Fan blade material fatigue degree online evaluation system based on deep voiceprint

Abstract

The application relates to the technical field of online monitoring of dynamic characteristics of large rotating machinery structures, and discloses a fan blade material fatigue degree online evaluation system based on deep voiceprints, which comprises the following steps: active power and structural vibration signals of a generator are synchronously collected by using unit self-operation data flow; before operation, data segments are checked according to SCADA event logs to eliminate power grid disturbance working conditions; and then, an aeroelastic transfer function representing a dynamic transfer relationship and a coherence function representing a vibration source are generated. The application utilizes two signals generated by unit self-operation, establishes a stable dynamic characteristic observation mode in an environment where a random excitation source cannot be measured by constructing a transfer function, makes long-term trend monitoring of the overall stiffness of the blade, which is a core structural parameter, independent of the dependence on an external known excitation source, and thus realizes evaluation of the structural dynamic balance state under continuous operation conditions.

Description

Technical Field

[0001] This invention relates to an online fatigue assessment system for wind turbine blade materials based on deep acoustic signatures, belonging to the field of online monitoring technology for dynamic characteristics of large rotating machinery structures. Background Technology

[0002] Currently, the establishment of its testing and analysis methods usually relies on the fact that the excitation and response of the structure are knowable or separable. By applying a known excitation to the system and measuring its response, technicians can infer the dynamic characteristic parameters of the system. This testing principle is effective in controlled environments.

[0003] However, when this principle is applied to large-scale wind turbine generators in operation, the special nature of their operating environment makes it difficult to meet the aforementioned prerequisites. Wind, as the excitation source that drives the generator and generates dynamic loads on it, exhibits random and unmeasurable characteristics in terms of intensity direction and frequency components. What technicians can reliably obtain using sensors is only the vibration response of structures such as blades and towers under this disordered load. This unknown nature of the excitation source introduces an inherent technical challenge in structural condition assessment: when the monitoring system records an increase in the vibration amplitude of the blades, it is difficult to effectively determine whether the increase is a normal operating condition fluctuation caused by enhanced wind load excitation or due to a decrease in the stiffness of the blade structure (i.e., a manifestation of material fatigue), making it more sensitive to load response. This coupling of excitation change and structural response sensitivity change in the vibration signal makes it difficult to determine the changes in the structural characteristics by analyzing only a single vibration signal.

[0004] To address this technical challenge, the industry has employed methods such as installing anemometers or establishing aerodynamic models to approximate the excitation source. However, these methods often face limitations in large-scale application in engineering practice due to insufficient representativeness of measurement points, model distortion, or high deployment costs. Specifically, existing technologies suffer from the following shortcomings: 1. The causal relationship between vibration response and load excitation is unclear; changes in a single vibration signal cannot be directly attributed to changes in the structure's inherent characteristics, affecting the reliability of fatigue assessment conclusions; 2. Technical solutions that directly measure or model random wind loads face high technical complexity and engineering implementation costs, limiting their widespread application; 3. Existing monitoring methods lack a direct... In the context of online observation that stably reflects changes in the intrinsic physical property of blade material stiffness, other monitoring approaches have emerged. For example, Chinese invention patent CN119470670B discloses an online monitoring method for offshore wind turbine blades based on acoustic signature recognition. This method attempts to assess blade cracks by collecting acoustic signature signals during wind turbine operation and combining them with data such as weather and power generation using a deep learning model. However, acoustic signals themselves are complex structural responses, and their characteristics are also significantly affected by random wind loads. Such methods still face inherent technical challenges in fundamentally decoupling excitation changes from structural property degradation, making it difficult to explicitly attribute signal changes to a decrease in blade stiffness. Therefore, the technical problem this invention aims to solve is how to process the unit's own operating signals to distinguish the respective effects of load fluctuations and changes in structural response sensitivity on vibration signals without requiring precise measurement of random wind excitation, thereby establishing an online assessment method that can accurately reflect the fatigue state of blade materials. Summary of the Invention

[0005] This invention provides an online fatigue assessment system for wind turbine blade materials based on depth acoustic signatures. Its main purpose is to solve the problem of how to distinguish the respective effects of load fluctuations and structural response sensitivity changes on vibration signals without the need for precise measurement of random wind excitation, so as to establish an online assessment method that can reflect the fatigue state of blade materials.

[0006] To achieve the above objectives, the present invention provides an online fatigue assessment system for wind turbine blade materials based on depth acoustic signatures, comprising:

[0007] A data acquisition module is configured to synchronously acquire, from the standard operating data stream of the wind turbine, the time series of the generator active power signal, the time series of the structural vibration signal, and the SCADA event log that records the unit's operating events;

[0008] A data validity arbitration module is configured to check the synchronously acquired data segments of active power signal and structural vibration signal before performing calculations, based on whether there are preset event codes in the SCADA event log that represent power grid disturbances, and only determine the data segments without event codes as valid data segments.

[0009] An edge computing module is configured to calculate the cross-spectral density between the active power signal and the structural vibration signal and the autospectral density of the active power signal only for valid data segments. Based on the calculation results, an aeroelastic transfer function characterizing the dynamic transmission relationship between the two is generated, as well as a coherence function characterizing the proportion of energy in the structural vibration signal caused by the linearity of the active power signal.

[0010] A cloud-based analysis module is configured to compare the aeroelastic transfer function of the current cycle with a preset baseline transfer function, and only when the value of the coherence function in the preset frequency band continuously meets a first threshold condition will the trend change representing the decrease in blade stiffness generated by the comparison be included in the fatigue assessment result.

[0011] Preferably, the edge computing module is configured to generate the aeroelastic transfer function through the following operations. , ,in, For frequency, This represents the cross-spectral density of the active power signal and the structural vibration signal. This represents the autospectral density of the active power signal.

[0012] Preferably, the cloud analysis module is also configured to compare the coherence function of the current period with the preset baseline coherence function, and when a decrease in the coherence function below the second threshold is detected within the preset frequency band, it is determined that there is a mechanical fault vibration source unrelated to aerodynamic load, and the blade stiffness decrease is distinguished accordingly.

[0013] Preferably, the cloud analysis module is further configured to activate a phase analysis diagnostic mode when it outputs an assessment result indicating a decrease in blade stiffness; in this mode, the data acquisition module is further configured to acquire an azimuth signal characterizing the impeller rotation phase; and the edge computing module is further configured to perform time-synchronous averaging processing on the structural vibration signal based on the azimuth signal to enhance the vibration component related to the rotation position of a specific blade, and locate the target blade that generates abnormal vibration based on the phase information of the enhanced vibration component.

[0014] Preferably, the time-synchronized averaging process includes: dividing the continuous structural vibration signal into multiple data segments with one revolution of the impeller as the cycle, and averaging all data segments point by point.

[0015] Preferably, the cloud analysis module is further configured to: perform time-scale characteristic analysis on the trend changes of the aeroelastic transfer function over time; the analysis includes calculating the rate of change of the trend change within a short time-scale window of hours; and calculating the monotonic increment of the trend change within a long time-scale window of weeks; and determining the physical cause of the trend change based on the characteristic differences between the rate of change and the monotonic increment.

[0016] Preferably, the data validity arbitration module has preset event codes that characterize power grid disturbances, including event codes for generator grid voltage drops or frequency anomalies; the arbitration module is configured to mark the synchronously acquired data segments within the corresponding time range as invalid and discard them when such codes are retrieved in the SCADA event log.

[0017] Preferably, the structural vibration signal originates from an acceleration sensor deployed on the wind turbine nacelle or tower.

[0018] Preferably, the baseline transfer function is established by statistically averaging multiple aeroelastic transfer functions calculated during the initial operation phase of the wind turbine.

[0019] Preferably, the data acquisition module is configured to perform synchronous acquisition at a frequency of not less than 10Hz; and the edge computing module is configured to process continuous valid data segments with an overlap rate of not less than 50%.

[0020] Compared with the prior art, the beneficial effects of the present invention are:

[0021] 1. By synchronously acquiring generator active power signals and unit structural vibration signals, and calculating the transfer function between the two, a stable dynamic characteristic observation method was established in the test environment of wind turbines, where the excitation source is unmeasurable and randomly changing. Since the high-frequency fluctuations of power generation and the fluctuations of aerodynamic loads acting on the blades share the same physical source, namely wind turbulence, the transfer function directly reflects the degree of structural vibration response caused by unit aerodynamic load fluctuations. This method utilizes the two signals that the unit itself inevitably generates during operation, enabling long-term trend monitoring of the core structural parameter of overall blade stiffness to be independent of external known excitation sources, making it possible to effectively evaluate the dynamic equilibrium state under real and continuous operating conditions.

[0022] 2. While calculating the transfer function, the coherence function between the power generation signal and the structural vibration signal is calculated in parallel. Both functions originate from the same set of input signals and the same calculation process, but provide physically orthogonal diagnostic information. The amplitude trend of the transfer function is used to characterize the change in the structure's sensitivity to aerodynamic excitation, while the coherence function is used to determine the proportion of aerodynamic excitation in the energy composition of the vibration response. When a mechanical fault vibration source, such as that generated by the transmission chain, appears in the system, this vibration source is incoherent with the aerodynamic excitation, which will inevitably lead to a decrease in the value of the coherence function in the corresponding frequency band. Therefore, by synchronously comparing the changes in the transfer function with the state of the coherence function, the system can clearly distinguish between the two completely different state evolutions: the intensification of the structure's response to aerodynamic loads and the intervention of new non-aerodynamic mechanical vibration sources, thus avoiding confusion about the source of the fault.

[0023] 3. After confirming a trend of decreasing overall unit stiffness through comparison of the aforementioned transfer function and coherence function, the azimuth signal of the impeller rotation is further introduced, and the structural vibration signal is processed by time-synchronous averaging. This processing step is triggered by the aforementioned stiffness decrease assessment result as a prerequisite. Time-synchronous averaging utilizes the rotation periodicity to enhance the deterministic components in the vibration signal that are strongly correlated with the blade rotation position, while weakening the non-periodic vibration components caused by random wind loads. The azimuth angle corresponding to the peak value of the enhanced vibration waveform directly points to the blade rotation position that contributes the most to the abnormal vibration. This deepens a global assessment of the overall system state into a component-level problem location with a clear physical location, constructing a complete diagnostic path from condition monitoring to fault location.

[0024] 4. When conducting long-term trend analysis on the transfer function fingerprint, the time-scale characteristics of its change process are also analyzed. The decrease in structural stiffness caused by material fatigue is a slow and cumulative change over a long period; while the changes in system dynamic characteristics caused by transient events such as blade icing are rapid changes in a short period of time and are reversible. By performing long-scale trend analysis and short-scale rate of change analysis on the time series of the transfer function fingerprint drift index in the cloud analysis module, and making logical decisions based on the differences between the two, the system can separate the permanent trend caused by the evolution of material properties from the temporary disturbances caused by external environmental factors. This mechanism ensures the purity of long-term data and the validity of conclusions in evaluating the core indicator of blade material fatigue. Attached Figure Description

[0025] Figure 1 This is a diagram showing the overall system architecture and online evaluation data flow of the present invention;

[0026] Figure 2This is a comparison of the time-domain characteristics of the AETF drift index of the material fatigue and blade icing in this invention.

[0027] Figure 3 This is a flowchart of the multi-state collaborative diagnosis and decision-making process of the cloud module of the present invention. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] An online fatigue assessment system for wind turbine blade materials based on deep acoustic signatures is applied to the online monitoring of the structural health status of in-service large wind turbine generators. The system architecture includes a data acquisition module, a data validity arbitration module, an edge computing module, and a cloud analysis module. The data acquisition module connects to the wind turbine generator's Supervisory Control and Data Acquisition (SCADA) system and condition monitoring sensors, acquiring the time series of the generator's active power signal at a synchronous acquisition frequency of no less than 10Hz. Time series of structural vibration signals originating from accelerometers in the nacelle or tower The system also includes SCADA event logs that record unit operation events. Since generator power output signals are affected by grid-side disturbances, to ensure the validity of the data used in subsequent analysis, a data validity arbitration module performs checks at the front end of the data processing flow. This module internally sets up a blacklist of event codes related to grid disturbances, including event codes for generator grid voltage drops or frequency anomalies. Before processing any segment of synchronously acquired active power signal and structural vibration signal data, the arbitration module searches the SCADA event logs within the corresponding timestamp range of that data segment. If an event code from the blacklist exists in the log, the data segment is deemed invalid and discarded. Only when no such event code exists is the data segment deemed valid and passed to the edge computing module.

[0030] After receiving a valid data segment, the edge computing module establishes a dynamic characteristic observation method. This module utilizes the correlation between the high-frequency fluctuations of power generation and the aerodynamic load fluctuations acting on the blades, which share the same physical source (i.e., wind turbulence), and uses the active power signal as a proxy for the aerodynamic load. The module's signal processing procedure is as follows: continuous valid data segments are framed with an overlap rate of no less than 50%; the active power signal time series within each frame is processed... Time series of structural vibration signals Perform a Fourier transform to obtain its spectrum. and ; Calculate the cross-spectral density between the two. Autospectral density of active power signal Finally, the aeroelastic transfer function characterizing the dynamic transmission relationship between the two is generated through the following calculations. , In the formula, For frequency; The cross-spectral density of the active power signal and the structural vibration signal; The autospectral density of the active power signal, amplitude Corresponding to the magnitude of the structural vibration caused by unit power fluctuation, this amplitude is directly proportional to the structural flexibility, i.e., inversely proportional to the overall effective stiffness of the blade. Simultaneously, to distinguish different vibration sources, the edge computing module, while calculating the transfer function, simultaneously calculates a coherence function characterizing the proportion of energy linearly caused by the active power signal in the structural vibration signal. The function value ranges from 0 to 1. When the value is close to 1, it indicates that the structural vibration at this frequency is mainly caused by active power fluctuations, and the aeroelastic transfer function... With coherence function It was also uploaded to the cloud analysis module.

[0031] The cloud-based analytics module performs long-term trend analysis and fatigue assessment. This module first executes a baseline calibration procedure: during the initial operation phase of the unit, it collects and statistically averages multiple aeroelastic transfer functions to establish a baseline transfer function. Similarly, a baseline coherence function is established. During long-term monitoring, the module will use the aeroelastic transfer function of the current period. With baseline transfer function Comparison is only possible if the coherence function The transfer function amplitude will only be increased when the value within the preset frequency band continuously meets the first threshold condition (e.g., continuously greater than 0.8). The monotonically increasing trend of the coherence function is included in the fatigue assessment results. When a decrease below the second threshold is detected in the coherence function within the preset frequency band, it is determined that there is a mechanical fault vibration source unrelated to aerodynamic load, and this is used to distinguish the decrease in blade stiffness, thereby avoiding confusion of fault sources. To ensure the universality of this technical solution for deployment on different units, the system needs to perform a standardized engineering calibration procedure to determine key calculation parameters during the initial operation. This procedure first identifies the center frequency of the first-order flapping mode of the blade based on the unit design data or by modal analysis of the vibration signals during the initial operation phase. And based on this, a width is set as The analysis frequency band was then used; subsequently, during a baseline data acquisition period covering at least 150 hours of effective operating conditions, the coherence function within the frequency band was statistically analyzed. The numerical distribution was used, and the 5th percentile was taken as the first threshold for judging the validity of the transfer function analysis. Simultaneously, a threshold lower than the baseline period was set. A fixed value of 25% of the mean is used as the second threshold to distinguish new sources of mechanical vibration; the warning threshold for the aeroelastic transfer function drift index is calculated by taking the mean of the index sample during the baseline period. with standard deviation Set its warning line as .

[0032] When the cloud-based analysis module confirms a trend of decreasing overall unit stiffness, a phase analysis diagnostic mode can be activated. In this mode, the data acquisition module is configured to also acquire an azimuth signal characterizing the impeller rotation phase. The edge computing module uses this azimuth signal as a reference to analyze the structural vibration signal. The process involves performing time-synchronized averaging, which includes dividing the continuous structural vibration signal into multiple data segments with one revolution of the impeller as the cycle, and averaging all data segments point by point. This process enhances deterministic vibration components related to the blade rotation position and weakens aperiodic vibration components. The azimuth angle corresponding to the peak value of the enhanced vibration waveform points to the blade rotation position that contributes the most to the abnormal vibration, thereby enabling diagnosis from condition monitoring to fault location. In addition, to distinguish between changes in dynamic characteristics caused by transient events such as blade icing and material fatigue, the cloud analysis module also analyzes the trend changes of the aeroelastic transfer function over time. The system performs time-scale characteristic analysis, which includes calculating the rate of change of trend changes within a short time-scale window of hours and the monotonic increment of trend changes within a long time-scale window of weeks. The stiffness decrease caused by material fatigue is a long-period, slow and cumulative change, while blade icing is a rapid change in a short period of time and is reversible. By making logical decisions based on the characteristic differences between the rate of change and the monotonic increment, the system can distinguish between permanent trends caused by the evolution of material properties and temporary disturbances caused by external environmental factors, so as to ensure the long-term validity of fatigue assessment results.

[0033] Example 1: On a continuously operating wind turbine generator, its monitoring system recorded a slow increasing trend in the amplitude of the nacelle vibration signal over a period of several months. This operating condition requires the maintenance team to distinguish between changes in excitation and changes in system characteristics. To assess the dynamic equilibrium state of the generator's structure, the aforementioned specific implementation method's assessment system was deployed. This system utilizes the generator's own SCADA data stream to synchronously acquire the generator's active power signal. Structural vibration signals output by the nacelle acceleration sensor During the six-month continuous monitoring period, the system's data validity arbitration module, based on synchronously acquired SCADA event logs, identified and eliminated three invalid data segments where the active power signal fluctuated drastically due to grid voltage dips. This prevented non-aerodynamic source disturbances from being introduced into subsequent dynamic characteristic analysis. For all data segments deemed valid, the edge computing module continuously calculated the aeroelastic transfer function. With coherence function The results are then uploaded to the cloud analysis module, which will then perform the current calculations. Baseline transfer function established during the initial system deployment Comparison; analysis results show that the coherence function Within the low-frequency range where the main energy is concentrated, its value remains consistently above 0.85; this phenomenon indicates that the structural vibration signal of the engine room and the active power signal of the generator maintain a high linear correlation, providing a prerequisite for the effectiveness of subsequent transfer function analysis and ruling out the possibility of new, independent mechanical fault vibration sources intervening during this period. Under these conditions, the aeroelastic transfer function... amplitude Trend analysis revealed that in the frequency band corresponding to the first-order flapping mode of the blade, Compared to baseline The amplitude of the value showed an irreversible, unidirectional increase.

[0034] Given that the amplitude of the transfer function characterizes the structural vibration response output caused by a unit aerodynamic load input, its unidirectional increase under the premise of stable coherence objectively indicates that the wind turbine structure is more sensitive to the same aerodynamic load, that is, the effective stiffness of the structure has decreased. This assessment conclusion provides a basis for decision-making for the operation and maintenance team, shifting the focus of maintenance work from judging the uncertainty of external wind conditions to confirming the changes in the characteristics of the structure itself. The system further activates the phase analysis diagnostic mode, and by introducing the impeller azimuth angle signal and performing time-synchronous averaging processing on the structural vibration signal, the phase with the largest contribution of abnormal vibration is finally pointed to one of the blades, realizing the location of the faulty component and providing direct guidance for subsequent shutdown and maintenance. This implementation process uses the active power signal as a proxy for the aerodynamic load and performs collaborative analysis of the transfer function and coherence function. Under the condition of not directly measuring the random excitation source, it transforms a causal fuzzy dynamic response analysis problem into a system characteristic identification problem that can be stably calibrated.

[0035] Example 2: To objectively verify the effectiveness of the aforementioned technical solution in distinguishing between changes in structural stiffness and changes in random excitation, this example designed and executed a dynamic characteristic test under a controlled environment. The purpose of the test was to quantitatively compare the performance difference between monitoring only the vibration amplitude and using the aeroelastic transfer function analysis method of this invention in identifying a decrease in structural stiffness, on a platform where aerodynamic loads and structural parameters can be precisely controlled. The test was conducted on a structural dynamic test bench with a wind tunnel. The test object was a scaled-down wind turbine blade, whose root was fixed to a vibration exciter by a set of adjustable preload bolts. The data acquisition system of the test platform included: a device installed in the blade... The system includes a triaxial accelerometer with a sampling frequency of 200Hz and a range of ±10g; a torque sensor mounted at the blade root with a range of 0-200N·m and an accuracy of 0.2%, providing a reference signal proportional to the aerodynamic load; and a hydraulic system for controlling the preload of the bolt assembly. This system can quantitatively adjust the installation stiffness of the blade. The data acquisition frequency is set primarily to balance the load of distortion-free capture and data processing of the blade's main modal frequencies (preliminary modal tests showed all frequencies were below 50Hz). Based on the Nyquist sampling theorem, a sampling frequency of 200Hz provides sufficient data resolution for subsequent analysis.

[0036] The experiment included two analysis groups: a control group, which analyzed only the root mean square value changes of the structural vibration signal; and a sample group based on the present invention, which used the method disclosed in the aforementioned specific embodiments, employing the torque sensor signal as a proxy for the generator's active power signal (i.e., the aerodynamic load proxy signal), and the acceleration sensor signal as the structural response signal, and calculating the aeroelastic transfer function between the two. and coherence function The test process is divided into four stages: Stage 1 (baseline state), the blade installation stiffness is set to 100%, the wind tunnel wind speed is maintained at 10 m / s, and the test is run for 30 minutes to establish the baseline transfer function. The system continuously collects and processes data during each test phase, including the baseline value of vibration amplitude; Phase 2 (load enhancement), maintaining 100% stiffness, increases the wind speed to 12 m / s; Phase 3 (stiffness reduction), restores the wind speed to 10 m / s, and simultaneously reduces the bolt preload to decrease the blade installation stiffness to 95%; Phase 4 (combined operating conditions), while maintaining 95% stiffness, increases the wind speed back to 12 m / s; During each test phase, the system continuously collects and processes data, including the vibration amplitude change rate and the aeroelastic transfer function drift exponent (defined as...). and The integral difference within the key frequency band is used as the evaluation index. Table 1 shows the record of key data at each stage.

[0037] Table 1: Comparison of Experimental Data

[0038]

[0039] Referring to Table 1, the vibration amplitude analysis method used in the control group showed an increase in amplitude in both stage two (load enhancement only) and stage three (stiffness decrease only), and the results could not distinguish between the two different physical causes. In the combined condition of stage four, the amplitude change was the largest, further mixing the effects of the two factors. In contrast, the method used in the sample group of this invention showed only a small fluctuation of 0.8% in the AETF drift index in stage two, showing low sensitivity to load changes. In stages three and four, regardless of the aerodynamic load level, the AETF drift index remained stable above 11%, and this change was directly consistent with the actual decrease in structural stiffness.

[0040] Example 3: This example combines Figures 1 to 3 This section describes an online fatigue assessment system for wind turbine blade materials based on depth acoustic signatures. Figure 1 As shown, the raw signals and rotor azimuth signals collected by the wind turbine sensors are sent to the data acquisition module. This module outputs structural vibration signals and synchronizes them with the SCADA event logs from the SCADA system to form a synchronization data segment. This data segment is filtered by the arbitration data validity module, and the valid synchronization data segments are transmitted to the dynamic characteristics calculation module. This module calculates and generates coherence functions and aeroelastic transfer functions, and inputs them into the trend analysis and evaluation module. This module performs baseline comparison based on the A1 baseline function library, outputs fatigue assessment results to the operation and maintenance team or monitoring system, and triggers the phase analysis diagnostic module when stiffness is confirmed to decrease. This diagnostic module combines the structural vibration signals and rotor azimuth signals to output the target blade positioning information to the operation and maintenance team or monitoring system.

[0041] like Figure 2 As shown, the vertical axis represents the AETF drift index, and the horizontal axis represents time in days. The solid line in the figure represents the long-term evolution of material fatigue, which shows a slow, unidirectional, and continuously accumulating growth trend. The dashed line represents the short-term reversible event of blade icing, which is characterized by a sharp, pulse-like increase that occurs at a specific point in time and then quickly returns to normal levels. This figure intuitively reveals that by analyzing the time-scale characteristics of the AETF drift index, it is possible to effectively distinguish between the permanent decrease in structural stiffness and the temporary disturbances caused by external environmental factors.

[0042] like Figure 3As shown, under normal monitoring conditions, the system continuously calculates the transfer function and coherence function and compares them with the baseline. If the coherence function is detected to be below the threshold in the preset frequency band, it is determined that there is an independent mechanical vibration source unrelated to aerodynamic load, and enters the mechanical fault pending investigation state. If the instantaneous rate of change of the transfer function is detected to exceed the icing threshold, a blade icing warning is triggered, and it can return to normal monitoring as the temperature rises. If the drift index of the transfer function is detected to exceed the threshold and the coherence function remains stable, it is determined to be a stiffness reduction warning, and after confirming the trend, the phase analysis diagnostic mode is activated. The target blade is located by collecting azimuth angle signals and performing time synchronization averaging, thus constructing a complete diagnostic path from condition monitoring, cause differentiation to fault location.

[0043] Example 4: This example elaborates on the in-depth diagnostic procedure from global state assessment to component-level fault location in the aforementioned technical solution, as well as the method for determining the key judgment thresholds; in one application scenario, the system's cloud analysis module analyzes the aeroelastic transfer function Long-term trend analysis confirmed that the AETF drift index of a wind turbine unit showed a continuous unidirectional increase exceeding the warning threshold, and the coherence function during the same period... To maintain stability, the warning threshold is not a fixed value, but rather is determined during the initial system deployment by adjusting the baseline transfer function. The threshold was determined by analyzing statistical data generated during the establishment process; it was set as the sum of the mean and three standard deviations of a series of AETF drift index sample values ​​calculated under healthy conditions. This provides a quantitative benchmark for the specific unit based on statistical confidence for subsequent anomaly detection. After the AETF drift index is confirmed to exceed this statistical threshold, the system determines that the overall stiffness of the unit is showing a downward trend. The cloud analysis module automatically sends a command to the edge computing module of the unit to activate the phase analysis diagnostic mode. In this mode, the data acquisition module continuously collects structural vibration signals. Simultaneously, an azimuth angle signal representing the impeller rotation phase is acquired. The edge computing module then executes an algorithm centered on time synchronization averaging, the operation of which is as follows: using the azimuth signal... The pulse signal indicating zero-degree impeller rotation serves as the synchronous trigger reference, transmitting continuous structural vibration signals. The data is divided into multiple segments, each corresponding to one rotation of the impeller in time. A first-in-first-out queue is created in memory to store the vibration data segments corresponding to the most recent 500 rotation cycles. All 500 data segments stored in the queue are averaged point-by-point to generate an enhanced vibration component containing only the length of a single rotation cycle—the time-synchronized average waveform. .

[0044] Time-synchronized average waveform In the waveform, the aperiodic vibration components caused by random wind loads cancel each other out due to the averaging process, while the deterministic vibration components related to a specific blade rotation position are preserved and enhanced. The edge computing module further refines this waveform. Peak detection is performed to determine the phase angle corresponding to the maximum amplitude. In this scenario, the system calculates the azimuth angle corresponding to the peak value as 125 degrees. According to the blade installation phase diagram of the unit, this angle corresponds to the rotational position of the second blade as it passes in front of the tower. Based on this, the system outputs a diagnostic conclusion, locating the source of the overall stiffness reduction from the entire impeller system to the second blade. To ensure the baseline transfer function... To ensure the effectiveness of the unit throughout its entire lifecycle, the cloud-based analytics module also includes a baseline adaptive update procedure. This procedure, on a one-year cycle, automatically retrieves all historical data segments from the past year that have been deemed valid by the data validity arbitration module and whose operating parameters are all within a stable range. It then uses this data to update the existing data. Weighted updates are performed to allow the diagnostic benchmark to adapt to the slow drift caused by normal wear and tear of the unit, ensuring the long-term accuracy of fatigue assessment.

[0045] Example 5: To improve the evaluation stability of the system under complex operating conditions of wind turbines, the system of this invention includes a pre-validation procedure for the original operating data before performing aeroelastic transfer function calculation. This procedure aims to screen the collected synchronous data segments to ensure that the data used for analysis comes from a time window where the aerodynamic characteristics of the wind turbine are relatively stable. Specifically, the system uses a sliding time window of 60 seconds as a unit to perform statistical analysis on the generator active power signal and the blade pitch angle signal from the SCADA system within the window. Only when the standard deviation of the active power signal within the window is less than a preset first power fluctuation threshold and the standard deviation of the blade pitch angle signal is less than a preset first blade pitch angle fluctuation threshold, is the data segment within the time window determined to be valid and sent to the subsequent transfer function calculation process.

[0046] To eliminate interference from the turbine's own control system behavior on dynamic characteristic assessment, this pre-verification procedure also includes an arbitration step based on event logs. While determining the validity of a data segment, the system simultaneously retrieves the corresponding SCADA event and fault logs for that time period. The system has a pre-set list of event codes related to active intervention by the control system, which includes event codes such as pitch system faults, yaw system actions, or emergency shutdowns. If any event code belonging to this list is found within the current time window, the data segment is marked as invalid and discarded, regardless of whether its power and pitch angle fluctuations meet the threshold conditions. Through this dual verification mechanism, only data from periods when the wind turbine is in a stable operating state and there is no significant intervention from the active control system is used for subsequent structural dynamic characteristic assessment, thereby ensuring the long-term consistency and reliability of the assessment results.

[0047] Example 6: Before deploying the system on a wind turbine located in a cold region with a history of blade icing events, to ensure the assessment system can accurately attribute the physical causes of changes in the AETF drift index, an offline model calibration and validation procedure based on historical data must be performed. This procedure determines a set of judgment thresholds for the time-scale characteristic analysis algorithm, tailored to the turbine model and environmental characteristics. The procedure first uses historical data from the turbine or similar models, corresponding to stages of material fatigue confirmed by other methods during past operation, to establish a fatigue evolution model. Analysts extract the AETF drift index time series corresponding to these stages and perform linear regression analysis on the data points within a 30-day sliding window to calculate its long-term monotonic trend slope. Through the analysis of multiple fatigue samples, a slope characteristic value representing material fatigue is determined. .

[0048] Subsequently, the procedures utilized historical data recorded during periods of icing phenomena associated with sudden drops in temperature, increased humidity, and abnormal decreases in power generation during past winter operations of the unit to establish an icing event model. Analysts extracted the AETF drift index time series corresponding to these periods and calculated the first difference of the drift index within a 1-hour sliding window to obtain its instantaneous rate of change. Through the analysis of multiple icing event samples, a characteristic value representing the rate of change of rapid icing processes was determined. Based on the above two characteristic values, the collaborative diagnosis and logical decision-making rules within the system are ultimately quantified and solidified: Rule 1, if the short-scale analysis detects an instantaneous rate of change of the drift index greater than... If the system outputs a blade icing warning, it will temporarily suppress the escalation of the fatigue warning; Rule 2: If long-scale analysis detects that the long-term monotonic trend slope of the drift index is consistently positive and greater than 0.5%, the system will output a blade icing warning and temporarily suppress the escalation of the fatigue warning. Meanwhile, if no obvious anomalies are found in short-scale analysis, the trend will be included in the fatigue accumulation assessment. Rule 3: If, after the system outputs an icing warning, it detects that the drift index falls back to the pre-warning level as the temperature rises, the event will be confirmed as a transient icing event, and its data will be removed from the calculation of the long-term fatigue trend. By executing this set of standardized offline calibration procedures, the original time-scale analysis method based on qualitative description has been transformed into an algorithm module with clear quantitative judgment criteria, so that the system can autonomously identify and separate two system state changes with different time evolution characteristics during actual online operation.

[0049] To further highlight the essential advantages of the present invention over existing conventional technologies in solving the problem of unclear attribution of response changes, the following comparative example 1 is provided for illustration.

[0050] Comparative Example 1: This comparative example aims to verify the effectiveness of a conventional technical approach that relies solely on analyzing the changes in the structural vibration signal itself in distinguishing between two operating conditions: aerodynamic load fluctuations and structural stiffness reduction. Except for the methods described below, the test objects, sensor deployment, data acquisition conditions, and operating condition settings used in this comparative example are kept as consistent as possible with those in Example 2 above. This comparative example does not calculate the aeroelastic transfer function between the generator active power signal and the structural vibration signal. Instead of analyzing the root mean square (RMS) value of the structural vibration signal, the unit's condition is assessed by directly analyzing the structural vibration signal.

[0051] The experiment was also conducted on a continuously operating wind turbine generator, with a monitoring period of 240 hours. To simulate common challenges in real-world operation, this monitoring period included two distinct phases: Phase A (Intensified Load Condition): From hour 48 to 72 of the monitoring period, the generator encountered sustained strong winds, with the average wind speed increasing by approximately 20% compared to normal. During this phase, the blade structural stiffness remained unchanged. Phase B (Decreased Stiffness Condition): From hour 180 to 204 of the monitoring period, the wind speed returned to normal levels, but the root connection stiffness of one blade was reduced by approximately 5% through simulation to simulate early material fatigue. Throughout the entire monitoring period, the system continuously recorded the structural vibration signals output by the nacelle acceleration sensor. The rate of change of the 10-minute root mean square value was calculated. Table 2 records the key evaluation indicators obtained by using the comparative method of this invention (analyzing only vibration signals) and the method of this invention (analyzing AETF drift index) in the two characteristic stages. The results are shown in Table 2.

[0052] Table 2: Comparison of comparative experimental data.

[0053]

[0054] Referring to Table 2, in Stage A (load enhancement only), the vibration amplitude analysis method used in the comparative study detected an increase of 18.7%, which is even higher than the increase caused by the decrease in true stiffness in Stage B. In actual operation and maintenance, this phenomenon can easily lead to misjudgments of structural health, potentially resulting in unnecessary downtime for maintenance. In contrast, the method of this invention, in Stage A, showed only a slight fluctuation of 1.1% in its core indicator, the AETF drift index, accurately reflecting the fact that the structural characteristics had not changed, demonstrating strong robustness to changes in purely aerodynamic loads. In stage B (stiffness decrease only), when the vibration amplitude changed by 14.2%, the AETF drift index of the method of this invention stably pointed to a drift of 10.8%, which showed a direct and clear correspondence with the actual decrease in structural stiffness. The experimental data of this comparative example proves that the conventional technical approach of analyzing only a single structural vibration signal is seriously coupled with the two completely different physical causes of aerodynamic load excitation and structural stiffness change, and cannot effectively distinguish between the two. Therefore, it is difficult to use it as a basis for assessing the fatigue of blade materials.

[0055] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0056] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. An online fatigue assessment system for wind turbine blade materials based on depth acoustic signatures, characterized in that, include: A data acquisition module is configured to synchronously acquire, from the standard operating data stream of the wind turbine, the time series of the generator active power signal, the time series of the structural vibration signal, and the SCADA event log that records the unit's operating events; A data validity arbitration module is configured to check the synchronously acquired data segments of active power signal and structural vibration signal before performing calculations, based on whether there are preset event codes in the SCADA event log that represent power grid disturbances, and only determine the data segments without event codes as valid data segments. An edge computing module is configured to calculate the cross-spectral density and auto-spectral density of the active power signal and the structural vibration signal, respectively, for only valid data segments. Based on the calculation results, it generates an aeroelastic transfer function characterizing the dynamic transmission relationship between the two, and a coherence function characterizing the proportion of energy in the structural vibration signal caused by the linearity of the active power signal. The edge computing module is configured to generate the aeroelastic transfer function through the following operations. , ,in, For frequency, This represents the cross-spectral density of the active power signal and the structural vibration signal. The autospectral density of the active power signal; A cloud-based analysis module is configured to compare the aeroelastic transfer function of the current cycle with a preset baseline transfer function, and only when the value of the coherence function in the preset frequency band continuously meets a first threshold condition will the trend change generated by the comparison and characterizing the decrease in blade stiffness be included in the fatigue assessment results. The cloud-based analysis module is also configured to compare the coherence function of the current cycle with the preset baseline coherence function, and when a decrease in the coherence function below a second threshold is detected in the preset frequency band, it is determined that there is a mechanical fault vibration source unrelated to aerodynamic load, and this is used to distinguish the decrease in blade stiffness.

2. The online fatigue assessment system for wind turbine blade materials based on depth acoustic signatures according to claim 1, characterized in that, The cloud analysis module is also configured to activate a phase analysis diagnostic mode when it outputs an assessment result indicating a decrease in blade stiffness; in this mode, the data acquisition module is also configured to acquire an azimuth signal characterizing the impeller rotation phase. Furthermore, the edge computing module is configured to: use the azimuth signal as a reference to perform time-synchronous averaging processing on the structural vibration signal, and locate the target blade that generates abnormal vibration based on the phase information of the enhanced vibration component.

3. The online fatigue assessment system for wind turbine blade materials based on depth acoustic signatures according to claim 2, characterized in that, The time-synchronized averaging process includes: dividing the continuous structural vibration signal into multiple data segments with one revolution of the impeller as the cycle, and averaging all data segments point by point.

4. The online fatigue assessment system for wind turbine blade materials based on depth acoustic signatures according to claim 1, characterized in that, The cloud-based analysis module is also configured to perform time-scale characteristic analysis on the trend changes of the aeroelastic transfer function over time; this analysis includes calculating the rate of change of the trend change within a short time-scale window in hours. And within a long-term timescale window measured in weeks, calculate the monotonic increments of trend changes; Based on the characteristic differences between the rate of change and the monotonic increment, the physical causes of the trend change are determined.

5. The online fatigue assessment system for wind turbine blade materials based on depth acoustic signatures according to claim 1, characterized in that, The data validity arbitration module includes preset event codes that characterize power grid disturbances, including event codes for generator grid voltage drops or frequency anomalies. The arbitration module is configured to mark the synchronously acquired data segments within the corresponding time range as invalid and discard them when this type of code is retrieved in the SCADA event log.

6. The online fatigue assessment system for wind turbine blade materials based on depth acoustic signatures according to claim 1, characterized in that, The structural vibration signal originates from the acceleration sensor deployed on the nacelle or tower of the wind turbine.

7. The online fatigue assessment system for wind turbine blade materials based on depth acoustic signatures according to claim 1, characterized in that, The baseline transfer function is established by statistically averaging multiple aeroelastic transfer functions calculated during the initial operation phase of the wind turbine.

Images