A method and system for monitoring dynamic response of a large size fan

Abstract

The application discloses a kind of dynamic response monitoring method and system of large-size fan, belong to electric power equipment remote monitoring technical field, the method is: obtaining the component structure image and sensor data of the component to be detected;According to the external structure change of the component structure image to the component to be detected is detected, obtains the external abnormal characteristics of the component to be detected;According to temperature change data and the material type and structure size of the component to be detected, simulate the internal structure change of the component to be detected, obtain the internal abnormal characteristics of the component to be detected;Based on external abnormal characteristics and internal abnormal characteristics, in combination with meteorological change situation and fan operation data, analyze fault reason, obtain fault diagnosis result.Therefore, by implementing the present application, it can solve the problem that the external and internal of fan component are detected quickly and accurately for fault reason and fault position without suspending the operation of fan in prior art, and a comprehensive and accurate fault diagnosis result is obtained.

Description

Technical Field

[0001] This application belongs to the field of remote monitoring technology for power equipment, specifically relating to a dynamic response monitoring method and system for large-size wind turbines. Background Technology

[0002] As the global energy structure shifts towards cleaner and lower-carbon energy, the proportion of wind power in the power system is increasing year by year. However, wind turbines are prone to structural loosening, corrosion, or foreign objects on their surfaces during long-term operation, leading to abnormal loads. Traditional inspection methods involve manual inspection of the core and vulnerable components of the turbine, but this method is inefficient and costly. Therefore, remote monitoring of wind turbines is currently a key means to ensure reliable operation of wind power generation and reduce maintenance costs.

[0003] Currently, structural images of components or operational data are commonly used to determine if a wind turbine is malfunctioning. External structural images can only identify external anomalies (e.g., foreign objects or cracks on the component surface). For internal anomalies, such as insufficient bearing lubrication or poor gear meshing, further inspection of the internal structure is required. However, due to safety constraints, current practice necessitates suspending wind turbine operation before internal inspection, leading to excessively high inspection costs. Therefore, a comprehensive inspection of the external and internal structures of wind turbine components without halting operation is crucial to accurately determine the cause and location of faults, providing support for timely and appropriate maintenance measures, and further improving the operational stability of the power system. Summary of the Invention

[0004] This application proposes a dynamic response monitoring method and system for large-size wind turbines, which can solve the problem in the prior art of quickly and accurately detecting the cause and location of faults in the external and internal parts of wind turbine components without stopping the operation of the wind turbine, and obtaining comprehensive and accurate fault diagnosis results.

[0005] The first aspect of this application provides a method for monitoring the dynamic response of a large-size wind turbine, the method comprising:

[0006] Acquire structural images and sensor data of the component to be tested in the wind turbine; wherein, the sensor data includes temperature change data, vibration data, and noise data;

[0007] Based on the component structure image and historical normal structure image, the external structural changes of the component to be detected are detected to obtain the external abnormal features of the component to be detected.

[0008] Based on the temperature change data and the material type and structural dimensions of the component under test, the internal structural changes of the component under test are simulated using the noise data and the vibration data to obtain the internal abnormal characteristics of the component under test;

[0009] Based on the external and internal anomaly characteristics, combined with the meteorological changes and wind turbine operation data in the current period, the cause of the failure of the component to be tested is analyzed, and the corresponding fault diagnosis result is obtained.

[0010] The above-described scheme detects anomalies in the external and internal structures of the components under test within the wind turbine. By comparing captured images of the component's structure with historical images of normal structures, it determines whether the external structure of the component has changed, thus obtaining external anomaly characteristics. Then, based on the vibrations and noise emitted by the component during operation, it judges whether internal structural changes have occurred. Furthermore, considering that different material types of the component react differently to temperature changes, the degree of internal structural change is quantified. This allows for the identification of internal structural anomalies in wind turbine components solely through sensor data, without interrupting wind turbine operation, saving significant inspection costs and improving inspection efficiency, while obtaining accurate internal anomaly characteristics. Finally, the impact of meteorological changes and wind turbine operating data on the damage to the component under test is analyzed to determine the causes of failure corresponding to the external and internal anomaly characteristics, resulting in a comprehensive and accurate fault diagnosis. This provides support for subsequent maintenance personnel to implement appropriate maintenance measures.

[0011] In one possible implementation of the first aspect, changes in the external structure of the component to be detected are detected based on the component structure image and historical normal structure images to obtain external abnormal features of the component to be detected, specifically:

[0012] The historical normal structure image of the component to be detected in the previous time period is obtained, the historical normal structure image is compared with the component structure image, and the abnormal area in the component structure image is cropped based on the comparison result to obtain the image to be identified.

[0013] Based on a preset component anomaly type library, the potential fault type of the image to be identified is predicted;

[0014] By using the image to be identified and the collected weight increase / decrease data of the component to be detected, the potential fault types are screened to determine the fault type of each component to be detected;

[0015] Based on the fault type and the image to be identified, construct the external abnormal features of the component to be detected.

[0016] The above scheme first crops out the parts that differ from historical normal structural images to obtain the image to be identified. Then, it matches the image with a component anomaly type library to predict the possible potential fault type for each image to be identified. To further determine what kind of external fault has occurred in the component to be inspected, the weight increase or decrease data of the component to be inspected is used to determine whether the component is aging, structurally deformed, or has foreign objects on its surface, thus achieving comprehensive detection of external anomalies of the component.

[0017] In one possible implementation of the first aspect, the potential fault types are screened using the image to be identified and the collected weight increase / decrease data of the component to be detected, to determine the fault type of each component to be detected, specifically as follows:

[0018] If there is only one potential fault type in an image to be identified, then that potential fault type is taken as the fault type of the component to be detected.

[0019] If there is more than one potential fault type in an image to be identified, the weight increase or decrease data of the component to be detected is obtained, and the surface texture and color distribution of the component to be detected are obtained from the image to be identified; the confidence level of each potential fault type is calculated based on the weight increase or decrease data, the surface texture and color distribution, and the potential fault type with the highest confidence level is selected as the fault type of the component to be detected.

[0020] The above solution combines weight gain / loss data, surface texture, and color distribution to more accurately determine the type of external fault in the component under test for multiple potential fault types. For uncertain fault types, the solution can determine whether the component under test has foreign objects, is corroded, or has abnormal textures by analyzing the component's weight changes, surface texture, and color distribution, thus achieving visual accuracy in determining the fault type of each component under test's external condition.

[0021] In one possible implementation of the first aspect, based on the temperature change data and the material type and structural dimensions of the component to be tested, the internal structural changes of the component to be tested are simulated using the noise data and the vibration data to obtain the internal abnormal characteristics of the component to be tested, specifically as follows:

[0022] Based on the noise and vibration characteristics of the component under test under normal operating conditions, abnormal noise signals and abnormal vibration signals are extracted from the noise data and vibration data, respectively.

[0023] By aligning and fusing abnormal noise signals and abnormal vibration signals, abnormal change characteristics can be obtained.

[0024] With the goal of maximizing the similarity to the abnormal change characteristics, the internal structural changes of the component under test are simulated based on the temperature change data and the material type, according to the normal structural dimensions of the component under test in the previous period, to obtain the internal abnormal characteristics of the component under test.

[0025] The above scheme takes into account the difficulty of directly photographing the internal structure of the component under test. Therefore, it collects noise and vibration data of the component during operation using sensors, and extracts abnormal data that differs from normal operating conditions from these data. Since vibration and noise are correlated, the acquired abnormal noise and vibration signals are fused to obtain abnormal change characteristics. Then, the operation of the component is simulated based on the normal structural dimensions of a previous period. When the component emits vibrations and noises highly similar to the abnormal change characteristics during the simulation, the simulation results at this point can be considered highly similar to the actual internal structure of the component under test. This allows for the determination of which internal structural changes have occurred in the component under test. Accurate internal abnormal characteristics can be safely obtained without stopping the wind turbine operation, saving significant inspection costs.

[0026] In one possible implementation of the first aspect, based on the normal structural dimensions of the component under test in the previous period, the internal structural changes of the component under test are simulated using the temperature change data and the material type to obtain the internal abnormal characteristics of the component under test, specifically:

[0027] Based on the temperature change data, the influence of temperature on each of the material types is simulated to obtain the first structural characteristics of each of the components to be tested;

[0028] Based on the connection structure between the component under test and other components, the degree of damage to the component under test caused by friction due to the material type during operation is simulated to obtain the second structural characteristics of each component under test;

[0029] Based on the first structural characteristic and the second structural characteristic, the internal structural changes of the component under test are continuously simulated on the basis of the normal structural dimensions. At the same time, the similarity between the noise signal and vibration signal generated by the component under test and the abnormal change characteristics is calculated under each internal structural change.

[0030] The simulation operation stops when the similarity is highest, and the current simulation result is compared with the normal structure size to construct the internal abnormal features.

[0031] To accurately determine the internal structural changes of a component, the above scheme considers that different material types of components will cause different structural changes under the same temperature change, resulting in different noise and vibration conditions. Therefore, a first structural characteristic of the component under test is constructed. Furthermore, considering the influence of the connections between components on noise and vibration, as well as component wear, a second structural characteristic of the component under test is constructed. Finally, based on the first and second structural characteristics, the internal structural changes of the component under test are accurately simulated to obtain internal anomaly features that more closely resemble actual conditions.

[0032] In one possible implementation of the first aspect, the abnormal noise signal and the abnormal vibration signal are aligned and fused to obtain abnormal change characteristics, specifically:

[0033] Calculate the cross-correlation coefficient and common peak value of the abnormal noise signal and the abnormal vibration signal;

[0034] Correlation characteristics were obtained by analyzing the correlation between abnormal noise signals and abnormal vibration signals in the time and frequency domains using wavelet transform.

[0035] Based on the cross-correlation coefficient, the common peak value, and the correlation features, a mapping relationship between abnormal noise signals and abnormal vibration signals is constructed. Then, the abnormal noise signals and abnormal vibration signals are fused using the mapping relationship to obtain abnormal change features.

[0036] The above scheme takes into account the correlation between the noise and vibration generated by the component during operation. By finding peak values ​​and aligning them in time, this correlation is captured, and relevant features are obtained to construct the corresponding mapping relationship, which provides support for accurately determining the location of structural changes in the component to be detected.

[0037] In one possible implementation of the first aspect, based on the external and internal anomaly characteristics, combined with the meteorological changes and wind turbine operation data within the current time period, the cause of the fault in the component to be tested is analyzed to obtain the corresponding fault diagnosis result, specifically as follows:

[0038] Extract the service life and operating condition change data of each component to be tested from the wind turbine operation data;

[0039] Based on the external and internal anomaly characteristics, corresponding damage association labels are constructed for the service life, the operating condition change data, and the meteorological change, respectively.

[0040] Using the external and internal abnormal features as result variables, causal reasoning is performed on the fault causes of the component to be detected to obtain the influence weight of each damage association label and the causal graph related to the fault causes;

[0041] By using the path verification method and the influence weights, the cause-effect graph is analyzed for direct and indirect causes to obtain the fault diagnosis results; wherein, the fault diagnosis results include the fault cause, fault location and fault type of the component under test.

[0042] Since the causes of external and internal anomalies differ, and various causes are coupled, the above-mentioned scheme constructs corresponding damage association labels for the service life, operating condition changes, and meteorological changes to support subsequent multi-factor analysis of fault causes. Then, using a reverse inference method, taking the structural anomalies of the component as the result, and the impact degree and outcome of the damage association labels on the component under test, fault analysis is performed. A cause-effect graph is constructed to describe the relationship between fault causes and component structural anomalies, and to characterize the coupling relationships between different fault causes. This enables the prediction of direct and indirect causes of the component under test, resulting in accurate and comprehensive fault diagnosis results.

[0043] In one possible implementation of the first aspect, based on the external anomaly characteristics and the internal anomaly characteristics, corresponding damage association labels are constructed for the service life, the operating condition change data, and the meteorological change conditions, respectively, specifically as follows:

[0044] Based on historical inspection records, the degree of damage to the component under inspection due to the service life is quantified, and the influence characteristics of the service life are constructed.

[0045] Based on the sampling frequency and component load variation curve, the degree of damage impact of the operating condition variation data on the component under test is quantified, and the impact characteristics of the operating condition variation data are constructed.

[0046] Based on the long-term and short-term characteristics of the meteorological changes, the impact characteristics of the meteorological changes are constructed.

[0047] Based on the external and internal abnormal features, damage association labels are generated for each of the impact features by searching for the causes of failures in the historical fault records of the wind turbine.

[0048] The above scheme uses existing data to quantify the degree of damage to the component under test caused by the service life, the changes in operating conditions, and the changes in weather conditions, thereby obtaining corresponding impact characteristics to characterize the damage that these factors will cause to the component, and providing data support for subsequent fault cause deduction.

[0049] In one possible implementation of the first aspect, using the external and internal abnormal features as result variables, causal reasoning is performed on the fault causes of the component to be detected to obtain the influence weight of each damage association label and a causal graph related to the fault causes, specifically:

[0050] Based on historical fault records, the external and internal abnormal features are used as outcome variables, the wind turbine operating data is used as intermediate response variables, and the damage association labels are used as potential cause variables.

[0051] Based on the relationship between the potential causal variables and the intermediate response variables, and the relationship between the outcome variables and the intermediate response variables, a causal graph skeleton is constructed through conditional independence testing.

[0052] Based on the causal effect of the potential causal variables on the intermediate response variables, generate the influence weight of each damage association label;

[0053] The causal graph skeleton is labeled according to the influence weights, and a causal forest is constructed on the labeled causal graph skeleton with the result variable as the endpoint to obtain the causal graph.

[0054] The above scheme uses existing historical fault records as causal events and the external and internal abnormal features as the results of these events. It deduces the probability of fault causes through the relationships provided by the causal events, resulting in a causal graph skeleton. Since different potential causal variables can cause varying degrees of change in wind turbine operating data—for example, the degree of wind speed change has a certain linear relationship with the degree of load increase on wind turbine blades—weights are assigned to each damage association label based on this causal effect. This further analyzes whether the fault is a direct or indirect cause, resulting in a causal graph that comprehensively displays the various fault causes of the component under test and the coupling relationships between these causes.

[0055] The second aspect of this application provides a dynamic response monitoring system for large-size wind turbines, the system comprising: a data acquisition module, an external feature acquisition module, an internal feature acquisition module, and a fault cause analysis module;

[0056] The data acquisition module is used to acquire component structure images and sensor data of the wind turbine's components to be tested; wherein the sensor data includes temperature change data, vibration data, and noise data.

[0057] The external feature acquisition module is used to detect changes in the external structure of the component to be detected based on the component structure image and historical normal structure images, and to obtain the external abnormal features of the component to be detected.

[0058] The internal feature acquisition module is used to simulate the internal structural changes of the component under test based on the temperature change data and the material type and structural dimensions of the component under test, through the noise data and the vibration data, to obtain the internal abnormal features of the component under test;

[0059] The fault cause analysis module is used to analyze the fault causes of the component under test based on the external and internal abnormal characteristics, combined with the meteorological changes and wind turbine operation data in the current period, and obtain the corresponding fault diagnosis results. Attached Figure Description

[0060] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0061] Figure 1 This is a schematic flowchart illustrating a dynamic response monitoring method for a large-size fan provided in one embodiment of this application.

[0062] Figure 2 This is a structural diagram of a dynamic response monitoring system for a large-size fan provided in one embodiment of this application. Detailed Implementation

[0063] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0064] It should be understood that the step numbers used in the text are for ease of description only and are not intended to limit the order in which the steps are performed.

[0065] First Embodiment

[0066] Wind turbines are key equipment in wind power generation systems, and their structural health directly affects the system's operational efficiency and safety. To improve the efficiency and timeliness of turbine component inspections, sensors and drones are commonly used to collect information on critical components such as blades, bearings, and gears for remote monitoring. However, monitoring methods for the internal structure of wind turbines are still relatively limited. Simply collecting temperature and vibration data is insufficient for a comprehensive analysis of the causes and locations of turbine failures; further on-site inspections are necessary to obtain accurate fault analysis results. Therefore, this application addresses these issues by further optimizing the internal structure inspection of wind turbine components. By using remotely collected multi-source data, it can accelerate and accurately detect the causes and locations of external and internal faults in wind turbine components, resulting in comprehensive and accurate fault diagnosis results.

[0067] like Figure 1 As shown, to address the problem in the prior art of quickly and accurately detecting the cause and location of faults in the external and internal components of wind turbines without stopping wind turbine operation, and obtaining comprehensive and accurate fault diagnosis results, the first embodiment of this application provides a detailed flowchart of a dynamic response monitoring method for large-size wind turbines. The dynamic response monitoring method for large-size wind turbines in this embodiment includes steps S1 to S4, detailed below:

[0068] Step S1: Obtain the component structure image and sensor data of the component to be tested in the wind turbine.

[0069] Step S1 in this embodiment mainly involves remotely collecting data related to the structure of the core components and vulnerable components of the wind turbine using various sensors and drones, thereby obtaining a comprehensive understanding of the internal and external structure of the wind turbine and providing sufficient data support for subsequent detection of component abnormalities.

[0070] The external structure of a wind turbine mainly includes components that are directly exposed to the environment and bear external influences such as wind loads and climate impacts. These include blades (used to capture wind energy), the tower (used to support the turbine nacelle and blades and increase the wind capture height), and the hub (used to connect the blades and the main shaft). These external components can be photographed from multiple angles using drones to obtain corresponding structural images. These images can then be used to determine whether there are any abnormal changes in the structure, surface, and installation position of the components, thus enabling fault detection of the wind turbine's external structure.

[0071] Among them, the structural anomalies of the blades mainly include cracks and fractures caused by long-term use, lightning damage or manufacturing defects, as well as surface erosion caused by long-term erosion by sand and raindrops; the structural anomalies of the tower mainly include tower tilting caused by strong wind load or insufficient bolt preload, and weld and bolt fractures caused by installation errors or corrosion.

[0072] The malfunctions of these external wind turbine components can be observed through images. However, drones cannot directly photograph components inside the wind turbine nacelle. Therefore, it is necessary to use sensors to collect multi-source data on the internal structure of the wind turbine to accurately determine which malfunctions have occurred inside the wind turbine without the need for manual disassembly of components.

[0073] In this embodiment of the application, multi-source sensor data, including temperature change data, vibration data, and noise data, are collected by temperature sensors, vibration sensors, and noise sensors installed inside the fan. This data is then used to locate the fault by detecting whether the component temperature is too high or whether the component emits abnormal vibrations and noise during operation.

[0074] The internal structure of a wind turbine is concentrated in the nacelle, where energy conversion and control are achieved. Key components include the gearbox (primarily for controlling blade speed), generator, control system (for adjusting blade angle), and cooling system (including radiators and refrigeration units to prevent overheating of the gearbox and generator). These components automatically generate heat during operation. If heat cannot be dissipated in time, it can lead to overheating, causing turbine malfunctions such as insulation aging, permanent magnet demagnetization, and mechanical deformation. Furthermore, malfunctions in the mechanical components can generate abnormal noise during operation, and the vibration frequency of the components can change. Therefore, collecting vibration and noise data can improve the accuracy of wind turbine fault diagnosis and significantly reduce unplanned downtime losses caused by manual inspections.

[0075] Among them, gearbox failures mainly include pitting and tooth breakage caused by poor lubrication or overload; generator failures mainly include generator short circuits caused by insulation aging, reduced power generation efficiency caused by high temperature, and demagnetization of permanent magnets; control system failures include bearing jamming caused by excessive bearing temperature and shortened capacitor lifespan caused by prolonged use of capacitors.

[0076] Optionally, when the remote monitoring system detects abnormalities in certain electrical parameters of the wind turbine's operating data, such as sudden voltage drops / surges, current overloads, large power fluctuations, and decreased insulation resistance, it will use drones and sensors to collect data on vulnerable and critical components of the wind turbine, obtaining component structure images and sensor data.

[0077] Step S2: Based on the component structure image and historical normal structure image, detect the external structural changes of the component to be detected to obtain the external abnormal features of the component to be detected.

[0078] Step S2 in this embodiment mainly involves using real-time captured images of the component structure of the component to be tested in the wind turbine to detect abnormal conditions on the outside of the wind turbine exposed to the external environment, further determining what faults have occurred on the outside of the component to be tested, obtaining the external abnormal characteristics of the component to be tested, and realizing comprehensive detection of external abnormalities of the component.

[0079] First, acquire historical normal structural images of the component to be inspected from the previous time period. The previous time period refers to the period during the last wind turbine inspection. These historical normal structural images are images of the component structure in a normal state taken during the last inspection. Then, compare the component structure images with historical normal structural images taken from the same angle, specifically comparing aspects such as the presence of cracks, surface textures, foreign objects, whether the installation position has shifted, and the wear condition of the component, to obtain detailed comparison results.

[0080] Based on the comparison results, abnormal areas that differ significantly from historical images in the component structure images are labeled and cropped. These cropped abnormal areas are then used as images to be identified for further identification of the component's fault type.

[0081] Based on a pre-defined component anomaly type library, the images to be identified are matched against data in the database to predict one or more potential fault types for each image. This component anomaly type library is built upon historical fault records of wind turbines, covering most common fault conditions for wind turbine components and storing a large number of images showing structural damage to each component to achieve accurate data matching.

[0082] Specifically, each image to be identified is labeled with the location of the wind turbine component from which it was taken. Based on the labeled location of the component, the corresponding image is found in the component anomaly type database and matched. When the confidence level of the matching result reaches the set threshold, it means that a fault image that is very similar to the image to be identified has been found. Based on the fault information of the fault image, the potential fault type of the image to be identified is determined.

[0083] For example, for an image to be identified about the leading edge of a blade, a small area of ​​the blade is a different color from the surrounding area. After searching the component anomaly type database, it is determined that there are multiple potential fault types in the image to be identified, including blade icing, coating peeling, or pitting.

[0084] If an image to be identified has only one potential fault type, then that potential fault type can be directly taken as the fault type of the component to be detected. However, for cases with multiple potential fault types, it is necessary to obtain the weight increase or decrease data of the component to be detected and to perform more detailed feature extraction on the image to be identified in order to help determine the most likely potential fault type.

[0085] The weight change data can be used to prove whether there are foreign objects or material detachment on the component under test. If the weight of the component does not change significantly, for example, if the component is deformed, then it is necessary to extract visual features related to the surface texture and color distribution of the component under test from the image to be identified, in order to further determine whether there are abnormalities such as cracks, corrosion or failure of composite material adhesion on the component under test.

[0086] By combining data on weight increase / decrease, surface texture, and color distribution, the confidence level of each potential fault type is calculated, and the potential fault type with the highest confidence level is selected as the fault type of the component to be tested.

[0087] Furthermore, for components such as towers, it is also possible to detect whether the components are vibrating or tilting. By using surrounding reference objects in the image, it can be determined whether the tilt of the component has worsened compared to the previous period, thereby determining whether the component has malfunctioned.

[0088] Step S3: Based on the temperature change data and the material type and structural dimensions of the component to be tested, the internal structural changes of the component to be tested are simulated using the noise data and the vibration data to obtain the internal abnormal characteristics of the component to be tested.

[0089] In step S3 of this embodiment, in order to accurately obtain the internal structure of the wind turbine without shutting down the machine, multiple sensors are set up to collect multi-source data from aspects such as temperature, vibration and noise, so as to detect the noise characteristics and vibration characteristics of the component under test in the current period, and then detect the internal structural changes of the component under test. The internal abnormal characteristics that can reflect what internal structural changes have occurred in the component under test and help determine the location of the fault are obtained. This enables efficient internal inspection of the wind turbine without unplanned shutdown, saving a lot of inspection costs.

[0090] First, as a control group, the noise and vibration characteristics of these components under normal operating conditions are obtained. Then, the corresponding abnormal noise and vibration signals are extracted from the noise and vibration data collected in the current time period. The normal data and abnormal data are compared to roughly identify whether there are any abnormalities inside the components under test in the current time period.

[0091] The noise generated by wind turbine components during operation includes mechanical noise, aerodynamic noise, electromagnetic noise, and structural noise. Mechanical noise includes gearbox noise, generator noise, and hydraulic system noise, which are generated by gear meshing, cooling fan rotation, and fluid pulsation caused by the operation of hydraulic pumps and valves, respectively. These are the main sources of noise during wind turbine operation. Aerodynamic noise is mainly generated by airflow passing over the blades. For example, the hissing noise generated by turbulence when airflow passes over the blade surface is typical. The spectrum of aerodynamic noise is usually low to mid-frequency with obvious noise peaks. Electromagnetic noise is caused by unbalanced electromagnetic forces causing electromagnetic vibrations in the generator, which are then radiated through solid structures. Structural noise is generated by the vibration of structures such as the nacelle and tower under power transmission and external excitation, which is then propagated through the structure.

[0092] Factors affecting wind turbine noise include wind speed, blade structure design, and operating conditions. For example, mechanical noise is more pronounced when components are old or poorly maintained. Gears may produce loud metallic clanging sounds due to poor meshing, tooth surface peeling, or broken teeth. Similarly, hydraulic systems may generate bubbling noise in the pipelines due to air in the oil or leaks. Therefore, analyzing collected noise data can pinpoint the noise source, providing support for subsequent fault location. Furthermore, combining data on vibration, temperature, and other factors can help determine the cause of component failures.

[0093] During operation, wind turbine components generate complex vibration signals due to mechanical motion, airflow disturbances, and other factors. These signals can be used for fault diagnosis of wind turbine components. For example, when a gearbox tooth breaks, the amplitude of the vibration signal generated by gear meshing increases suddenly, and the sidebands broaden. When the generator rotor is eccentric, the amplitude of its vibration signal changes significantly, possibly accompanied by electrical frequency modulation.

[0094] Based on the above description, noise and vibration data are compared with the noise and vibration characteristics of the component under normal operating conditions in terms of signal amplitude, frequency, and bandwidth. Signal data that are significantly different from normal characteristics are extracted as abnormal noise and abnormal vibration signals.

[0095] Since both vibration and noise signals are related to mechanical vibration and aerodynamics, and there is a certain coupling relationship between them but they are also different, the accuracy and reliability of fault detection can be improved by combining abnormal noise signals and abnormal vibration signals for analysis, so as to obtain accurate internal abnormal characteristics of the component under test.

[0096] In this embodiment, the abnormal noise signal is first denoised to remove environmental interference, the trend term of the abnormal vibration signal is eliminated, and the effective frequency band is extracted, thus completing signal preprocessing. Then, based on the coupling relationship between the two, the cross-correlation coefficient and common peak value of the preprocessed abnormal noise signal and abnormal vibration signal are calculated.

[0097] For example, when bearing damage occurs, noise impact and vibration peaks will appear simultaneously. At this time, the noise signal and vibration signal generated by the bearing will have a certain correlation, and the cross-correlation coefficient of the signals can be calculated accordingly. When a gear tooth breaks, the vibration signal and noise signal generated during gear meshing will have a peak at the same point in time, and this peak can be used as the common peak value.

[0098] Abnormal noise and vibration signals are decomposed into wavelets of different frequency bands by wavelet transform, and the correlation of the signals in the time and frequency domains is analyzed to obtain correlation features.

[0099] In the time domain, different groups of wavelets are translated or scaled to capture the local features of the signal on the time axis. By calculating the energy of the wavelet coefficients within each time window, the local intensity of the signal is reflected, and the local features of the signal on the time axis are acquired. Because the impact caused by component failure will appear in the wavelet coefficients of vibration and noise signals at specific frequency bands, the signal is decomposed into multiple scales to obtain wavelet coefficients from low frequency to high frequency bands. From these, time-domain abrupt change points are found, and the correlation between abnormal noise signals and abnormal vibration signals in the time domain is determined more precisely.

[0100] In the frequency domain, time-frequency analysis is performed by calculating the energy proportion of each wavelet group. The frequency band with concentrated energy in each wavelet group is identified as the dominant frequency band. The correlation between abnormal noise signals and abnormal vibration signals in the frequency domain is determined by the dominant frequency band.

[0101] Then, by combining the correlation in the time domain and the correlation in the frequency domain, correlation features are constructed.

[0102] Optionally, the number of wavelet transform decomposition layers can be determined based on the highest frequency and sampling rate of the abnormal noise signal and abnormal vibration signal.

[0103] By analyzing the correlation of signals in the time domain, transient events can be located, and the exact time period of the fault can be pinpointed, providing support for subsequent fault cause identification by combining meteorological and operational condition changes. By analyzing the correlation of signals in the frequency domain, periodic impacts can be detected, providing a basis for determining the fault type.

[0104] By combining the cross-correlation coefficient, the common peak value, and the correlation features, a mapping relationship between abnormal noise signals and abnormal vibration signals is constructed. Then, the abnormal noise signals and abnormal vibration signals are fused using this mapping relationship to obtain abnormal change characteristics.

[0105] Then, aiming for the highest similarity to the abnormal change features, an internal structural change simulation is performed based on the normal structural dimensions of the component under test in the previous time period. Simultaneously, during the internal structural change process, the similarity between the noise and vibration signals emitted by the simulation results and the abnormal change features is calculated. The simulation stops when the similarity reaches its highest value, completing the simulation of the actual changes in the component under test. The simulation result corresponding to the highest value is compared with the normal structural dimensions, and the corresponding internal abnormal features are obtained by acquiring the internal structural changes of the component under test.

[0106] Considering that each component under test has different material type and structural shape, and that the connection structure with other connected components will also generate different degrees of friction, the characteristics of the component under test are also evaluated based on temperature and the connection structure, so as to more accurately simulate the changes in the internal structure of the component under test under environmental influence.

[0107] Temperature changes have significantly different effects on components made of different materials, primarily depending on the material's physical properties (such as coefficient of thermal expansion, thermal conductivity, and temperature limit). The material types of the components to be tested in a wind turbine are mainly categorized as metals (such as gearbox housings and towers), composite materials (such as blades), polymer materials (such as component sealing layers and cable insulation layers), ceramic and permanent magnet materials (such as permanent magnets and ceramic insulators), lubricating materials (such as grease), and electronic materials (such as circuit boards). This application's embodiments simulate the degree of temperature influence on each material type to further obtain the different responses of each component to temperature changes, thus obtaining the first structural characteristics of each component related to temperature.

[0108] For example, both are gearbox housings, but within the same temperature range, aluminum alloy products are more prone to deformation than iron products. Therefore, temperature has a greater impact on aluminum alloy gearbox housings, and the internal structural deformation of aluminum alloy gearbox housings is more pronounced.

[0109] Besides temperature, differences in the connection structure or method between the component under test and other components can also cause varying degrees of damage to the component under test. For example, even with bearing connections, sliding bearings generate less friction than rolling bearings; similarly, with gear meshing, helical gears, due to their larger contact area, may generate higher friction than spur gears. Considering that friction generated by the same material type can also lead to different degrees of deformation for each component due to differences in the connection structure or method, this application embodiment further simulates the degree of damage to each component under test caused by friction during operation based on the connection structure or method, thereby obtaining a second structural characteristic for each component under test.

[0110] Combining the first structural feature and the second structural characteristic, based on the normal structural dimensions of the component under test in the previous period and the operating conditions of the wind turbine in the current period, the internal structural changes of the component under test are continuously simulated to obtain the internal abnormal features of the component under test.

[0111] Step S4: Based on the external and internal abnormal features, combined with the meteorological changes and wind turbine operation data in the current period, analyze the cause of the fault of the component to be tested, and obtain the corresponding fault diagnosis result.

[0112] Because external and internal anomalies contain a large amount of fault information about the components to be tested, in order to further determine the cause of the fault, assist maintenance personnel in quickly carrying out repairs and implementing better protective measures, it is also necessary to combine the current meteorological changes and wind turbine operating data to deduce the cause of the fault and obtain a comprehensive and accurate fault diagnosis result.

[0113] Data such as the service life and operating condition changes of each component to be tested are extracted from the wind turbine operation data. Then, based on external and internal abnormal characteristics, corresponding damage association labels are constructed for the above data and meteorological changes, so as to associate the abnormal conditions of the components with the corresponding failure causes.

[0114] Specifically, based on the historical inspection records of the wind turbine, the correlation between the service life of components and the frequency of component damage is analyzed, the relationship between the current service life of components and the degree of damage impact is quantified, and the influence characteristics of service life are constructed. For example, as the operating time increases, blades are more prone to cracking due to accumulated mechanical fatigue. When the service life of the blades exceeds 10 years, the influence characteristics of service life are more obvious than when they were first put into use, and there are more associated damage tags.

[0115] Based on the sampling frequency and component load variation curves, the degree of damage impact of the operating condition changes on the component under test is quantified. This is because the component load variation curve of the wind turbine is closely related to the turbine's operating conditions and can reflect whether the turbine is operating under load within a certain period. When the component load variation curve changes significantly, it indicates that the component is likely to experience fatigue damage; when the component load variation curve shows extreme conditions, it indicates that the component is likely to suffer a reduction in structural safety due to extreme operating conditions. The sampling frequency affects the accuracy of the impact characteristics of the operating condition change data; the higher the sampling frequency, the more accurate the impact characteristics obtained.

[0116] Then, based on the long-term and short-term characteristics of meteorological changes, the impact characteristics of the meteorological changes are constructed. This is because meteorological changes affect the switching of wind turbine operating modes. For example, in high wind speeds, to avoid damage to blades, towers, and other structures due to overload, blade operation is suspended through feathering, mechanical braking, and yaw to avoid wind; in sandstorm weather, derated operation is required to protect the motor and frequency converter, etc.

[0117] Furthermore, to improve the accuracy of the impact characteristics of meteorological changes, this embodiment of the application further divides the impact of meteorological changes into long-term and short-term characteristics. Short-term characteristics include wind speed turbulence intensity, temperature gradient, and lightning density, while long-term characteristics include the average number of freezing rain days per year and the frequency of sandstorms. By using long-term and short-term characteristics, the constructed impact characteristics of meteorological changes can reflect the impact of extreme weather and seasonal changes on the component under test.

[0118] In addition, to further improve data accuracy, this application embodiment also verifies meteorological data according to timestamps and wind turbine latitude and longitude to reduce data latency.

[0119] Finally, based on the external and internal abnormal features of the component to be tested, the damage association label for each influencing feature is obtained by searching for the fault events caused by each influencing feature in the existing historical fault records of the wind turbine.

[0120] Then, based on the existing external and internal abnormal features containing the fault conditions of the component to be tested, a causal graph related to each damage association label and fault cause is constructed, which facilitates the reverse deduction of the fault cause of the component to be tested.

[0121] First, it's necessary to clarify the causal relationship. Based on historical fault records, external and internal anomalies are used as outcome variables, wind turbine operating data as intermediate response variables, and damage association labels as potential causal variables. Wind turbine operating data, such as speed, power, current, and voltage, are used as intermediate response variables because they are results of the fault. Damage association labels, as factors that may cause the fault in the component under test, are used as potential causal variables, forming the basis for subsequent fault cause screening.

[0122] Then, through conditional independence testing, based on the relationships that the potential causal variables cause changes in the intermediate response variables and that the intermediate response variables are representations of the outcome variables, a causal graph skeleton is constructed between the damage association labels and external and internal anomalous features.

[0123] For example, a conditional independent test is used to examine whether changes in wind speed can significantly predict changes in blade vibration. A pre-defined model is used to fit the effect of wind speed on vibration while controlling for confounding variables such as temperature, thus constructing a directed path between wind speed and vibration. To improve the accuracy of the test results, it is also necessary to eliminate interference from other variables during the test.

[0124] Optionally, the Fisher-Z test method is used in the embodiments of this application.

[0125] Simultaneously, based on the causal effect of the latent causal variable on the intermediate response variable, an influence weight is generated for each damage association label. The causal effect describes the degree of influence of the latent causal variable on the intermediate response variable; for example, for every 1 m / s increase in wind speed, the RMS value of vibration will increase by 0.2 g. The influence weight characterizes the probability that the damage association label is a cause of component failure.

[0126] Next, nodes in the cause-effect graph skeleton are labeled according to the influence weights. Then, using the outcome variable as the endpoint, a causal forest is constructed on the labeled cause-effect graph skeleton in a way that distinguishes between direct and indirect causes. This results in a causal graph of fault causes related to external and internal anomalies. This causal graph comprehensively displays the various fault causes of the component under test and the coupling relationships between these causes. The distinction between direct and indirect causes is based on the influence weights.

[0127] For example, if a generator experiences leakage, the direct cause is insulation aging, and the indirect cause is insulation failure due to long-term high-temperature operation. Therefore, the cause-effect diagram will distinguish the direct and indirect causes of the generator failure based on the length of the path, which will facilitate the generation of more comprehensive fault diagnosis results.

[0128] Finally, the cause-effect graph is analyzed using the path verification method to determine the direct and indirect causes of the fault in the component under test. By identifying external and internal abnormal features, the location and type of the fault are determined, resulting in accurate and comprehensive fault diagnosis. This provides support for timely implementation of appropriate maintenance measures and can further improve the operational stability of the power system.

[0129] Implementing the embodiments of this application has the following beneficial effects:

[0130] This application embodiment detects abnormalities in the external and internal structures of components under test within a wind turbine. By comparing captured images of the component's structure with historical images of normal structures, it determines whether the external structure of the component has changed, thus obtaining external anomaly characteristics. Then, based on the vibrations and noise emitted by the component during operation, it judges whether internal structural changes have occurred. Furthermore, considering that different material types of the component react differently to temperature changes, the degree of internal structural change is quantified. This allows for the identification of internal structural anomalies in wind turbine components solely through sensor data, without interrupting wind turbine operation, saving significant inspection costs and improving inspection efficiency, while obtaining accurate internal anomaly characteristics. Finally, the impact of meteorological changes and wind turbine operating data on damage to the component under test is analyzed to determine the causes of failure corresponding to the external and internal anomaly characteristics, resulting in a comprehensive and accurate fault diagnosis. This provides support for subsequent maintenance personnel to implement appropriate maintenance measures.

[0131] Furthermore, in order to implement the dynamic response monitoring system for large-size wind turbines corresponding to the above method embodiments, and to achieve the corresponding functions and technical effects, Figure 2 A structural diagram of a dynamic response monitoring system for large-size wind turbines is provided. For ease of explanation, only the parts relevant to this embodiment are shown. The dynamic response monitoring system for large-size wind turbines provided in this application embodiment includes:

[0132] The data acquisition module 201 is used to acquire component structure images and sensor data of the component to be tested in the wind turbine; wherein, the sensor data includes temperature change data, vibration data and noise data.

[0133] The external feature acquisition module 202 is used to detect changes in the external structure of the component to be detected based on the component structure image and historical normal structure images, and to obtain the external abnormal features of the component to be detected.

[0134] In this embodiment of the application, the historical normal structure image of the component to be detected in the previous time period is obtained, the historical normal structure image is compared with the component structure image, and the abnormal area in the component structure image is cropped based on the comparison result to obtain the image to be identified.

[0135] Based on a preset component anomaly type library, the potential fault type of the image to be identified is predicted;

[0136] By using the image to be identified and the collected weight increase / decrease data of the component to be detected, the potential fault types are screened to determine the fault type of each component to be detected;

[0137] Based on the fault type and the image to be identified, construct the external abnormal features of the component to be detected.

[0138] The internal feature acquisition module 203 is used to simulate the internal structural changes of the component under test based on the temperature change data and the material type and structural dimensions of the component under test, through the noise data and the vibration data, to obtain the internal abnormal features of the component under test.

[0139] In this embodiment of the application, based on the noise and vibration characteristics of the component to be tested under normal operating conditions, abnormal noise signals and abnormal vibration signals are extracted from the noise data and the vibration data, respectively.

[0140] By aligning and fusing abnormal noise signals and abnormal vibration signals, abnormal change characteristics can be obtained.

[0141] With the goal of maximizing the similarity to the abnormal change characteristics, the internal structural changes of the component under test are simulated based on the temperature change data and the material type, according to the normal structural dimensions of the component under test in the previous period, to obtain the internal abnormal characteristics of the component under test.

[0142] The fault cause analysis module 204 is used to analyze the fault cause of the component to be tested based on the external abnormal characteristics and the internal abnormal characteristics, combined with the meteorological changes and wind turbine operation data in the current period, and obtain the corresponding fault diagnosis result.

[0143] In this embodiment of the application, based on the temperature change data, the degree of influence of temperature on each of the material types is simulated to obtain the first structural characteristics of each of the components to be tested;

[0144] Based on the connection structure between the component under test and other components, the degree of damage to the component under test caused by friction due to the material type during operation is simulated to obtain the second structural characteristics of each component under test;

[0145] Based on the first structural characteristic and the second structural characteristic, the internal structural changes of the component under test are continuously simulated on the basis of the normal structural dimensions. At the same time, the similarity between the noise signal and vibration signal generated by the component under test and the abnormal change characteristics is calculated under each internal structural change.

[0146] The simulation operation stops when the similarity is highest, and the current simulation result is compared with the normal structure size to construct the internal abnormal features.

[0147] In some embodiments, the data acquisition module 201 specifically comprises:

[0148] The external structure of a wind turbine mainly includes components that are directly exposed to the environment and bear external influences such as wind loads and climate impacts. These include blades (used to capture wind energy), the tower (used to support the turbine nacelle and blades and increase the wind capture height), and the hub (used to connect the blades and the main shaft). These external components can be photographed from multiple angles using drones to obtain corresponding structural images. These images can then be used to determine whether there are any abnormal changes in the structure, surface, and installation position of the components, thus enabling fault detection of the wind turbine's external structure.

[0149] Among them, the structural anomalies of the blades mainly include cracks and fractures caused by long-term use, lightning damage or manufacturing defects, as well as surface erosion caused by long-term erosion by sand and raindrops; the structural anomalies of the tower mainly include tower tilting caused by strong wind load or insufficient bolt preload, and weld and bolt fractures caused by installation errors or corrosion.

[0150] The malfunctions of these external wind turbine components can be observed through images. However, drones cannot directly photograph components inside the wind turbine nacelle. Therefore, it is necessary to use sensors to collect multi-source data on the internal structure of the wind turbine to accurately determine which malfunctions have occurred inside the wind turbine without the need for manual disassembly of components.

[0151] In this embodiment of the application, multi-source sensor data, including temperature change data, vibration data, and noise data, are collected by temperature sensors, vibration sensors, and noise sensors installed inside the fan. This data is then used to locate the fault by detecting whether the component temperature is too high or whether the component emits abnormal vibrations and noise during operation.

[0152] The internal structure of a wind turbine is concentrated in the nacelle, where energy conversion and control are achieved. Key components include the gearbox (primarily for controlling blade speed), generator, control system (for adjusting blade angle), and cooling system (including radiators and refrigeration units to prevent overheating of the gearbox and generator). These components automatically generate heat during operation. If heat cannot be dissipated in time, it can lead to overheating, causing turbine malfunctions such as insulation aging, permanent magnet demagnetization, and mechanical deformation. Furthermore, malfunctions in the mechanical components can generate abnormal noise during operation, and the vibration frequency of the components can change. Therefore, collecting vibration and noise data can improve the accuracy of wind turbine fault diagnosis and significantly reduce unplanned downtime losses caused by manual inspections.

[0153] Among them, gearbox failures mainly include pitting and tooth breakage caused by poor lubrication or overload; generator failures mainly include generator short circuits caused by insulation aging, reduced power generation efficiency caused by high temperature, and demagnetization of permanent magnets; control system failures include bearing jamming caused by excessive bearing temperature and shortened capacitor lifespan caused by prolonged use of capacitors.

[0154] Optionally, when the remote monitoring system detects abnormalities in certain electrical parameters of the wind turbine's operating data, such as sudden voltage drops / surges, current overloads, large power fluctuations, and decreased insulation resistance, it will use drones and sensors to collect data on vulnerable and critical components of the wind turbine, obtaining component structure images and sensor data.

[0155] In some embodiments, the external feature acquisition module 202 specifically comprises:

[0156] First, acquire historical normal structural images of the component to be inspected from the previous time period. The previous time period refers to the period during the last wind turbine inspection. These historical normal structural images are images of the component structure in a normal state taken during the last inspection. Then, compare the component structure images with historical normal structural images taken from the same angle, specifically comparing aspects such as the presence of cracks, surface textures, foreign objects, whether the installation position has shifted, and the wear condition of the component, to obtain detailed comparison results.

[0157] Based on the comparison results, abnormal areas that differ significantly from historical images in the component structure images are labeled and cropped. These cropped abnormal areas are then used as images to be identified for further identification of the component's fault type.

[0158] Based on a pre-defined component anomaly type library, the images to be identified are matched against data in the database to predict one or more potential fault types for each image. This component anomaly type library is built upon historical fault records of wind turbines, covering most common fault conditions for wind turbine components and storing a large number of images showing structural damage to each component to achieve accurate data matching.

[0159] Specifically, each image to be identified is labeled with the location of the wind turbine component from which it was taken. Based on the labeled location of the component, the corresponding image is found in the component anomaly type database and matched. When the confidence level of the matching result reaches the set threshold, it means that a fault image that is very similar to the image to be identified has been found. Based on the fault information of the fault image, the potential fault type of the image to be identified is determined.

[0160] For example, for an image to be identified about the leading edge of a blade, a small area of ​​the blade is a different color from the surrounding area. After searching the component anomaly type database, it is determined that there are multiple potential fault types in the image to be identified, including blade icing, coating peeling, or pitting.

[0161] If an image to be identified has only one potential fault type, then that potential fault type can be directly taken as the fault type of the component to be detected. However, for cases with multiple potential fault types, it is necessary to obtain the weight increase or decrease data of the component to be detected and to perform more detailed feature extraction on the image to be identified in order to help determine the most likely potential fault type.

[0162] The weight change data can be used to prove whether there are foreign objects or material detachment on the component under test. If the weight of the component does not change significantly, for example, if the component is deformed, then it is necessary to extract visual features related to the surface texture and color distribution of the component under test from the image to be identified, in order to further determine whether there are abnormalities such as cracks, corrosion or failure of composite material adhesion on the component under test.

[0163] By combining data on weight increase / decrease, surface texture, and color distribution, the confidence level of each potential fault type is calculated, and the potential fault type with the highest confidence level is selected as the fault type of the component to be tested.

[0164] Furthermore, for components such as towers, it is also possible to detect whether the components are vibrating or tilting. By using surrounding reference objects in the image, it can be determined whether the tilt of the component has worsened compared to the previous period, thereby determining whether the component has malfunctioned.

[0165] In some embodiments, the internal feature acquisition module 203 specifically comprises:

[0166] First, as a control group, the noise and vibration characteristics of these components under normal operating conditions are obtained. Then, the corresponding abnormal noise and vibration signals are extracted from the noise and vibration data collected in the current time period. The normal data and abnormal data are compared to roughly identify whether there are any abnormalities inside the components under test in the current time period.

[0167] The noise generated by wind turbine components during operation includes mechanical noise, aerodynamic noise, electromagnetic noise, and structural noise. Mechanical noise includes gearbox noise, generator noise, and hydraulic system noise, which are generated by gear meshing, cooling fan rotation, and fluid pulsation caused by the operation of hydraulic pumps and valves, respectively. These are the main sources of noise during wind turbine operation. Aerodynamic noise is mainly generated by airflow passing over the blades. For example, the hissing noise generated by turbulence when airflow passes over the blade surface is typical. The spectrum of aerodynamic noise is usually low to mid-frequency with obvious noise peaks. Electromagnetic noise is caused by unbalanced electromagnetic forces causing electromagnetic vibrations in the generator, which are then radiated through solid structures. Structural noise is generated by the vibration of structures such as the nacelle and tower under power transmission and external excitation, which is then propagated through the structure.

[0168] Factors affecting wind turbine noise include wind speed, blade structure design, and operating conditions. For example, mechanical noise is more pronounced when components are old or poorly maintained. Gears may produce loud metallic clanging sounds due to poor meshing, tooth surface peeling, or broken teeth. Similarly, hydraulic systems may generate bubbling noise in the pipelines due to air in the oil or leaks. Therefore, analyzing collected noise data can pinpoint the noise source, providing support for subsequent fault location. Furthermore, combining data on vibration, temperature, and other factors can help determine the cause of component failures.

[0169] During operation, wind turbine components generate complex vibration signals due to mechanical motion, airflow disturbances, and other factors. These signals can be used for fault diagnosis of wind turbine components. For example, when a gearbox tooth breaks, the amplitude of the vibration signal generated by gear meshing increases suddenly, and the sidebands broaden. When the generator rotor is eccentric, the amplitude of its vibration signal changes significantly, possibly accompanied by electrical frequency modulation.

[0170] Based on the above description, noise and vibration data are compared with the noise and vibration characteristics of the component under normal operating conditions in terms of signal amplitude, frequency, and bandwidth. Signal data that are significantly different from normal characteristics are extracted as abnormal noise and abnormal vibration signals.

[0171] Since both vibration and noise signals are related to mechanical vibration and aerodynamics, and there is a certain coupling relationship between them but they are also different, the accuracy and reliability of fault detection can be improved by combining abnormal noise signals and abnormal vibration signals for analysis, so as to obtain accurate internal abnormal characteristics of the component under test.

[0172] In this embodiment, the abnormal noise signal is first denoised to remove environmental interference, the trend term of the abnormal vibration signal is eliminated, and the effective frequency band is extracted, thus completing signal preprocessing. Then, based on the coupling relationship between the two, the cross-correlation coefficient and common peak value of the preprocessed abnormal noise signal and abnormal vibration signal are calculated.

[0173] For example, when bearing damage occurs, noise impact and vibration peaks will appear simultaneously. At this time, the noise signal and vibration signal generated by the bearing will have a certain correlation, and the cross-correlation coefficient of the signals can be calculated accordingly. When a gear tooth breaks, the vibration signal and noise signal generated during gear meshing will have a peak at the same point in time, and this peak can be used as the common peak value.

[0174] Abnormal noise and vibration signals are decomposed into wavelets of different frequency bands by wavelet transform, and the correlation of the signals in the time and frequency domains is analyzed to obtain correlation features.

[0175] In the time domain, different groups of wavelets are translated or scaled to capture the local features of the signal on the time axis. By calculating the energy of the wavelet coefficients within each time window, the local intensity of the signal is reflected, and the local features of the signal on the time axis are acquired. Because the impact caused by component failure will appear in the wavelet coefficients of vibration and noise signals at specific frequency bands, the signal is decomposed into multiple scales to obtain wavelet coefficients from low frequency to high frequency bands. From these, time-domain abrupt change points are found, and the correlation between abnormal noise signals and abnormal vibration signals in the time domain is determined more precisely.

[0176] In the frequency domain, time-frequency analysis is performed by calculating the energy proportion of each wavelet group. The frequency band with concentrated energy in each wavelet group is identified as the dominant frequency band. The correlation between abnormal noise signals and abnormal vibration signals in the frequency domain is determined by the dominant frequency band.

[0177] Then, by combining the correlation in the time domain and the correlation in the frequency domain, correlation features are constructed.

[0178] Optionally, the number of wavelet transform decomposition layers can be determined based on the highest frequency and sampling rate of the abnormal noise signal and abnormal vibration signal.

[0179] By analyzing the correlation of signals in the time domain, transient events can be located, and the exact time period of the fault can be pinpointed, providing support for subsequent fault cause identification by combining meteorological and operational condition changes. By analyzing the correlation of signals in the frequency domain, periodic impacts can be detected, providing a basis for determining the fault type.

[0180] By combining the cross-correlation coefficient, the common peak value, and the correlation features, a mapping relationship between abnormal noise signals and abnormal vibration signals is constructed. Then, the abnormal noise signals and abnormal vibration signals are fused using this mapping relationship to obtain abnormal change characteristics.

[0181] Then, aiming for the highest similarity to the abnormal change features, an internal structural change simulation is performed based on the normal structural dimensions of the component under test in the previous time period. Simultaneously, during the internal structural change process, the similarity between the noise and vibration signals emitted by the simulation results and the abnormal change features is calculated. The simulation stops when the similarity reaches its highest value, completing the simulation of the actual changes in the component under test. The simulation result corresponding to the highest value is compared with the normal structural dimensions, and the corresponding internal abnormal features are obtained by acquiring the internal structural changes of the component under test.

[0182] Considering that each component under test has different material type and structural shape, and that the connection structure with other connected components will also generate different degrees of friction, the characteristics of the component under test are also evaluated based on temperature and the connection structure, so as to more accurately simulate the changes in the internal structure of the component under test under environmental influence.

[0183] Temperature changes have significantly different effects on components made of different materials, primarily depending on the material's physical properties (such as coefficient of thermal expansion, thermal conductivity, and temperature limit). The material types of the components to be tested in a wind turbine are mainly categorized as metals (such as gearbox housings and towers), composite materials (such as blades), polymer materials (such as component sealing layers and cable insulation layers), ceramic and permanent magnet materials (such as permanent magnets and ceramic insulators), lubricating materials (such as grease), and electronic materials (such as circuit boards). This application's embodiments simulate the degree of temperature influence on each material type to further obtain the different responses of each component to temperature changes, thus obtaining the first structural characteristics of each component related to temperature.

[0184] For example, both are gearbox housings, but within the same temperature range, aluminum alloy products are more prone to deformation than iron products. Therefore, temperature has a greater impact on aluminum alloy gearbox housings, and the internal structural deformation of aluminum alloy gearbox housings is more pronounced.

[0185] Besides temperature, differences in the connection structure or method between the component under test and other components can also cause varying degrees of damage to the component under test. For example, even with bearing connections, sliding bearings generate less friction than rolling bearings; similarly, with gear meshing, helical gears, due to their larger contact area, may generate higher friction than spur gears. Considering that friction generated by the same material type can also lead to different degrees of deformation for each component due to differences in the connection structure or method, this application embodiment further simulates the degree of damage to each component under test caused by friction during operation based on the connection structure or method, thereby obtaining a second structural characteristic for each component under test.

[0186] Combining the first structural feature and the second structural characteristic, based on the normal structural dimensions of the component under test in the previous period and the operating conditions of the wind turbine in the current period, the internal structural changes of the component under test are continuously simulated to obtain the internal abnormal features of the component under test.

[0187] In some embodiments, the fault cause analysis module 204 specifically comprises:

[0188] Because external and internal anomalies contain a large amount of fault information about the components to be tested, in order to further determine the cause of the fault, assist maintenance personnel in quickly carrying out repairs and implementing better protective measures, it is also necessary to combine the current meteorological changes and wind turbine operating data to deduce the cause of the fault and obtain a comprehensive and accurate fault diagnosis result.

[0189] Data such as the service life and operating condition changes of each component to be tested are extracted from the wind turbine operation data. Then, based on external and internal abnormal characteristics, corresponding damage association labels are constructed for the above data and meteorological changes, so as to associate the abnormal conditions of the components with the corresponding failure causes.

[0190] Specifically, based on the historical inspection records of the wind turbine, the correlation between the service life of components and the frequency of component damage is analyzed, the relationship between the current service life of components and the degree of damage impact is quantified, and the influence characteristics of service life are constructed. For example, as the operating time increases, blades are more prone to cracking due to accumulated mechanical fatigue. When the service life of the blades exceeds 10 years, the influence characteristics of service life are more obvious than when they were first put into use, and there are more associated damage tags.

[0191] Based on the sampling frequency and component load variation curves, the degree of damage impact of the operating condition changes on the component under test is quantified. This is because the component load variation curve of the wind turbine is closely related to the turbine's operating conditions and can reflect whether the turbine is operating under load within a certain period. When the component load variation curve changes significantly, it indicates that the component is likely to experience fatigue damage; when the component load variation curve shows extreme conditions, it indicates that the component is likely to suffer a reduction in structural safety due to extreme operating conditions. The sampling frequency affects the accuracy of the impact characteristics of the operating condition change data; the higher the sampling frequency, the more accurate the impact characteristics obtained.

[0192] Then, based on the long-term and short-term characteristics of meteorological changes, the impact characteristics of the meteorological changes are constructed. This is because meteorological changes affect the switching of wind turbine operating modes. For example, in high wind speeds, to avoid damage to blades, towers, and other structures due to overload, blade operation is suspended through feathering, mechanical braking, and yaw to avoid wind; in sandstorm weather, derated operation is required to protect the motor and frequency converter, etc.

[0193] Furthermore, to improve the accuracy of the impact characteristics of meteorological changes, this embodiment of the application further divides the impact of meteorological changes into long-term and short-term characteristics. Short-term characteristics include wind speed turbulence intensity, temperature gradient, and lightning density, while long-term characteristics include the average number of freezing rain days per year and the frequency of sandstorms. By using long-term and short-term characteristics, the constructed impact characteristics of meteorological changes can reflect the impact of extreme weather and seasonal changes on the component under test.

[0194] In addition, to further improve data accuracy, this application embodiment also verifies meteorological data according to timestamps and wind turbine latitude and longitude to reduce data latency.

[0195] Finally, based on the external and internal abnormal features of the component to be tested, the damage association label for each influencing feature is obtained by searching for the fault events caused by each influencing feature in the existing historical fault records of the wind turbine.

[0196] Then, based on the existing external and internal abnormal features containing the fault conditions of the component to be tested, a causal graph related to each damage association label and fault cause is constructed, which facilitates the reverse deduction of the fault cause of the component to be tested.

[0197] First, it's necessary to clarify the causal relationship. Based on historical fault records, external and internal anomalies are used as outcome variables, wind turbine operating data as intermediate response variables, and damage association labels as potential causal variables. Wind turbine operating data, such as speed, power, current, and voltage, are used as intermediate response variables because they are results of the fault. Damage association labels, as factors that may cause the fault in the component under test, are used as potential causal variables, forming the basis for subsequent fault cause screening.

[0198] Then, through conditional independence testing, based on the relationships that the potential causal variables cause changes in the intermediate response variables and that the intermediate response variables are representations of the outcome variables, a causal graph skeleton is constructed between the damage association labels and external and internal anomalous features.

[0199] For example, a conditional independent test is used to examine whether changes in wind speed can significantly predict changes in blade vibration. A pre-defined model is used to fit the effect of wind speed on vibration while controlling for confounding variables such as temperature, thus constructing a directed path between wind speed and vibration. To improve the accuracy of the test results, it is also necessary to eliminate interference from other variables during the test.

[0200] Optionally, the Fisher-Z test method is used in the embodiments of this application.

[0201] Simultaneously, based on the causal effect of the latent causal variable on the intermediate response variable, an influence weight is generated for each damage association label. The causal effect describes the degree of influence of the latent causal variable on the intermediate response variable; for example, for every 1 m / s increase in wind speed, the RMS value of vibration will increase by 0.2 g. The influence weight characterizes the probability that the damage association label is a cause of component failure.

[0202] Next, nodes in the cause-effect graph skeleton are labeled according to the influence weights. Then, using the outcome variable as the endpoint, a causal forest is constructed on the labeled cause-effect graph skeleton in a way that distinguishes between direct and indirect causes. This results in a causal graph of fault causes related to external and internal anomalies. This causal graph comprehensively displays the various fault causes of the component under test and the coupling relationships between these causes. The distinction between direct and indirect causes is based on the influence weights.

[0203] For example, if a generator experiences leakage, the direct cause is insulation aging, and the indirect cause is insulation failure due to long-term high-temperature operation. Therefore, the cause-effect diagram will distinguish the direct and indirect causes of the generator failure based on the length of the path, which will facilitate the generation of more comprehensive fault diagnosis results.

[0204] Finally, the cause-effect graph is analyzed using the path verification method to determine the direct and indirect causes of the fault in the component under test. By identifying external and internal abnormal features, the location and type of the fault are determined, resulting in accurate and comprehensive fault diagnosis. This provides support for timely implementation of appropriate maintenance measures and can further improve the operational stability of the power system.

[0205] Implementing the embodiments of this application has the following beneficial effects:

[0206] This application embodiment detects abnormalities in the external and internal structures of components under test within a wind turbine. By comparing captured images of the component's structure with historical images of normal structures, it determines whether the external structure of the component has changed, thus obtaining external anomaly characteristics. Then, based on the vibrations and noise emitted by the component during operation, it judges whether internal structural changes have occurred. Furthermore, considering that different material types of the component react differently to temperature changes, the degree of internal structural change is quantified. This allows for the identification of internal structural anomalies in wind turbine components solely through sensor data, without interrupting wind turbine operation, saving significant inspection costs and improving inspection efficiency, while obtaining accurate internal anomaly characteristics. Finally, the impact of meteorological changes and wind turbine operating data on damage to the component under test is analyzed to determine the causes of failure corresponding to the external and internal anomaly characteristics, resulting in a comprehensive and accurate fault diagnosis. This provides support for subsequent maintenance personnel to implement appropriate maintenance measures.

[0207] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of this application. It should be understood that the above descriptions are merely specific embodiments of this application and are not intended to limit the scope of protection of this application. In particular, it should be noted that any modifications, equivalent substitutions, or improvements made by those skilled in the art within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method of monitoring the dynamic response of a large scale fan, characterized by, The method comprises the following steps: obtaining a component structure image and sensor data of a component to be detected of a fan; wherein the sensor data comprises temperature change data, vibration data and noise data; detecting external structural changes of the component to be detected according to the component structure image and historical normal structure images, and obtaining external abnormal features of the component to be detected; simulating internal structural changes of the component to be detected according to the temperature change data, material types and structural dimensions of the component to be detected, and obtaining internal abnormal features of the component to be detected through the noise data and the vibration data, specifically: extracting abnormal noise signals and abnormal vibration signals from the noise data and the vibration data according to noise characteristics and vibration characteristics of the component to be detected under normal working conditions; aligning and fusing the abnormal noise signals and the abnormal vibration signals to obtain abnormal change features; taking the highest similarity with the abnormal change features as the target, simulating internal structural changes of the component to be detected based on normal structural dimensions of the component to be detected in a previous period through the temperature change data and the material types to obtain internal abnormal features of the component to be detected, specifically: simulating the influence degree of temperature on each material type according to the temperature change data to obtain first structural characteristics of each component to be detected; simulating the damage degree of the component to be detected due to friction of the material types in the running process according to the connection structure between the component to be detected and other components to obtain second structural characteristics of each component to be detected; continuously simulating internal structural changes of the component to be detected based on the normal structural dimensions through the first structural characteristics and the second structural characteristics, and calculating the similarity between noise signals and vibration signals generated by the component to be detected under each internal structural change and the abnormal change features; stopping the simulation operation when the similarity is the highest, and comparing the current simulation result with the normal structural dimensions to construct the internal abnormal features; analyzing the fault causes of the component to be detected based on the external abnormal features and the internal abnormal features, combining weather changes and fan operation data in a current period to obtain corresponding fault diagnosis results; the method of detecting external structural changes of the component to be detected according to the component structure image and the historical normal structure image to obtain external abnormal features of the component to be detected, specifically: obtaining the historical normal structure image of the component to be detected in a previous period, comparing the historical normal structure image with the component structure image, and cutting the abnormal area in the component structure image based on the comparison result to obtain a to-be-recognized image; predicting potential fault types of the to-be-recognized image according to a preset component abnormal type library; screening the potential fault types through the to-be-recognized image and collected weight increase and decrease data of the component to be detected to determine the fault type of each component to be detected; According to the fault type and the image to be identified, an external abnormal feature of the component to be detected is constructed; The abnormal noise signal and the abnormal vibration signal are aligned and fused to obtain an abnormal change feature, specifically as follows: The cross-correlation coefficient and the common peak value of the abnormal noise signal and the abnormal vibration signal are calculated; The correlation of the abnormal noise signal and the abnormal vibration signal in the time domain and the frequency domain is analyzed through wavelet transform to obtain a correlation feature; According to the cross-correlation coefficient, the common peak value and the correlation feature, a mapping relationship of the abnormal noise signal and the abnormal vibration signal is constructed, and then data fusion is performed on the abnormal noise signal and the abnormal vibration signal through the mapping relationship to obtain the abnormal change feature.

2. The method of claim 1, wherein The potential fault type is screened through the image to be identified and the weight increase and decrease data of the component to be detected, and the fault type of each component to be detected is determined, specifically as follows: If there is only one potential fault type for one image to be identified, the potential fault type is taken as the fault type of the component to be detected; If there is more than one potential fault type for one image to be identified, the weight increase and decrease data of the component to be detected is obtained, and the surface texture and color distribution of the component to be detected are obtained from the image to be identified; the confidence of each potential fault type is calculated according to the weight increase and decrease data, the surface texture and the color distribution, and the potential fault type with the highest confidence is selected as the fault type of the component to be detected.

3. The method of claim 1, wherein Based on the external abnormal feature and the internal abnormal feature, the weather change and the fan operation data in the current period are combined to analyze the fault cause of the component to be detected, and a corresponding fault diagnosis result is obtained, specifically as follows: The service life and the working condition change data of each component to be detected are extracted from the fan operation data; According to the external abnormal feature and the internal abnormal feature, corresponding damage association labels are constructed for the service life, the working condition change data and the weather change; The external abnormal feature and the internal abnormal feature are taken as result variables, and the fault cause of the component to be detected is causally inferred to obtain the influence weight of each damage association label and a causal diagram related to the fault cause; The direct cause and the indirect cause of the causal diagram are analyzed through the path inspection method and the influence weight to obtain the fault diagnosis result; wherein the fault diagnosis result includes the fault cause, the fault position and the fault type of the component to be detected.

4. The method of claim 3, wherein According to the external abnormal feature and the internal abnormal feature, corresponding damage association labels are constructed for the service life, the working condition change data and the weather change, specifically as follows: Based on historical inspection records, the damage influence degree of the component to be detected by the service life is quantified to construct the influence feature of the service life; Based on the sampling frequency and the component load change curve, the damage influence degree of the component to be detected by the working condition change data is quantified to construct the influence feature of the working condition change data; According to the long-term and short-term characteristics of the meteorological change, an influence feature of the meteorological change is constructed; With the external abnormal feature and the internal abnormal feature as results, a failure cause in a historical failure record of the fan is searched, and the damage correlation label is generated for each influence feature.

5. The method of claim 3, wherein With the external abnormal feature and the internal abnormal feature as result variables, a failure cause of the to-be-detected component is causally inferred, and an influence weight of each damage correlation label and a causal graph related to the failure cause are obtained, specifically as follows: Based on the historical failure record, the external abnormal feature and the internal abnormal feature are taken as result variables, fan operation data are taken as intermediate response variables, and the damage correlation label is taken as a latent cause variable; Through conditional independence testing, a causal graph skeleton is constructed based on a relationship between the latent cause variable and the intermediate response variable and a relationship between the result variable and the intermediate response variable; According to a causal effect of the latent cause variable on the intermediate response variable, an influence weight of each damage correlation label is generated; According to the influence weight, the causal graph skeleton is labeled, and a causal forest of the labeled causal graph skeleton is constructed with the result variable as a terminal point, and the causal graph is obtained.

6. A dynamic response monitoring system for a large scale fan, the system comprising: Comprise: a data acquisition module, an external feature acquisition module, an internal feature acquisition module, and a failure cause analysis module; The data acquisition module is configured to acquire a component structure image and sensor data of a to-be-detected component of a fan; wherein the sensor data comprises temperature change data, vibration data, and noise data; The external feature acquisition module is configured to detect an external structure change of the to-be-detected component according to the component structure image and a historical normal structure image, and obtain an external abnormal feature of the to-be-detected component; The internal feature acquisition module is configured to simulate internal structure changes of the component to be detected based on the temperature change data and the material type and structure size of the component to be detected, and obtain internal abnormal features of the component to be detected based on the noise data and the vibration data. Specifically, the abnormal noise signal and the abnormal vibration signal are extracted from the noise data and the vibration data based on noise characteristics and vibration characteristics of the component to be detected under normal working conditions. The abnormal noise signal and the abnormal vibration signal are aligned and fused to obtain abnormal change features. The internal structure changes of the component to be detected are simulated based on the temperature change data and the material type on the basis of the normal structure size of the component to be detected in the previous period, with the highest similarity to the abnormal change features as the target, to obtain internal abnormal features of the component to be detected. Specifically, the influence of temperature on each material type is simulated based on the temperature change data to obtain first structure characteristics of each component to be detected. The damage degree of the component to be detected due to friction of the material type during operation is simulated based on the connection structure between the component to be detected and other components to obtain second structure characteristics of each component to be detected. The internal structure changes of the component to be detected are continuously simulated based on the normal structure size by using the first structure characteristics and the second structure characteristics, and the similarity between the noise signal and the vibration signal generated by the component to be detected under each internal structure change and the abnormal change features is calculated. When the similarity is the highest, the simulation operation is stopped, and the current simulation result is compared with the normal structure size to construct the internal abnormal features. The fault cause analysis module is configured to analyze the fault cause of the component to be detected based on the external abnormal features and the internal abnormal features, in combination with weather change conditions and fan operation data in the current period, to obtain a corresponding fault diagnosis result. The external structure changes of the component to be detected are detected based on the component structure image and historical normal structure images to obtain external abnormal features of the component to be detected. Specifically, The historical normal structure image of the component to be detected in the previous period is obtained, and the historical normal structure image is compared with the component structure image. Based on the comparison result, an abnormal region in the component structure image is cropped to obtain a to-be-recognized image. A potential fault type of the to-be-recognized image is predicted based on a preset component abnormal type library. The potential fault type is screened based on the to-be-recognized image and collected weight increase and decrease data of the component to be detected, to determine a fault type of each component to be detected. The external abnormal features of the component to be detected are constructed based on the fault type and the to-be-recognized image. The abnormal noise signal and the abnormal vibration signal are aligned and fused to obtain abnormal change features. Specifically, Cross-correlation coefficients and common peak values of the abnormal noise signal and the abnormal vibration signal are calculated. The correlation characteristics are obtained by analyzing the correlation of the abnormal noise signal and the abnormal vibration signal in time domain and frequency domain through wavelet transform; According to the cross-correlation coefficient, the common peak value and the correlation characteristics, a mapping relationship of the abnormal noise signal and the abnormal vibration signal is constructed, and then data fusion is performed on the abnormal noise signal and the abnormal vibration signal through the mapping relationship to obtain abnormal change characteristics.

Images