A method and system for detecting fan faults

Abstract

The present invention discloses a method and system for detecting fan faults, including: obtaining a multi-channel sound signal sequence during the operation of the fan, splicing the multi-channel sound signal sequence into a single-channel mixed sequence, and constructing a data source based on the single-channel mixed sequence and the multi-channel sound signal sequence; segmenting each sequence in the data source, performing time-frequency transformation on each obtained subsequence, and constructing a subgraph based on a new subsequence of each sequence; fusing the subgraphs corresponding to all sequences, and using a trained graph network model to perform fault detection on the obtained fused subgraph. The sound signal is converted into a graph signal for fault detection, and the information of the multi-channel sound signal is integrated to improve the accuracy, while achieving end-to-end detection.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular to a method and system for detecting fan faults. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] As the core component of the wind power generation system, the normal operation of the wind turbine is crucial to the stability of energy production. However, due to factors such as the wind farm environment, long-term operation and manufacturing defects, wind turbine blades may face various faults, such as wear, cracks, yaw, etc. Wind turbine blade failure will not only reduce power generation efficiency, but may also cause equipment damage and increase maintenance costs. Therefore, early detection and diagnosis of blade failures are crucial to ensure the reliability and long-term operation of wind power generation systems.

[0004] Traditional wind turbine monitoring methods mainly rely on manual detection, pattern recognition and other methods. Although these methods can detect some common faults, their ability to diagnose minor defects and early faults in blades is relatively limited. At the same time, there are problems such as high cost and inability to identify new fault types.

[0005] Among many monitoring methods, sound signals carry a wealth of information. By analyzing the sound generated during the operation of the fan, the abnormal state of the blades can be more accurately identified, which can reflect the characteristics of the health status of the equipment. However, the current sound-based fan blade fault detection technology mainly relies on some traditional signal processing technologies. Since traditional signal processing technologies mainly rely on mathematical processing, they perform well in single, ideal scenarios. However, in actual application scenarios with more complex environments, compared with the detection methods combined with deep learning, traditional signal processing technologies have problems such as low detection accuracy and insufficient versatility in scenarios. Summary of the invention

[0006] In order to solve the above problems, the present invention proposes a fan fault detection method and system, which converts sound signals into graphical models for fault detection, integrates the information of multi-channel sound signals, improves the accuracy, and realizes end-to-end detection.

[0007] In order to achieve the above object, the present invention adopts the following technical solution:

[0008] In a first aspect, the present invention provides a fan fault detection method, comprising:

[0009] Acquire a multi-channel sound signal sequence during the operation of the fan, splice the multi-channel sound signal sequence into a single-channel mixed sequence, and construct a data source based on the single-channel mixed sequence and the multi-channel sound signal sequence;

[0010] Each sequence in the data source is segmented, each subsequence is transformed in time and frequency, and a subgraph is constructed based on the new subsequence of each sequence;

[0011] The subgraphs corresponding to all sequences are fused, and the trained graph network model is used to perform fault detection on the obtained fused subgraphs.

[0012] As an optional implementation, a beamforming method is used to splice a multi-channel sound signal sequence into a single-channel mixed sequence. After the single-channel mixed sequence and the multi-channel sound signal sequence are paralleled, a data source including N+1 one-dimensional data sequences is obtained, where N is the number of channels of the multi-channel sound signal sequence.

[0013] As an optional implementation, each sequence is divided into equal parts, each sequence is converted into m subsequences, and the multiple subsequences are normalized.

[0014] As an alternative implementation, with a new subsequence The value at is taken as the eigenvalue, the point represented by the eigenvalue is taken as the node of the subgraph, and l is the length of the new subsequence.

[0015] As an optional implementation, determine the points represented by the feature values ​​involved in each sequence, measure the similarity of the two points with the Euclidean distance between the two points, and use the K nearest neighbor method to add edges between the k nearest points of each point, with the edges being the edges of the subgraph, and the weight of each edge being the Euclidean distance between the two points.

[0016] As an optional implementation method, after concatenating and merging the subgraphs corresponding to all sequences, the feature matrix composed of the eigenvalues ​​of the points in the fused subgraphs and the connection matrix composed of the presence or absence of edges and the weights of the edges between different points are used as the input of the graph network model to obtain a detection result of whether there is a fault or not.

[0017] As an optional implementation method, by integrating remote monitoring of the health status of the wind turbine on the server side, the transmission and display of wind turbine images, videos and sound signals can be realized, the changing trend analysis of different blade signals can be realized, and the feature extraction of different blade signals and the display of the surrounding environment status can be realized.

[0018] In a second aspect, the present invention provides a fan fault detection system, comprising:

[0019] A preprocessing module is configured to obtain a multi-channel sound signal sequence during the operation of the fan, splice the multi-channel sound signal sequence into a single-channel mixed sequence, and construct a data source based on the single-channel mixed sequence and the multi-channel sound signal sequence;

[0020] A subgraph construction module is configured to segment each sequence in the data source, perform time-frequency transformation on each obtained subsequence, and then construct a subgraph based on a new subsequence of each sequence;

[0021] The fault detection module is configured to fuse the subgraphs corresponding to all the sequences, and perform fault detection on the obtained fused subgraphs using the trained graph network model.

[0022] In a third aspect, the present invention provides an electronic device comprising a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method described in the first aspect is performed.

[0023] In a fourth aspect, the present invention provides a computer-readable storage medium for storing computer instructions, wherein when the computer instructions are executed by a processor, the method described in the first aspect is performed.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] The present invention proposes a fan fault detection method and system, which adopts a microphone array composed of multiple microphones to collect multi-channel sound signals during the operation of the fan, splices the multi-channel sound signal sequence into a single-channel mixed sequence, and then forms a final data source with the multi-channel sound signal sequence, captures sound sources from multiple directions and dimensions, focuses sensitivity on a specific direction, and retains the information of the original data as much as possible, thereby improving detection sensitivity and accuracy, and reducing detection costs compared to traditional manual detection methods.

[0026] The present invention proposes a fan fault detection method and system, which divides each sequence in the data source. Since the signal strengths obtained from different channels may be different, the subsequences are normalized to effectively eliminate the data strength differences collected from different data sources. Then, in order to express the sequence characteristics, the subsequences are transformed in time and frequency, and the values ​​at the middle positions of the new subsequences are selected as characteristic values. The points where the characteristic values ​​are located are used as nodes in the subgraph, and the edges constructed by points with closer distances are used as the edges of the subgraph. Thus, a subgraph of each sequence is obtained, so that the sound signal is converted into a graph model for fault detection, and the detection from multi-channel sound signals to sound signal diagnosis results is realized. The information of multi-channel sound signals is integrated to improve the accuracy, and end-to-end detection is realized at the same time, which saves detection steps and is easy to operate.

[0027] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0029] Figure 1 This is a flow chart of the fan fault detection method provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0030] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0031] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0032] It should be noted that the terms used herein are only for describing specific embodiments, and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that the terms "include" and "comprise" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0033] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0034] Example 1

[0035] This embodiment provides a wind turbine fault detection method, which quickly identifies various types of blade faults by real-time monitoring and analysis of sound signals during wind turbine operation, thereby improving the reliability and maintenance efficiency of the wind power generation system, while realizing remote fault monitoring of the wind turbine and improving the convenience of monitoring.

[0036] like Figure 1 As shown, specifically including:

[0037] Acquire a multi-channel sound signal sequence during the operation of the fan, splice the multi-channel sound signal sequence into a single-channel mixed sequence, and construct a data source based on the single-channel mixed sequence and the multi-channel sound signal sequence;

[0038] Each sequence in the data source is segmented, each subsequence is Fourier transformed, and a subgraph is constructed based on the new subsequence of each sequence;

[0039] The subgraphs corresponding to all sequences are fused, and the trained graph network model is used to perform fault detection on the obtained fused subgraphs.

[0040] Because the sites where wind turbines are arranged are often accompanied by strong winds and prone to noise, this embodiment uses a microphone array to collect multi-channel sound signals during the operation of the wind turbine. The microphone array is a collection of multiple microphones, which provides a more advanced means for sound capture than a traditional single microphone. Its main advantage lies in its powerful spatial capture capability. Through the collaborative work of multiple microphones, it can capture sound sources from multiple directions and dimensions, thereby providing a more accurate and efficient solution for sound source positioning, background noise suppression and sound field reconstruction; in addition, the microphone array uses its complex signal processing capabilities to separate and enhance sounds from different directions, providing a powerful boost for the analysis and processing of blade rotation sounds in complex noise environments; at the same time, for the fault detection algorithm used in this embodiment below, the microphone array is more suitable than conventional microphones, and the fusion of multi-channel data information is more accurate. The acquired data is then uploaded to the cloud through wireless transmission and finally integrated onto the platform.

[0041] In this embodiment, a beamforming method is used to splice the multi-channel sound signal sequence collected by the microphone array into a single-channel mixed sequence. After the single-channel mixed sequence and the multi-channel sound signal sequence are arranged in parallel, N+1 one-dimensional data sequences are obtained, where N is the number of channels of the multi-channel sound signal sequence.

[0042] The core idea of ​​the beamforming algorithm is to weight and phase-adjust the signal of each sensor to form a synthetic beam in a specific direction, thereby enhancing the signal in the target direction and weakening interference in other directions: Among them, x n (t) is the output of the nth channel of the microphone array, t is time, N is the number of microphones, ω n is the weight of the corresponding microphone, τ n is the time delay corresponding to the microphone.

[0043] By adjusting the weights and time delays, beamforming can be achieved to focus the sensitivity of the system in a specific direction, thereby obtaining a fused single-channel mixed sequence, which is used as a data source together with the initial N-channel data sequence.

[0044] Because the single-channel mixed sequence and the initial N-channel sequence are data collected at the same time length, and the length of the data sequence is also the same, the N+1 one-dimensional data sequences are combined into one data source, and each sequence is divided into m equal parts, and each sequence is converted into m subsequences;

[0045]

[0046] In this embodiment, N is 6.

[0047] Since the signal strength obtained from different channels may be different, the data from different data sources must be normalized. For the data from the same data source, the difference from the traditional normalization is that the normalization in this embodiment is based on the multiple subsequences obtained after segmentation. Through the normalization process, the difference in the data strength collected from different data sources is effectively eliminated, paving the way for splicing and composing a graph network.

[0048]

[0049] Among them, x i represents the i-th segment of a sequence, x min is the minimum value of the sequence, x max is the maximum value of the sequence.

[0050] After normalization, in order to better reflect the characteristics of a data sequence, this embodiment performs Fourier transform on the data sequence to convert the data from time domain data to frequency domain data. Different from the traditional Fourier transform, the Fourier transform of this embodiment is based on the Fourier transform of the segmented subsequences. For each subsequence, the new subsequence obtained after Fourier transform is:

[0051] If the length of the new subsequence is l, then select The value at is used as the eigenvalue of the new subsequence, and the eigenvalue is used as the point in the graph to represent the new subsequence. A subgraph is constructed based on the points represented by the eigenvalues ​​of the new subsequences of each sequence.

[0052] Specifically, signal sequences containing more related information tend to share fault information, and when two signal sequences contain more related information, the feature points representing the two sequences are often similar. Therefore, this embodiment uses the Euclidean distance between two points as a standard for measuring the similarity of the two points. The closer the Euclidean distance, the more similar the two points are.

[0053] Then the Euclidean distance between points a and b is:

[0054]

[0055] Among them, x a ,y a are the horizontal and vertical coordinates of point a on the new subsequence, x b ,y b are the horizontal and vertical coordinates of point b on the new subsequence, and point a and point b are both The point where the value is located.

[0056] Then, the K nearest neighbor method is used to construct edges for points with close distances, and the weight of each edge is the distance between the two points; finally, the points represented by the eigenvalues ​​of the new subsequence of each sequence are used as the nodes of the subgraph, and the constructed edges are used as the edges of the subgraph, so as to obtain the subgraph corresponding to each sequence.

[0057] K nearest neighbors is a common mathematical method for classifying points. The specific process includes: when a point is selected for classification, the distance between it and all other points is traversed and calculated, and k nearest points are selected. These points can be considered to be of the same category. In this embodiment, after finding the k nearest points, edges are added between these points. At the same time, in this method, different values ​​of k will have a huge impact on the results. After repeated experiments, the best effect is achieved when the k value is 3.

[0058] Similarly, other sequences also construct subgraphs in the same manner. In this embodiment, seven data sequences are taken as an example, corresponding to seven subgraphs.

[0059] When performing fault diagnosis on the collected sound signal, the data of the seven channels need to be fused for diagnosis. Correspondingly, after constructing seven sub-graphs, the seven sub-graphs are fused; because the difference in signal strength of different channels has been considered and normalized in the previous steps, the seven sub-graphs obtained can be directly spliced ​​and fused.

[0060] After obtaining the fused sub-graph, the trained graph network model is used to perform fault detection on the signal sequence converted into a graph. The graph network model used in this embodiment is GNN, which is a graph neural network model commonly used for classification. The input of this model is the feature matrix of the points of the graph and the connection matrix of the graph, and the output is the category of the graph. In this embodiment, the feature matrix of the point is the eigenvalue of the point, and the connection matrix of the graph is a matrix constructed by whether there are edges and the weights of the edges between different points. For the category of the graph, it is divided into two categories: with or without faults. After training, the model can classify the graph converted from the sound signal.

[0061] This embodiment implements a detection model from multi-channel sound signals to diagnostic results of sound signals, integrates the information of multi-channel sound signals, improves accuracy, and realizes end-to-end detection, saving detection steps and being more convenient to operate.

[0062] Since wind turbines are often placed in outdoor areas with strong winds and the wind turbines themselves are relatively large, close-range monitoring is inconvenient. Therefore, it is very important to realize remote fault detection of wind turbines. This embodiment uses Raspberry Pi to obtain long-distance wind turbine data and builds a service platform on the server side to implement remote monitoring of the health status of the wind turbine. The construction of the integrated system side is mainly based on Web technology, and is committed to realizing a series of functions to comprehensively monitor and analyze the status of the wind turbine. Specifically including:

[0063] (1) Wind turbine image and video transmission and presentation function: By combining Web technology with Raspberry Pi, the system can realize real-time transmission and display of wind turbine images and videos, allowing users to remotely monitor the wind turbine operating status and obtain image and video information in a timely manner, thereby gaining a more comprehensive understanding of the actual working conditions.

[0064] (2) Real-time presentation of sound signals: Through the web interface, technicians can view the sound signals of the fan in real time, helping to detect potential faults or abnormal conditions and improving the accuracy of fault diagnosis.

[0065] (3) Trend analysis function for different blade signals: A trend analysis function is provided for the sensor data of different blades, allowing users to view the changing trends of blade signals, which helps to promptly discover blade performance problems and provides important reference for maintenance.

[0066] (4) Feature extraction function for different blade signals: By providing the feature extraction function for different blade signals and displaying them on the system interface, it helps to gain a deeper understanding of the performance differences of each blade and provide data support for optimizing wind turbine operation.

[0067] (5) Display of blade surrounding environment conditions: The web interface displays data on the blade surrounding environment conditions, including temperature, wind speed, humidity, air pressure and other parameters. This helps to comprehensively consider the impact of environmental factors on wind turbine operation and provide comprehensive information for a reasonable evaluation of wind turbine performance.

[0068] (6) Remote status monitoring: Through Web technology, users can remotely monitor the status of the fan at any time and anywhere, improving the practicality and convenience of the system and enabling users to understand the operating status of the fan in real time at different locations.

[0069] (7) Real-time data upload and interface presentation: Realize real-time upload of wind turbine data to ensure that users can obtain the latest operating information; at the same time, through the intuitive Web interface, clearly present data in the form of charts, images, etc., so that users can intuitively understand the performance indicators and operating status of the wind turbine.

[0070] In summary, the construction of an integrated system based on Web technology not only provides the functions of remote monitoring and real-time data display, but also helps users to have a deeper understanding of all aspects of wind turbine operation through various analysis tools, providing strong support for operation, maintenance and management.

[0071] Example 2

[0072] This embodiment provides a fan fault detection system, including:

[0073] A preprocessing module is configured to obtain a multi-channel sound signal sequence during the operation of the fan, splice the multi-channel sound signal sequence into a single-channel mixed sequence, and construct a data source based on the single-channel mixed sequence and the multi-channel sound signal sequence;

[0074] A subgraph construction module is configured to segment each sequence in the data source, perform time-frequency transformation on each obtained subsequence, and then construct a subgraph based on a new subsequence of each sequence;

[0075] The fault detection module is configured to fuse the subgraphs corresponding to all the sequences, and perform fault detection on the obtained fused subgraphs using the trained graph network model.

[0076] It should be noted that the above modules correspond to the steps described in Example 1, and the examples and application scenarios implemented by the above modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above Example 1. It should be noted that the above modules, as part of the system, can be executed in a computer system such as a set of computer executable instructions.

[0077] In further embodiments, there is also provided:

[0078] An electronic device includes a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method described in Embodiment 1 is performed. For the sake of brevity, it will not be described in detail here.

[0079] It should be understood that in this embodiment, the processor may be a central processing unit CPU, and the processor may also be other general-purpose processors, digital signal processors DSP, application-specific integrated circuits ASIC, off-the-shelf programmable gate arrays FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0080] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0081] A computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by a processor, the method described in Example 1 is completed.

[0082] The method in Example 1 can be directly embodied as a hardware processor, or a combination of hardware and software modules in the processor. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware. To avoid repetition, it is not described in detail here.

[0083] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0084] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A fan fault detection method, characterized in that: include: Acquire a multi-channel sound signal sequence during the operation of the fan, splice the multi-channel sound signal sequence into a single-channel mixed sequence, and construct a data source based on the single-channel mixed sequence and the multi-channel sound signal sequence; Each sequence in the data source is segmented, each subsequence is transformed in time and frequency, and a subgraph is constructed based on the new subsequence of each sequence; with new subsequence The value at is taken as the eigenvalue, and the point represented by the eigenvalue is taken as the node of the subgraph. l is the length of the new subsequence; Determine the points represented by the eigenvalues ​​involved in each sequence, measure the similarity of the two points by the Euclidean distance between the two points, and use the K nearest neighbor method to add edges between the k nearest points of each point. The edges are the edges of the subgraph, and the weight of each edge is the Euclidean distance between the two points. The subgraphs corresponding to all sequences are fused, and the trained graph network model is used to perform fault detection on the obtained fused subgraphs. The model used is the graph neural network, GNN.

2. A fan fault detection method according to claim 1, characterized in that: Using a beamforming method, a multi-channel sound signal sequence is spliced ​​into a single-channel mixed sequence. After the single-channel mixed sequence and the multi-channel sound signal sequence are paralleled, a data source including N+1 one-dimensional data sequences is obtained, where N is the number of channels of the multi-channel sound signal sequence.

3. A fan fault detection method according to claim 1, characterized in that: Each sequence is divided equally, each sequence is converted into m subsequences, and the multiple subsequences are normalized.

4. A fan fault detection method according to claim 1, characterized in that: After the subgraphs corresponding to all sequences are spliced ​​and fused, the feature matrix composed of the eigenvalues ​​of the points in the fused subgraphs and the connection matrix composed of the presence or absence of edges and the weights of the edges between different points are used as the input of the graph network model to obtain the detection result of whether there is a fault or not.

5. A fan fault detection method according to claim 1, characterized in that: By integrating remote monitoring of the health status of the wind turbine on the server side, the transmission and display of wind turbine images, videos and sound signals can be realized, the changing trend analysis of different blade signals can be realized, and the feature extraction of different blade signals and the display of the surrounding environment status can be realized.

6. A fan fault detection system, characterized in that: include: A preprocessing module is configured to obtain a multi-channel sound signal sequence during the operation of the fan, splice the multi-channel sound signal sequence into a single-channel mixed sequence, and construct a data source based on the single-channel mixed sequence and the multi-channel sound signal sequence; A subgraph construction module is configured to segment each sequence in the data source, perform time-frequency transformation on each obtained subsequence, and then construct a subgraph based on a new subsequence of each sequence; with new subsequence The value at is taken as the eigenvalue, and the point represented by the eigenvalue is taken as the node of the subgraph. l is the length of the new subsequence; Determine the points represented by the eigenvalues ​​involved in each sequence, measure the similarity of the two points by the Euclidean distance between the two points, and use the K nearest neighbor method to add edges between the k nearest points of each point. The edges are the edges of the subgraph, and the weight of each edge is the Euclidean distance between the two points. The fault detection module is configured to fuse the subgraphs corresponding to all sequences, and use the trained graph network model to perform fault detection on the obtained fused subgraphs. The model used is a graph neural network, GNN.

7. An electronic device, characterized in that: The method comprises a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method according to any one of claims 1 to 5 is completed.

8. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the method described in any one of claims 1 to 5.

Images