Fan fault detection method and system based on AI auscultation, and fan safety system

Abstract

The present application relates to a fan fault detection method and system based on AI auscultation, and a fan safety system, which collects a third sound signal during the operation of the fan blades, performs digital-to-analog conversion and noise reduction processing on the third sound signal to obtain a sound signal C, extracts the actual features of the sound signal C, and compares them with the standard features to calculate the first similarity; when the first similarity is less than or equal to the first preset value, the actual features are compared with any one of the fault feature sets, and the second similarity is calculated respectively, the second similarity is arranged in descending order, and the first M fault features corresponding to the second similarity are taken to obtain the corresponding fault type or a combination of M fault types; after detecting a fault, an alarm is sounded and fault diagnosis advice and maintenance plans are provided. The above scheme can use AI auscultation combined with deep learning technology to extract and compare acoustic features of the real-time collected working noise to detect whether the blades are faulty in real time.

Description

Technical Field

[0001] The present application relates to the technical field of fan fault detection, and in particular to a fan fault detection method and system based on AI auscultation, and a fan safety system. Background Art

[0002] A wind turbine is an electric device that converts wind energy into mechanical work, which drives the rotor to rotate and finally outputs alternating current. A wind turbine generally consists of a wind wheel, a generator, a rectifier, a tower, a speed limit safety mechanism, and an energy storage device.

[0003] As energy crisis and environmental problems become increasingly serious, countries around the world are competing to develop renewable emerging energy. my country's long-term plan clearly supports "research and development of large wind power generation equipment", and wind power equipment has developed rapidly. While wind power is developing rapidly, the high operation and maintenance costs of wind turbines have affected the economic benefits of wind farms. Wind farms are generally located in remote areas with harsh environments, and the cabins are located at an altitude of 70m to 90m, which makes it difficult to maintain and repair the units, increasing the operation and maintenance costs of the units. In addition, the high operation and maintenance costs increase the operating costs of wind farms and reduce the economic benefits of wind power.

[0004] Wind turbine blades are one of the core components of wind turbines and are key equipment for energy conversion. Their status has a direct impact on the performance and power generation quality of the entire machine. Due to long-term operation under variable loads and harsh environments, they frequently fail. Therefore, blade status monitoring plays a very important role in ensuring the efficiency of wind turbines and improving the safety and reliability of the units.

[0005] The traditional means of detecting blade damage and faults is to use telescopes for observation and manual inspection by rappelling with ropes, which has the disadvantages of low detection efficiency, high labor intensity for workers, high detection cost, low safety, long detection time, and large power generation loss due to downtime. An existing detection scheme is to use drones to inspect the surface defects of wind turbine blades from a high altitude. After the original image of the wind turbine blade surface taken by the drone is transmitted back, it is manually identified whether the surface of the wind turbine blade has defects. However, this method still requires manual identification of defects and cannot produce real-time detection results. At the same time, this method can only periodically detect whether the blade has faults. Before the detection, the blade may have been operating for a long time due to faults, and it is impossible to detect whether the blade has faults in real time. Summary of the invention

[0006] Based on this, it is necessary to provide a fan fault detection method and system based on AI auscultation, and a fan safety system, in view of the current situation that the existing wind turbine blade detection is still at a low level of manual operation, with low efficiency, high detection cost, and inability to detect in real time. AI auscultation can be combined with deep learning technology to extract and compare the acoustic features of the working noise collected in real time, so as to detect in real time whether the blades are faulty, prevent problems before they happen, and achieve the best detection effect at the lowest cost, thereby providing highly cost-effective technical support for the safe operation of the wind turbine blade structure, greatly reducing the production risk of the wind farm, improving the production efficiency of the wind farm, and solving the problems that the existing wind turbine blade detection is still at a low level of manual operation, with low efficiency, high detection cost, and inability to detect in real time.

[0007] The fan fault detection method based on AI auscultation includes:

[0008] The first sound signal of the fan blade during the fault-free operation is collected, and the first sound signal is converted into digital-to-analog and subjected to noise reduction processing to obtain a sound signal A, and the characteristics of the sound signal A are extracted to obtain a standard characteristic; the second sound signals of the fan blade during the operation under N different fault types are collected respectively, and the N second sound signals are converted into digital-to-analog and subjected to noise reduction processing to obtain N sound signals B, and the fault characteristics of the N sound signals B are extracted respectively to obtain a fault characteristic set, wherein N is greater than or equal to 50;

[0009] Collecting a third sound signal during the operation of the fan blade, performing digital-to-analog conversion and noise reduction processing on the third sound signal to obtain a sound signal C, extracting actual features of the sound signal C, and comparing the actual features with the standard features to calculate a first similarity;

[0010] When the first similarity is less than or equal to a first preset value, the actual feature is compared with any one of the fault feature set, and the second similarities are calculated respectively to obtain N second similarities, and the second similarities are arranged in descending order, and the first M fault features corresponding to the second similarities are taken to obtain the corresponding fault type or a combination of M fault types, where M is less than or equal to 5;

[0011] When a fault is detected, an alarm will be issued and fault diagnosis advice and maintenance plans will be provided.

[0012] Preferably, in the above-mentioned fan fault detection method based on AI auscultation, the characteristics of the sound signal A, the fault characteristics of the sound signal B, and the actual characteristics of the sound signal C all include a time domain waveform diagram, a spectrum diagram, a voiceprint diagram and a frequency band energy ratio.

[0013] Preferably, in the above-mentioned fan fault detection method based on AI auscultation, the calculation of the first similarity and the calculation of the second similarity are respectively analyzed and compared using three methods: structural similarity, cosine distance, and Euclidean distance to obtain three similarity results, and the largest of the three similarities is taken as the final similarity.

[0014] Preferably, in the above-mentioned fan fault detection method based on AI auscultation, after the steps of detecting a fault, giving an alarm and providing fault diagnosis advice and maintenance solutions, the method further comprises:

[0015] Manually confirm the fault type on site and mark it on the actual characteristics of the sound signal C;

[0016] The fault feature set is stored.

[0017] Preferably, in the above-mentioned fan fault detection method based on AI auscultation, after the steps of detecting a fault, giving an alarm and providing fault diagnosis advice and maintenance solutions, the method further comprises:

[0018] Generate operational reports periodically.

[0019] Preferably, in the above-mentioned fan fault detection method based on AI auscultation, the digital-to-analog conversion and noise reduction processing of the first sound signal, the digital-to-analog conversion and noise reduction processing of the N second sound signals, and the digital-to-analog conversion and noise reduction processing of the third sound signal all include a filtering processing step through a Butterworth filter, the order of the Butterworth filter is 30 to 50, the lower cutoff frequency is 50 Hz to 200 Hz, the upper cutoff frequency is 10 kHz to 15 kHz, and its square amplitude-frequency response function is:

[0020]

[0021] in, N is the order of the filter, Ωc is the 3dB cutoff frequency, Ωu is the upper cutoff frequency, and Ωl is the lower cutoff frequency.

[0022] A fan fault detection system based on AI auscultation includes a signal acquisition module, a signal processing module, a feature extraction module, a similarity calculation module, a judgment module, a fault determination module and an alarm module, wherein the signal acquisition module is used to collect a first sound signal during the fault-free operation of the fan blade, to collect a second sound signal during the operation of the fan blade under N different fault types, and to collect a third sound signal during the operation of the fan blade, and the signal processing module is used to perform digital-to-analog conversion and noise reduction processing on the first sound signal, to perform digital-to-analog conversion and noise reduction processing on N second sound signals, and to perform digital-to-analog conversion and noise reduction processing on the third sound signal. Processing, the feature extraction module is used to extract the features of the sound signal A, to extract the fault features of N sound signals B respectively, and to extract the actual features of the sound signal C, the similarity calculation module is used for the first similarity and for calculating the second similarity, the judgment module is used to judge whether the first similarity is less than or equal to a first preset value, the fault determination module is used to arrange the second similarity in descending order, take the first M fault features corresponding to the second similarity, and obtain the corresponding fault type or a combination of M fault types, and the alarm module is used to alarm and provide fault diagnosis advice and maintenance plans after detecting a fault.

[0023] Preferably, in the above-mentioned fan fault detection system based on AI auscultation, the feature extraction module includes a time domain waveform graph extraction unit, a spectrum graph extraction unit, a voiceprint graph extraction unit and a frequency band energy ratio extraction unit, and the similarity calculation module includes a structural similarity calculation unit, a cosine distance calculation unit and a Euclidean distance calculation unit.

[0024] Preferably, the above-mentioned fan fault detection system based on AI auscultation also includes a report generation module, and the report generation module is used to regularly generate operation reports.

[0025] The wind turbine safety system includes an infrared online temperature measurement module, a behavioral environment analysis module, an equipment remote inspection module and the above-mentioned AI stethoscope-based wind turbine fault detection system. The infrared online temperature measurement module is used for equipment and components that need to be measured for temperature during each inspection at this stage; the behavioral environment analysis module is used to support the analysis of people wandering, gathering and intrusion around the wind turbine tower base to prevent theft and sabotage by outsiders; the equipment remote inspection module is used to combine video capabilities, audio capabilities and temperature measurement capabilities to carry out manual and intelligent inspections of electrical equipment in the station as planned.

[0026] The technical solution adopted in this application can achieve the following beneficial effects:

[0027] In the fan fault detection method and system based on AI auscultation and the fan safety system disclosed in the embodiments of the present application, a third sound signal is collected during the operation of the fan blade, and the actual feature of the third sound signal is first compared with the standard feature, and a first similarity is calculated to determine whether the fan blade is faulty at this time. In the event of a fault, it is necessary to further detect the type of fault that occurs in the fan blade, so the actual feature is compared with any one of the fault feature sets, and the similarity between the actual feature and each fault feature in the fault feature set is calculated to obtain N second similarities, and the second similarities are arranged in descending order, and the first M fault features corresponding to the second similarities are taken to obtain the corresponding faults. type or a combination of M fault types. After detecting a fault, an alarm is issued and fault diagnosis opinions and maintenance plans are provided, thereby completing the fault detection of the wind turbine blades. The third sound signal can be continuously collected and corresponding detection can be performed, so as to detect in real time whether the blades are faulty and prevent problems before they occur. No human participation is required in the fault detection process, and the best detection effect can be achieved at the lowest cost, thereby providing highly cost-effective technical support for the safe operation of the wind turbine blade structure, greatly reducing the production risk of the wind farm, improving the production efficiency of the wind farm, and solving the problems of existing wind turbine blade detection still being at a low level of manual operation, low efficiency, high detection cost, and inability to detect in real time.

[0028] At the same time, the actual features are first compared with the standard features to calculate the first similarity. Since the amount of data of the standard features is small, the processing difficulty of calculating the first similarity is relatively small at this time. Only when the first similarity is less than or equal to the first preset value, the actual features are compared with any one in the fault feature set, and the second similarity is calculated respectively. Since the amount of data in the fault feature set is large, the processing difficulty of calculating the second similarity is relatively large. Therefore, only when the first similarity is less than or equal to the first preset value, the actual features are compared with any one in the fault feature set, which can effectively reduce the difficulty of data processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 A schematic diagram of a fan fault detection method based on AI auscultation disclosed in an embodiment of the present application;

[0030] Figure 2 A schematic diagram of a fan fault detection system based on AI auscultation disclosed in an embodiment of the present application;

[0031] Figure 3 This is a schematic diagram of a wind turbine safety system disclosed in an embodiment of the present application. DETAILED DESCRIPTION

[0032] In order to facilitate the understanding of the present application, the present application will be described more fully below with reference to the relevant drawings. The preferred embodiments of the present application are given in the drawings. However, the present application can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present application more thoroughly and comprehensively understood.

[0033] It should be noted that when an element is referred to as being "disposed on" another element, it may be directly on the other element or there may be a centered element. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be a centered element at the same time. The terms "vertical", "horizontal", "left", "right", "top", "bottom", "bottom end", "top end" and similar expressions used herein are for illustrative purposes only and are not intended to be the only implementation method.

[0034] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application. The term "and / or" used herein includes any and all combinations of one or more of the related listed items.

[0035] Please refer to Figure 1 The present application embodiment discloses a fan fault detection method based on AI auscultation, comprising the following steps:

[0036] The first sound signal during the trouble-free operation of the fan blades can be collected by using a sound pickup sonar or an acoustic array sensor. The first sound signal is the sound signal during the normal operation of the fan blades without faults. Then, the first sound signal is converted into digital and analog and subjected to noise reduction processing to obtain a sound signal A. The features of the sound signal A are extracted to obtain standard features. Common sound features include time domain features, frequency domain features, and time-frequency domain features, which can be specifically subdivided into waveform, energy, soundprint, main frequency, short-time energy, zero-crossing rate, correlation, etc. The existing partial discharge data can be analyzed accordingly to find the features of the partial discharge sound. The obtained standard features correspond to the fact that the fan blades are trouble-free. That is to say, if the sound features generated during the operation of the fan blades are the same or basically similar to the standard features, it means that the fan blades are trouble-free at this time. If the sound features generated during the operation of the fan blades are different from the standard features, it means that the fan blades are faulty at this time.

[0037] The second sound signals of the fan blades during operation under N different fault types are collected respectively. The collected second sound signals are N, and the N second sound signals correspond to N different fault types one by one. The second sound signals of the fan blades during operation under N different fault types can be collected by using a sound pickup sonar or an acoustic array sensor. The second sound signal is the sound signal of the fan blades in operation when the fault occurs. Then, the N second sound signals are converted into digital analogs and subjected to noise reduction processing to obtain N sound signals B. The fault features of the N sound signals B are extracted respectively to obtain a fault feature set, where N is greater than or equal to 50. The obtained fault feature set corresponds to a fault of the fan blades. That is, if the sound features generated during the operation of the fan blades are the same or substantially similar to any one or more fault features in the fault feature set, it indicates that the fan blades are faulty at this time, and the fault type of the fan blades is one or more fault types corresponding to any one or more fault features. The fault feature set is obtained through a large amount of data neural network training, and integrates more fault types.

[0038] The third sound signal during the operation of the fan blades can be collected by using a sound pickup sonar or a sound array sensor. Specifically, the sound pickup sonar or the sound array sensor can be set on the bearing seat of the fan main shaft. The third sound signal is the sound signal of the fan blades during the actual operation on site. The third sound signal is then converted into an analog digital signal and subjected to noise reduction processing to obtain a sound signal C. The actual features of the sound signal C are extracted and compared with the standard features to calculate the first similarity, that is, the similarity between the actual features and the standard features is calculated.

[0039] When the first similarity is greater than the first preset value, that is, the actual feature is the same as or substantially similar to the standard feature, it indicates that the fan blade is not faulty at this time. When the first similarity is less than or equal to the first preset value, that is, the actual feature is different from the standard feature, it indicates that the fan blade is faulty at this time. At this time, only the fan blade is detected to be faulty, but the fault type of the fan blade cannot be obtained. Therefore, it is necessary to further detect the fault type of the fan blade, so the actual feature is compared with any one of the fault feature sets, and the second similarity is calculated respectively. The similarity between the actual feature and each fault feature in the fault feature set is calculated to obtain N second similarities, and the second similarities are arranged in descending order, that is, sorted from large to small, and the first M fault features corresponding to the second similarity are taken to obtain the corresponding fault type or a combination of M fault types, where M is less than or equal to 5, indicating that the fan blade has the above-mentioned fault at this time.

[0040] Alternatively, the fault features greater than the second preset value among the N second similarities are taken to obtain the corresponding fault type or a combination of M fault types, indicating that the above fault has occurred in the fan blade at this time. The first preset value and the second preset value can be 50%, 60% or 70%, and the application does not limit the size of the first preset value and the second preset value.

[0041] After a fault is detected, an alarm will be issued and fault diagnosis suggestions and maintenance plans will be provided for reference by maintenance personnel.

[0042] In the fan fault detection method based on AI auscultation disclosed in the embodiment of the present application, by collecting the third sound signal during the operation of the fan blade, the actual feature of the third sound signal is first compared with the standard feature, and the first similarity is calculated to determine whether the fan blade is faulty at this time. In the case of a fault, it is necessary to further detect the type of fault that occurs in the fan blade, so the actual feature is compared with any one of the fault feature sets, and the similarity between the actual feature and each fault feature in the fault feature set is calculated to obtain N second similarities, and the second similarities are arranged in descending order, and the first M fault features corresponding to the second similarities are taken to obtain the corresponding fault type or M It can detect a combination of fault types, and after detecting a fault, it will alarm and provide fault diagnosis opinions and maintenance plans, thereby completing the fault detection of wind turbine blades. It can continuously collect third sound signals and perform corresponding tests, so as to detect in real time whether the blades are faulty and prevent problems before they occur. No human participation is required in the fault detection process, and the best detection effect can be achieved at the lowest cost, thereby providing highly cost-effective technical support for the safe operation of the wind turbine blade structure, greatly reducing the production risk of the wind farm, improving the production efficiency of the wind farm, and solving the problems of existing wind turbine blade detection still being at a low level of manual operation, low efficiency, high detection cost, and inability to detect in real time.

[0043] At the same time, the actual features are first compared with the standard features to calculate the first similarity. Since the amount of data of the standard features is small, the processing difficulty of calculating the first similarity is relatively small at this time. Only when the first similarity is less than or equal to the first preset value, the actual features are compared with any one in the fault feature set, and the second similarity is calculated respectively. Since the amount of data in the fault feature set is large, the processing difficulty of calculating the second similarity is relatively large. Therefore, only when the first similarity is less than or equal to the first preset value, the actual features are compared with any one in the fault feature set, which can effectively reduce the difficulty of data processing.

[0044] As described above, the features of the sound signal A are extracted to obtain the standard features, the fault features of N sound signals B are extracted respectively to obtain the fault feature set, and the actual features of the sound signal C are extracted. Optionally, the features of the sound signal A, the fault features of the sound signal B, and the actual features of the sound signal C can all include a time domain waveform, a spectrum, a voiceprint, and a frequency band energy ratio. Through a large number of experiments, it is found that the features of the sound signals collected during the operation of the fan blades are basically reflected in the time domain waveform, the spectrum, the voiceprint, and the frequency band energy ratio. Therefore, by extracting the time domain waveform, the spectrum, the voiceprint, and the frequency band energy ratio from the sound signal A, the sound signal B, and the sound signal C, the features of the sound signals collected during the operation of the fan blades can be more comprehensively reflected, so that the differences between the features can be more accurately distinguished, which is convenient for calculating the similarity, and then it is convenient to obtain the fault type of the fan blades, thereby improving the accuracy and efficiency of the fan fault detection method.

[0045] We can analyze the existing partial discharge data to find the characteristics of partial discharge sound. For partial discharge sound data, we mainly analyze the time domain waveform, spectrum, soundprint, and frequency band energy ratio of the sound. From the spectrum diagram and soundprint diagram of the signal, we can see that there are obvious signals in the range of 10kHz to 30kHz, including both audible sound frequency bands and ultrasonic frequency bands, and the frequency band range is relatively wide. In general, there is little ultrasonic interference in the environment, so we can judge that the signal is an ultrasonic signal. To this end, we mainly conduct detailed feature analysis on the signals in the range of 10kHz to 30kHz.

[0046] Preferably, the first similarity and the second similarity are calculated respectively using structural similarity, cosine distance, and Euclidean distance for analysis and comparison to obtain three similarity results, and the largest of the three similarities is taken as the final similarity. That is to say, among the three results calculated by the three methods, the largest one is taken as the value of the first similarity or the second similarity. The calculation results of the above three similarities are highly accurate and can improve the accuracy of fault detection.

[0047] As a preferred embodiment, after the step of detecting a fault, giving an alarm and providing fault diagnosis advice and a maintenance plan, the fan fault detection method disclosed in the present application may further include:

[0048] Manually confirm the fault type on site and mark it on the actual characteristics of the sound signal C;

[0049] Store the fault feature set.

[0050] After detecting the fault, giving an alarm and providing fault diagnosis advice and maintenance plans, on-site staff need to repair the fan blades to determine whether the alarmed fault type matches the actual fault type on site. If they match, they are marked on the actual features of the sound signal C and then stored in the fault feature set. The next time a sound signal similar to the actual features of the sound signal C appears, the fault type can be directly determined. If they do not match, the actual fault type is also marked on the actual features of the sound signal C and then stored in the fault feature set. The next time a sound signal similar to the actual features of the sound signal C appears, the fault type can be directly determined. This enables the fan fault detection method to have AI learning capabilities, increase the data in the fault feature set, and improve the accuracy of the fault type alarmed by the fan fault detection method.

[0051] As mentioned above, the collected sound signal needs to be converted into digital-to-analog and subjected to noise reduction processing. Furthermore, the digital-to-analog conversion and noise reduction processing of the first sound signal, the digital-to-analog conversion and noise reduction processing of the N second sound signals, and the digital-to-analog conversion and noise reduction processing of the third sound signal all include a filtering processing step through a Butterworth filter. The order of the Butterworth filter is 30 to 50, the lower cutoff frequency is 50 Hz to 200 Hz, and the upper cutoff frequency is 10 kHz to 15 kHz. Its square amplitude-frequency response function is:

[0052]

[0053] in, N is the order of the filter, Ωc is the 3dB cutoff frequency, Ωu is the upper cutoff frequency, Ω l is the lower cut-off frequency.

[0054] The first sound signal, the second sound signal and the third sound signal are filtered to reduce interference signals in the first sound signal, the second sound signal and the third sound signal, which greatly improves the accuracy of fault type judgment and prevents interference signals in the first sound signal, the second sound signal and the third sound signal from affecting the detection results.

[0055] As a preferred step, after detecting a fault, an alarm is issued and a fault diagnosis suggestion and a maintenance plan are provided,

[0056] The fan fault detection method disclosed in the present application may also include:

[0057] Generate operation reports regularly to help operation and maintenance personnel understand the equipment operation status and provide reference for the research, development and improvement of wind turbine blades.

[0058] Please refer to Figure 2The embodiment of the present application also discloses a fan fault detection system based on AI auscultation, including a signal acquisition module, a signal processing module, a feature extraction module, a similarity calculation module, a judgment module, a fault determination module and an alarm module, wherein the signal acquisition module is used to collect a first sound signal during the fault-free operation of the fan blade, to collect a second sound signal during the operation of the fan blade under N different fault types, and to collect a third sound signal during the operation of the fan blade, and the signal processing module is used to perform digital-to-analog conversion and noise reduction processing on the first sound signal, to perform digital-to-analog conversion and noise reduction processing on the N second sound signals, and to perform digital-to-analog conversion and noise reduction processing on the first sound signal. Three sound signal digital-to-analog conversion and noise reduction processing, the feature extraction module is used to extract the features of the sound signal A, to extract the fault features of N sound signals B respectively, and to extract the actual features of the sound signal C, the similarity calculation module is used for the first similarity and for calculating the second similarity, the judgment module is used to judge whether the first similarity is less than or equal to the first preset value, the fault determination module is used to arrange the second similarity in descending order, take the first M fault features corresponding to the second similarity, and obtain the corresponding fault type or a combination of M fault types, and the alarm module is used to alarm and provide fault diagnosis advice and maintenance plans after detecting a fault.

[0059] In the wind turbine fault detection system based on AI auscultation disclosed in the embodiment of the present application, the signal acquisition module can continuously collect the third sound signal and perform corresponding detection, so as to detect in real time whether the blade is faulty and prevent it before it happens. No human participation is required in the fault detection process, and the best detection effect can be achieved at the lowest cost, thereby providing highly cost-effective technical support for the safe operation of the wind turbine blade structure, greatly reducing the production risk of the wind farm, improving the production efficiency of the wind farm, and solving the problems of existing wind turbine blade detection still being at a low level of manual operation, low efficiency, high detection cost, and inability to detect in real time.

[0060] At the same time, the actual feature is first compared with the standard feature to calculate the first similarity. Since the amount of data of the standard feature is small, the processing difficulty of calculating the first similarity is relatively small at this time. Only when the first similarity is less than or equal to the first preset value, the actual feature is compared with any one in the fault feature set, and the second similarity is calculated respectively. Since the amount of data in the fault feature set is large, the processing difficulty of calculating the second similarity is relatively large. Therefore, only when the first similarity is less than or equal to the first preset value, the actual feature is compared with any one in the fault feature set, which can effectively reduce the processing difficulty of the similarity calculation module.

[0061] Preferably, the feature extraction module may include a time domain waveform extraction unit, a spectrum extraction unit, a voiceprint extraction unit and a frequency band energy ratio extraction unit. Through a large number of experiments, it is found that the characteristics of the sound signals collected during the operation of the fan blades are basically reflected in the time domain waveform, spectrum, voiceprint and frequency band energy ratio. Therefore, by extracting the time domain waveform, spectrum, voiceprint and frequency band energy ratio in the sound signal A, the sound signal B and the sound signal C, the characteristics of the sound signals collected during the operation of the fan blades can be more comprehensively reflected, so that the differences between the features can be distinguished more accurately, which is convenient for calculating the similarity, and then it is convenient to obtain the fault type of the fan blade, thereby improving the accuracy and efficiency of the fan fault detection method.

[0062] The similarity calculation module may include a structural similarity calculation unit, a cosine distance calculation unit, and a Euclidean distance calculation unit. The above three similarity calculation units have high accuracy and can improve the accuracy of fault detection.

[0063] Preferably, the wind turbine fault detection system disclosed in the present application may also include a report generation module, which is used to regularly generate operation reports to facilitate operation and maintenance personnel to understand the equipment operation status and provide a reference for the research and development and improvement of wind turbine blades.

[0064] Please refer to Figure 3 The embodiment of the present application also discloses a wind turbine safety system, including an infrared online temperature measurement module, a behavioral environment analysis module, an equipment remote inspection module and the wind turbine fault detection system based on AI stethoscope as described in any of the above embodiments. The infrared online temperature measurement module is used for the equipment and components that need to be measured for temperature during each inspection at this stage; the behavioral environment analysis module is used to support the analysis of personnel wandering, gathering and intrusion around the wind turbine tower base to prevent theft and sabotage by outsiders; the equipment remote inspection module is used to combine video capabilities, audio capabilities and temperature measurement capabilities to perform manual and intelligent inspections on electrical equipment in the station as planned.

[0065] It should be noted that the equipment remote inspection module includes an online intelligent inspection unit, an online manual inspection unit and an offline manual inspection unit.

[0066] Among them, the online intelligent inspection unit targets specific scenarios (box transformers, frequency conversion cabinets, cabin stator / rotor power distribution cabinets and cabin cabinets, etc.), cooperates with intelligent algorithm front-ends, thermal imaging front-ends, auscultation front-ends and other intelligent devices to perform intelligent analysis on the data collected, generates alarm data in a timely manner, and finally manually confirms a set of online intelligent inspection processes.

[0067] At the same time, the online manual inspection unit can perform online manual inspections through the large screen of the centralized control center, the platform client on the computer PC, and the mobile phone by means of automatic capture, video playback, alarm event query, and inspection record query.

[0068] In addition, offline manual inspections can issue inspection tasks based on the inspection content covered by the wind turbine inspection card. Inspectors receive inspection tasks through handheld individual devices and efficiently record inspection result data at the inspection site. They can also take photos and record videos of the site to enrich the inspection record content. When an abnormality occurs that cannot be quantified, they can also seek support from resident experts for diagnosis in a timely manner.

[0069] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0070] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be construed as limiting the scope of the patent application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent application shall be subject to the attached claims.

Claims

1. A fan fault detection method based on AI auscultation, characterized in that: include: Collecting a first sound signal during the trouble-free operation of the fan blade, performing digital-to-analog conversion and noise reduction processing on the first sound signal to obtain a sound signal A, extracting features of the sound signal A to obtain a standard feature; The second sound signals of the fan blades during operation under N different fault types are collected respectively, and the N second sound signals are converted into digital-to-analog signals and subjected to noise reduction processing to obtain N sound signals B, and the fault features of the N sound signals B are extracted respectively to obtain a fault feature set, where N is greater than or equal to 50; Collecting a third sound signal during the operation of the fan blade, performing digital-to-analog conversion and noise reduction processing on the third sound signal to obtain a sound signal C, extracting actual features of the sound signal C, and comparing the actual features with the standard features to calculate a first similarity; When the first similarity is less than or equal to a first preset value, the actual feature is compared with any one of the fault feature set, and the second similarities are calculated respectively to obtain N second similarities, and the second similarities are arranged in descending order, and the first M fault features corresponding to the second similarities are taken to obtain the corresponding fault type or a combination of M fault types, where M is less than or equal to 5; After a fault is detected, an alarm will be issued and fault diagnosis advice and maintenance plans will be provided.

2. The fan fault detection method based on AI auscultation according to claim 1 is characterized in that: The characteristics of the sound signal A, the fault characteristics of the sound signal B, and the actual characteristics of the sound signal C all include a time domain waveform diagram, a spectrum diagram, a voiceprint diagram, and a frequency band energy ratio.

3. The fan fault detection method based on AI auscultation according to claim 1 is characterized in that: The calculation of the first similarity and the calculation of the second similarity are both performed by using three methods, namely, structural similarity, cosine distance, and Euclidean distance, to perform analysis and comparison, and three similarity results are obtained, and the largest of the three similarities is taken as the final similarity.

4. The fan fault detection method based on AI auscultation according to claim 1 is characterized in that: After the fault is detected, an alarm is issued and a fault diagnosis suggestion and a maintenance plan are provided, the following steps are also included: Manually confirm the fault type on site and mark it on the actual characteristics of the sound signal C; The fault feature set is stored.

5. The fan fault detection method based on AI auscultation according to claim 1 is characterized in that: After the fault is detected, an alarm is issued and a fault diagnosis suggestion and a maintenance plan are provided, the following steps are also included: Generate operational reports periodically.

6. The fan fault detection method based on AI auscultation according to claim 1 is characterized in that: The digital-to-analog conversion and noise reduction processing of the first sound signal, the digital-to-analog conversion and noise reduction processing of the N second sound signals, and the digital-to-analog conversion and noise reduction processing of the third sound signal all include a filtering processing step through a Butterworth filter, the order of the Butterworth filter is 30 to 50, the lower cutoff frequency is 50 Hz to 200 Hz, the upper cutoff frequency is 10 kHz to 15 kHz, and the square amplitude-frequency response function is: in, N is the order of the filter, Ωc is the 3dB cutoff frequency, Ωu is the upper cutoff frequency, Ω l is the lower cutoff frequency.

7. The fan fault detection system based on AI auscultation is characterized by: The invention comprises a signal acquisition module, a signal processing module, a feature extraction module, a similarity calculation module, a judgment module, a fault determination module and an alarm module, wherein the signal acquisition module is used to collect a first sound signal during the fault-free operation of the fan blade, to collect a second sound signal during the operation of the fan blade under N different fault types, and to collect a third sound signal during the operation of the fan blade, the signal processing module is used to perform digital-to-analog conversion and noise reduction processing on the first sound signal, to perform digital-to-analog conversion and noise reduction processing on N second sound signals, and to perform digital-to-analog conversion and noise reduction processing on the third sound signal, the feature extraction module is used to perform digital-to-analog conversion and noise reduction processing on the N second sound signals, and to perform digital-to-analog conversion and noise reduction processing on the third sound signal, and ... the feature extraction module is used to perform digital-to-analog conversion and noise reduction processing on the N second sound signals, and the feature extraction module is used to perform digital-to-analog conversion and The extraction module is used to extract the characteristics of the sound signal A, to extract the fault characteristics of N sound signals B respectively, and to extract the actual characteristics of the sound signal C. The similarity calculation module is used for the first similarity and for calculating the second similarity. The judgment module is used to judge whether the first similarity is less than or equal to a first preset value. The fault determination module is used to arrange the second similarity in descending order, take the first M fault characteristics corresponding to the second similarity, and obtain the corresponding fault type or a combination of M fault types. The alarm module is used to alarm and provide fault diagnosis advice and maintenance plans after detecting a fault.

8. The fan fault detection system based on AI auscultation according to claim 7 is characterized in that: The feature extraction module includes a time domain waveform extraction unit, a spectrum extraction unit, a voiceprint extraction unit and a frequency band energy ratio extraction unit, and the similarity calculation module includes a structural similarity calculation unit, a cosine distance calculation unit and a Euclidean distance calculation unit.

9. The fan fault detection system based on AI auscultation according to claim 7 is characterized in that: It also includes a report generation module, which is used to generate operation reports regularly.

10. The fan safety system is characterized by: The invention comprises an infrared online temperature measurement module, a behavior environment analysis module, an equipment remote inspection module and a fan fault detection system based on AI auscultation as described in any one of claims 7 to 9, wherein the infrared online temperature measurement module is used for the equipment and components that need to be temperature measured in each inspection at this stage; the behavior environment analysis module is used to support the analysis of personnel wandering, gathering and intrusion around the wind turbine tower base to prevent theft and sabotage by outsiders; The equipment remote inspection module is used to combine video capabilities, audio capabilities and temperature measurement capabilities to perform manual and intelligent inspections on electrical equipment within the station as planned.

Images