Fan blade grading analysis fault monitoring system and monitoring method

Abstract

The invention belongs to the technical field of application of computer science, signal processing, mode recognition, artificial intelligence and the like, and particularly discloses a fan blade grading analysis fault monitoring system and method, and the system comprises a data preprocessing module, an abnormity coarse screening module, a spectrogram generation module, a fault fine classification module and an alarm sending module. The method has the beneficial effects that a grading thought is introduced, a simple and rapid fast-dtw algorithm is used on a fan side to perform preliminary screening on audio signals, whether abnormity exists or not is judged, and if abnormity is found through preliminary screening, further detailed classification is performed, so that the specific fault type of the blade is accurately identified; according to the method, the fault detection accuracy and efficiency are improved, unnecessary calculation burden is reduced through grading processing, optimal configuration of resources is achieved, and an innovative solution is provided for equipment maintenance and fault early warning in the wind power generation industry by combining fault detection of the fan blade with the modern deep learning technology.

Description

Technical Field

[0001] The present invention belongs to the fields of computer science, signal processing, pattern recognition, artificial intelligence and other applied technologies, and specifically relates to a wind turbine blade graded analysis fault monitoring system and monitoring method, which is applied to the monitoring of wind power generation equipment. The system uses voiceprint analysis technology to perform abnormality identification and fault classification on wind turbine blades. Background Art

[0002] With the rapid development of renewable energy, wind power generation has received more and more attention as a clean energy.

[0003] As a key component of a wind turbine generator set, the operating status of wind turbine blades directly affects the efficiency and safety of the entire power generation system. Therefore, real-time monitoring and fault diagnosis of wind turbine blades are particularly important. Traditional wind turbine blade fault detection methods usually rely on regular maintenance and manual inspections, which are not only costly but also difficult to achieve real-time monitoring and early fault warning.

[0004] In the existing technology, audio signal analysis is an effective means of diagnosing wind turbine blade faults. By collecting the audio signals generated by the wind turbine blades during operation, their frequency and time domain characteristics can be analyzed to determine whether the blades have cracks, lightning strikes, fractures and other faults. However, these methods often require complex signal processing and feature extraction techniques, which require large amounts of calculation and are prone to false alarms or missed alarms when there are no obvious fault features.

[0005] Based on the above problems, the present invention provides a fan blade graded analysis fault monitoring system and monitoring method. Summary of the invention

[0006] Purpose of the invention: The purpose of the present invention is to provide a wind turbine blade graded analysis fault monitoring system and monitoring method, aiming to improve the accuracy and real-time performance of fault detection through the wind turbine blade graded analysis fault monitoring method, optimize the use of computing resources, and provide an efficient and reliable fault monitoring solution for the wind power industry, so as to improve the operating safety and maintenance efficiency of wind turbines.

[0007] Technical solution: The first aspect of the present invention provides a wind turbine blade graded analysis fault monitoring system, including a data preprocessing module, an abnormality coarse screening module, a spectrogram generation module, a fault fine classification module and an alarm sending module; the data preprocessing module performs denoising preprocessing on the audio signal of the wind turbine blade; the abnormality coarse screening module uses the edge computing power on the wind turbine side to apply the fast-dtw algorithm to the denoised audio signals in pairs, so as to quickly confirm whether there are abnormal blades; the spectrogram generation module converts the preprocessed audio signal into a Mel spectrogram; the fault fine classification module uses the core computing power on the station side to perform detailed classification of fault types on the spectrogram through a deep convolutional neural network based on Vision Transformer; the alarm sending module pushes detailed alarm information of wind turbine blade faults to the alarm center.

[0008] The second aspect of the present invention provides a monitoring method for a fan blade graded analysis fault monitoring system, comprising the following steps: S1, data preprocessing by a data preprocessing module, denoising preprocessing of the audio signal of the fan blade; S2, abnormal coarse screening by an abnormal coarse screening module, using the edge computing power on the fan side, applying the fast-dtw algorithm to the denoised audio signals in pairs, so as to quickly confirm whether there are abnormal blades; S3, spectrogram generation by a spectrogram generation module, converting the preprocessed audio signal into a Mel spectrogram; S4, fault fine classification by a fault fine classification module, using the core computing power on the station side, and performing detailed classification of the fault type on the spectrogram through a deep convolutional neural network based on Vision Transformer; S5, alarm sending by an alarm sending module, pushing detailed alarm information of the fan blade fault to the alarm center.

[0009] As a preferred embodiment, the S1 data preprocessing step further includes:

[0010] a. Receive the original audio signal;

[0011] b. Use a bandpass filter to remove background noise components in the signal;

[0012] c. Use window function to process the signal to reduce spectrum leakage;

[0013] d. Normalize the denoised signal.

[0014] As a preferred embodiment, the S2 abnormality rough screening step further includes:

[0015] a. Use the fast-dtw algorithm to calculate the similarity between each pair of audio signals. The calculation formula is:

[0016]

[0017] Among them, S a and S b are two audio signals, d represents the distance metric, and π represents the optimal alignment path.

[0018] Definition of minimum value in the cost matrix;

[0019] b. Determine whether there are abnormal leaves based on the calculated similarity score. If it exceeds the preset threshold, it is abnormal.

[0020] As a preferred embodiment, the S3 spectrogram generation step includes:

[0021] a. Perform Fourier transform on the denoised audio signal;

[0022] b. Applying a Mel filter bank to convert the frequency domain signal into a Mel frequency representation;

[0023] c. Calculate the power spectral density of the Mel filter bank and generate a Mel spectrogram.

[0024] As a preferred implementation, S4 uses a deep learning model based on Vision Transformer (ViT) to classify the Mel-spectrogram features of the wind turbine blades.

[0025] The ViT model includes a multi-head self-attention mechanism that processes multiple attention heads in parallel, with each head independently computing attention scores and outputs.

[0026] As a preferred embodiment, the ViT model includes the following sub-steps:

[0027] a. Divide the Mel spectrogram into P×P patches;

[0028] b. Flatten each patch and map it to a feature vector of dimension D through a linear layer.

[0029] As a preferred implementation, the multi-head attention mechanism includes:

[0030] a. Split the input feature tensor into multiple heads, each head processing a different part of the input data;

[0031] b. Each header calculates the set of query, key, and value, expressed as:

[0032]

[0033] Among them, Q, K, V are the corresponding query, key and value matrices, d k is the dimension of the key vector;

[0034] c. Combine the output of each head into a unified output, usually by concatenation and then passing it through a linear layer;

[0035] d. The combined output is used in subsequent Transformer encoder layers or classification tasks.

[0036] As a preferred embodiment, the input feature tensor of the ViT model has dimensions (N, S, D), where N is the batch size, S is the sequence length equal to P×P, and D is the feature dimension of each sequence element.

[0037] As a preferred embodiment, the output of the ViT model is a feature tensor transformed by the final linear layer, and its dimension is (N, num_classes); wherein num_classes is the number of fault categories.

[0038] The third aspect of the present invention provides a computer-readable storage medium, on which is stored a monitoring method program for a fan blade graded analysis fault monitoring system. When the monitoring method for the fan blade graded analysis fault monitoring system is executed by a processor, the steps of the monitoring method for the fan blade graded analysis fault monitoring system are implemented.

[0039] The fourth aspect of the present invention provides an electronic device, which includes a memory, a processor, and the wind blade graded analysis fault monitoring system stored in the memory and executable on the processor, and when the processor executes the wind blade graded analysis fault monitoring system, the steps of the monitoring method of the wind blade graded analysis fault monitoring system are implemented.

[0040] Compared with the prior art, the fan blade classification analysis fault monitoring system and monitoring method of the present invention have the following beneficial effects:

[0041] The present invention introduces a hierarchical idea, that is, firstly, a simple and fast fast-dtw algorithm is used on the wind turbine side to perform a preliminary screening of the audio signal to determine whether there is an abnormality. As a time series similarity measurement method, the fast-dtw algorithm can effectively identify abnormal patterns in the blade audio signal, and has high computational efficiency, which is suitable for real-time processing by edge computing devices; if an abnormality is found in the preliminary screening, the system will use a deep learning model based on Vision Transformer (ViT) to further classify the audio signal in detail. As an advanced image processing model, the ViT model processes multiple attention heads in parallel through a multi-head self-attention mechanism, which can capture the complex features and long-distance dependencies in the audio signal, thereby accurately identifying the specific fault type of the blade.

[0042] The method of the present invention not only improves the accuracy and efficiency of fault detection, but also reduces unnecessary computing burden through hierarchical processing, realizes optimal allocation of resources, and provides an innovative solution for equipment maintenance and fault warning in the wind power industry by combining wind turbine blade fault detection with modern deep learning technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0044] Figure 1 It is a system structure block diagram of the present invention;

[0045] Figure 2 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0046] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0047] In the description of the present invention, it should be noted that the orientations or positional relationships indicated by the terms "top", "bottom", "one side", "the other side", "front", "back", "middle part", "inside", "top end", "bottom end", etc. are based on the orientations or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention; the terms "first", "second", and "third" are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance; in addition, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be an indirect connection through an intermediate medium, or it can be a connection between the two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0048] like Figure 1A wind turbine blade graded analysis fault monitoring system shown in the figure includes a data preprocessing module, an abnormality coarse screening module, a spectrogram generation module, a fault fine classification module and an alarm sending module; the data preprocessing module performs denoising preprocessing on the audio signal of the wind turbine blade; the abnormality coarse screening module uses the edge computing power on the wind turbine side to apply the fast-dtw algorithm to the denoised audio signals in pairs, so as to quickly confirm whether there are abnormal blades; the spectrogram generation module converts the preprocessed audio signal into a Mel spectrogram; the fault fine classification module uses the core computing power on the station side to perform detailed classification of fault types on the spectrogram through a deep convolutional neural network based on Vision Transformer; the alarm sending module pushes detailed alarm information of wind turbine blade faults to the alarm center.

[0049] like Figure 2 As shown, the monitoring method of the present invention specifically comprises the following steps:

[0050] S1. Data preprocessing:

[0051] a. Use 32 high-sensitivity microphones evenly distributed around the fan to collect the audio signals generated by the fan blades when they are running. The sampling rate of each microphone is 48kHz to ensure that the high-frequency fault signals that may be generated by the blades can be captured.

[0052] b. Perform bandpass filtering on the audio signal collected by each microphone. The passband of the filter is set to 1kHz to 10kHz. This frequency band can cover the typical fault frequencies that may occur in wind turbine blades.

[0053] c. Divide the filtered signal into 25ms frames with an overlap time of 10ms between frames. Use the Hanning window function to perform windowing on each frame to reduce spectrum leakage.

[0054] d. Normalize the windowed signal so that the amplitude range of the signal is unified to [-1,1] for subsequent processing.

[0055] S2. Abnormal rough screening:

[0056] a. Use the Fast-DTW algorithm to perform pairwise comparisons on the audio signals after data preprocessing and calculate the similarity between each pair of audio signals. The window size of the Fast-DTW algorithm is set to 100 ms to adapt to the changes within the blade rotation cycle.

[0057] b. Construct a cost matrix and find the optimal alignment path through a dynamic programming algorithm to calculate the similarity score. The formula is:

[0058]

[0059] Among them, S a and S b are two audio signals, d represents the distance metric, and π represents the optimal alignment path, which is defined by the minimum value in the cost matrix;

[0060] c. Set a threshold, such as 0.8, and when the similarity score exceeds this threshold, it is considered that there is a significant difference between the signals, which may indicate that one of the leaves is abnormal.

[0061] S3. Spectrogram generation:

[0062] a. Perform a Fast Fourier Transform (FFT) on the audio signal confirmed to be abnormal, converting it from the time domain to the frequency domain, using a 1024-point FFT window.

[0063] b. Apply 40 Mel filters to process the frequency domain signal, and the bandwidth of each filter is set to 70 Hz to cover the audible frequency range of the human ear.

[0064] c. Calculate the power spectral density of each filter group and generate the final Mel spectrogram for subsequent feature extraction.

[0065] S4. Fault classification:

[0066] a. The Mel spectrogram is divided into 16x16 pixel patches, each patch is flattened and mapped to a feature vector of dimension 768 through a linear layer.

[0067] b. The model contains 12 Transformer encoder layers, each encoder layer includes 8 attention heads, and the dimension of each head is 768 / 8=96.

[0068] c. Further extract features through the Transformer encoder layer and output the probability distribution of fault types through the classification head. The output dimension of the classification head is (N, num_classes); where num_classes is the number of fault categories.

[0069] S5. Alarm sending:

[0070] a. Once the ViT model determines that a blade fault exists, the system will generate detailed fault warning information, including the fault type, possible causes and recommended maintenance measures.

[0071] b. Through the wireless communication module, the alarm information is sent to the central monitoring system or the maintenance personnel's mobile device to ensure the real-time and reliability of the message.

[0072] Example

[0073] The monitoring system of the present invention ( Figure 1 ) is used to implement the method described ( Figure 2 shown):

[0074] a. Data preprocessing module: performs denoising preprocessing on the audio signal of the fan blades; this module is used to implement the function of step 1 in Example 1, which will not be repeated here.

[0075] b. Abnormal rough screening module: using the edge computing power on the fan side, the fast-dtw algorithm is applied to the denoised audio signals in pairs to quickly confirm whether there are abnormal blades; this module is used to implement the function of step 2 in embodiment 1 and will not be repeated here.

[0076] c. Spectrogram generation module: converts the preprocessed audio signal into a Mel-type spectrogram; this module is used to implement the function of step 3 in Example 1, and will not be described in detail here.

[0077] d. Fault classification module: Utilize the core computing power of the station side to perform detailed classification of fault types on the spectrogram through a deep convolutional neural network based on Vision Transformer; this module is used to implement the function of step 4 in Example 1 and will not be repeated here.

[0078] e. Alarm sending module: pushes detailed alarm information of wind turbine blade failure to the alarm center. This module is used to implement the function of step 5 in the first embodiment, which will not be described in detail here.

[0079] Specifically, in order to achieve Figure 2 The monitoring method shown, an embodiment of the present invention provides a computer-readable storage medium, the computer-readable storage medium stores a monitoring method program of a fan blade graded analysis fault monitoring system, and when the monitoring method of the fan blade graded analysis fault monitoring system is executed by a processor, the steps of the monitoring method of the fan blade graded analysis fault monitoring system are implemented.

[0080] Specifically, for Figure 1 , Figure 2 The monitoring system and monitoring method shown, an embodiment of the present invention provides an electronic device, the electronic device includes a memory, a processor and the fan blade graded analysis fault monitoring system stored in the memory and executable on the processor, when the processor executes the fan blade graded analysis fault monitoring system, the steps of the monitoring method of the fan blade graded analysis fault monitoring system are implemented.

[0081] It should be noted that, in this article, terms such as "comprises", "includes" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of more restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0082] 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 the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. A fan blade classification analysis fault monitoring system, characterized by: It includes data preprocessing module, abnormality rough screening module, spectrogram generation module, fault classification module and alarm sending module; The data preprocessing module performs denoising preprocessing on the audio signal of the fan blades; The abnormal rough screening module uses the edge computing power on the fan side to apply the fast-dtw algorithm to the denoised audio signals in pairs, so as to quickly confirm whether there are abnormal blades; The spectrogram generation module converts the preprocessed audio signal into a mel spectrogram; The fault classification module uses the core computing power of the station side to perform detailed classification of fault types on the spectrogram through a deep convolutional neural network based on Vision Transformer; The alarm sending module pushes detailed alarm information of wind turbine blade failure to the alarm center.

2. A monitoring method for a fan blade graded analysis fault monitoring system according to claim 1, characterized in that: The following steps are included: S1, data preprocessing of the data preprocessing module, performing denoising preprocessing on the audio signal of the fan blade; S2, the abnormal rough screening module uses the edge computing power on the wind turbine side to apply the fast-dtw algorithm to the denoised audio signals in pairs, so as to quickly confirm whether there are abnormal blades; S3, spectrogram generation by a spectrogram generation module, converting the preprocessed audio signal into a mel spectrogram; S4, the fault classification module uses the core computing power of the station side to perform detailed classification of fault types on the spectrogram through a deep convolutional neural network based on VisionTransformer; S5. The alarm sending module sends an alarm and pushes the detailed alarm information of the fan blade failure to the alarm center.

3. The monitoring method of a fan blade classification analysis fault monitoring system according to claim 2 is characterized by: The data preprocessing in S1 includes the following steps: a. Receive the original audio signal; b. Use a bandpass filter to remove background noise components in the signal; c. Use window function to process the signal to reduce spectrum leakage; d. Normalize the denoised signal.

4. The monitoring method of a fan blade classification analysis fault monitoring system according to claim 2 is characterized by: The rough screening of abnormalities in S2 comprises the following steps: a. Use the fast-dtw algorithm to calculate the similarity between each pair of audio signals. The calculation formula is: Among them, S a and S b are two audio signals, d represents the distance metric, and π represents the optimal alignment path, which is defined by the minimum value in the cost matrix; b. Determine whether there are abnormal leaves based on the calculated similarity score. If it exceeds the preset threshold, it is abnormal.

5. The monitoring method of a fan blade classification analysis fault monitoring system according to claim 2 is characterized by: The generation of the spectrogram in S3 comprises the following steps: a. Perform Fourier transform on the denoised audio signal; b. Applying a Mel filter bank to convert the frequency domain signal into a Mel frequency representation; c. Calculate the power spectral density of the Mel filter bank and generate a Mel spectrogram.

6. The monitoring method of a fan blade classification analysis fault monitoring system according to claim 2 is characterized by: In the S4, a deep learning model based on Vision Transformer (ViT) is used to classify the Mel-spectrogram features of the fan blades; Among them, the deep learning model of Vision Transformer (ViT) includes a multi-head self-attention mechanism, which processes multiple attention heads in parallel, and each head independently calculates the attention score and output.

7. The monitoring method of a fan blade classification analysis fault monitoring system according to claim 6, characterized in that: The method of classifying the Mel-spectrogram features of the fan blades using a deep learning model based on Vision Transformer (ViT) includes the following sub-steps: a. Divide the Mel spectrogram into P×P patches; b. Flatten each patch and map it to a feature vector of dimension D through a linear layer.

8. The monitoring method of a fan blade classification analysis fault monitoring system according to claim 6, characterized in that: The multi-head self-attention mechanism includes: a. Split the input feature tensor into multiple heads, each head processing a different part of the input data; b. Each header calculates the set of query, key, and value, expressed as: where Q, K, V are the corresponding query, key, and value matrices, d k is the dimension of the key vector; c. Combine the output of each head into a unified output, usually by concatenation and then passing it through a linear layer; d. The combined output is used in subsequent Transformer encoder layers or classification tasks.

9. The monitoring method of a fan blade classification analysis fault monitoring system according to claim 8, characterized in that: The input feature tensor of the multi-head self-attention mechanism of the Vision Transformer (ViT) deep learning model has dimensions (N, S, D); Where N is the batch size, S is the sequence length equal to P×P, and D is the feature dimension of each sequence element.

10. The monitoring method of a fan blade classification analysis fault monitoring system according to claim 8, characterized in that: The output of the multi-head self-attention mechanism of the Vision Transformer (ViT) deep learning model is a feature tensor transformed by the final linear layer, with a dimension of (N, num_classes); Where num_classes is the number of fault classes.

Images