early warning platform is connected to the data computing terminal and is used to realize real-time monitoring of wind turbine blade acoustic data, abnormal alarms, display of crack diagnosis results, and annotation and management of sound data through visualization technology, spatial information flow display technology, and historical flow display technology.

9. The deep learning-based acoustic signature recognition and crack diagnosis system for wind turbine blade defects according to claim 8, characterized in that: The non-contact voiceprint acquisition unit includes: A circular array consisting of at least eight digital microphones; The processor is a high-performance processor with four or more cores and a clock speed of ≥1.5GHz; The memory includes ≥64GB of ROM and ≥1GB of RAM; The communication module supports Ethernet, Bluetooth, or WiFi communication and is used for data transmission with the data computing terminal.

10. The deep learning-based acoustic signature recognition and crack diagnosis system for wind turbine blade defects according to claim 9, characterized in that: The monitoring and early warning platform includes: The data dashboard module is used to display wind turbine models, statistical information, alarm information, and sound information. The statistical information includes the total number of wind turbines and the total number of alarmed wind turbines. The alarm center module is used to record the alarm records of the wind turbines and supports filtering by wind turbine name, time, and alarm type. The equipment details page module is used to display the health status of the wind turbine, as well as the model judgment results for each audio track, and provides viewing functions for voiceprint playback, spectrogram, amplitude graph, and frequency domain analysis graph; The data analysis module is used to filter target equipment from the wind turbine list and display its real-time data, microphone measurement data, model results and waveforms, and spectrum analysis results.

Images