A fan blade health state evaluation method and system

Abstract

The application relates to the field of state evaluation, and discloses a fan blade health state evaluation method and system, which are used for intelligent evaluation of the health state of a blade. The method comprises the following steps: acquiring a blade vibration acceleration signal, extracting quantiles based on historical sliding time window statistics as a dynamic reference for amplitude normalization, resampling the signal from a time domain to an angle domain through cubic spline interpolation, enhancing a signal-to-noise ratio through multi-rotation phase alignment and average processing, reconstructing a phase space for the angle domain signal, extracting a recurrence rate and a certainty index, calibrating a distance threshold value through a percentile method, comparing the two indexes with fixed health references respectively, obtaining a comprehensive deviation degree, and outputting an evaluation result of the health or damage of the blade after comparison with a preset threshold value. The application adopts a dynamic updating reference to adapt to natural aging of equipment, adopts double-index fusion to improve overall evaluation comprehensiveness, and adopts self-adaptive parameters to enhance overall working condition applicability, so that accurate and reliable evaluation of the health state of the blade is realized.

Description

Technical Field

[0001] This invention relates to the field of condition assessment, and more particularly to a method and system for assessing the health status of wind turbine blades. Background Technology

[0002] As a crucial component of global renewable energy, wind power generation continues to grow in installed capacity. Wind turbine blades, as the core component for wind energy capture, operate under complex conditions such as alternating loads, wind shear, turbulence, and lightning strikes, making them highly susceptible to structural damage such as cracks, delamination, and erosion. Blade failure not only causes significant economic losses but can also lead to safety accidents. Therefore, real-time and accurate health status assessment of wind turbine blades has become a core technical requirement in the wind power operation and maintenance field.

[0003] In summary, existing wind turbine blade health status assessment technologies have the following main shortcomings: Existing angle domain resampling methods mostly employ simple linear interpolation, which has limited interpolation accuracy in non-uniform speed operation ranges. At the same time, phase space reconstruction is performed directly after resampling without phase alignment and averaging of multi-turn signals. Random noise and transient interference are not effectively suppressed, resulting in a high proportion of noise pseudo-recursive points in the recursion matrix of subsequent recursive quantitative analysis (RQA) analysis, thus reducing the reliability of the indicators. Existing methods typically extract only one of the recursion rate or determinism as the criterion. However, in the early stages of leaf damage, the decrease in determinism indicators (weakened systematic regularity) may be the first sign, while the recursion rate may not yet show significant changes; as the damage intensifies, the recursion rate begins to deviate significantly. A single indicator cannot cover the entire process of damage evolution, easily leading to early missed detections or late misjudgments. Existing multi-indicator fusion methods often employ simple equal-weighting or subjective weighting, failing to consider the different contributions of recursion rate and deterministic indicators to damage characterization. Furthermore, the threshold relationship between comprehensive indicators and health assessment is not clear enough, resulting in insufficient interpretability and consistency of assessment results.

[0004] Therefore, we propose a method and system for assessing the health status of wind turbine blades to address the aforementioned issues. Summary of the Invention

[0005] This invention provides a method and system for assessing the health status of wind turbine blades, which is used for intelligent assessment of the health status of blades.

[0006] The first aspect of this invention provides a method for assessing the health status of wind turbine blades. The method includes: acquiring a vibration signal of the wind turbine blade; performing amplitude processing on the vibration signal to obtain a normalized signal; acquiring a rotational speed signal of the wind turbine blade; resampling the normalized signal from the time domain to the angle domain based on the rotational speed signal to obtain an angle-domain vibration signal; reconstructing the phase space of the angle-domain vibration signal to construct a phase space point set; calculating a recursion matrix based on the phase space point set, and extracting a recursion rate and a deterministic index from the recursion matrix; determining assessment feature values ​​based on the recursion rate, the deterministic index, and preset benchmark feature parameters; and outputting the health status assessment result of the wind turbine blade based on the assessment feature values.

[0007] Optionally, in a first implementation of the first aspect of the present invention, the method includes: The cycle time information of the wind turbine blades is extracted based on the rotation speed signal; Based on a preset number of sampling points per cycle, the normalized signal corresponding to the cycle time information is interpolated and resampled to obtain the angular domain vibration signal with equal angular intervals between adjacent sampling points.

[0008] Optionally, in a second implementation of the first aspect of the present invention, it further includes: Select the angular domain vibration signals from multiple consecutive operating cycles that meet the preset operating conditions; The selected angular domain vibration signals are aligned according to the same rotation position and averaged to obtain an enhanced angular domain vibration signal, which is then used as the input for the phase space reconstruction.

[0009] Optionally, in a third implementation of the first aspect of the present invention, the method includes: Extract time-series data from the angular domain vibration signal; Based on a preset delay time and embedding dimension, the time series data is reconstructed into multiple phase space vectors containing a fixed number of data points, and all the phase space vectors constitute the phase space point set.

[0010] Optionally, in a fourth implementation of the first aspect of the present invention, the delay time is determined by calculating the local minimum value of the statistical dependence degree as a function of the delay amount; The embedding dimension is determined by calculating the number of data points corresponding to a false nearest neighbor ratio that is lower than a preset threshold.

[0011] Optionally, in a fifth implementation of the first aspect of the present invention, the method includes: Calculate the distance between each pair of phase space points in the phase space point set; The distance value is compared with a distance threshold to determine the recursive and non-recursive points in the recursive matrix; The recursion rate is calculated based on the proportion of the recursion points in the recursion matrix. Identify valid diagonals in the recursive matrix whose length is greater than or equal to a preset length, and calculate the deterministic index based on the ratio of the number of recursive points on the valid diagonals to the total number of recursive points.

[0012] Optionally, in a sixth implementation of the first aspect of the present invention, the method for determining the distance threshold includes: Obtain the set of distances between all pairs of phase space points in the phase space point set; The distance value corresponding to the preset real percentile position in the distance set is extracted as the distance threshold.

[0013] Optionally, in a seventh implementation of the first aspect of the present invention, the method further includes calculating the recursive accumulation energy index (CEI): ; Where DET is the deterministic indicator and RR is the recursion rate. The total number of valid diagonals identified. This represents the total number of all recursive points.

[0014] Optionally, in an eighth implementation of the first aspect of the present invention, the method includes: Calculate the first deviation of the recursion rate from the benchmark recursion rate, and the second deviation of the deterministic index from the benchmark deterministic index. The benchmark recursion rate and the benchmark deterministic index together constitute the benchmark feature parameter. The first deviation and the second deviation are weighted and summed to obtain the evaluation feature value; The evaluation feature value is compared with a preset deviation threshold. If the evaluation feature value is greater than the preset deviation threshold, an evaluation result indicating that the wind turbine blades are damaged is output.

[0015] A second aspect of the present invention provides a wind turbine blade health status assessment system, comprising: an amplitude processing module for acquiring vibration signals of the wind turbine blade and performing amplitude processing on the vibration signals to obtain a normalized signal; a resampling module for acquiring rotational speed signals of the wind turbine blade and resampling the normalized signal from the time domain to the angle domain based on the rotational speed signals to obtain an angle domain vibration signal; a phase space reconstruction module for performing phase space reconstruction on the angle domain vibration signal to construct a phase space point set; a feature extraction module for calculating a recursion matrix based on the phase space point set and extracting a recursion rate and a deterministic index from the recursion matrix; and a status assessment module for determining assessment feature values ​​based on the recursion rate, the deterministic index, and preset benchmark feature parameters, and outputting a wind turbine blade health status assessment result based on the assessment feature values.

[0016] The mechanism of this invention is as follows: by comparing the deviation from the health benchmark, the abnormal vibration mode caused by blade structural damage is transformed into a quantitative shift of recursive characteristics, thereby realizing a model-independent and condition-resistant intelligent assessment of the blade health status. Beneficial effects: Through the dynamic benchmark update mechanism, it can capture short-term abnormal deviations in state and effectively filter out gradual interference caused by long-term natural aging of equipment. It has a significant sensitivity advantage, especially in the early stage of damage, and can achieve true early warning. Abnormal rotation speeds are eliminated by screening based on operating conditions, and coherent cancellation of noise and coherent enhancement of effective signals are achieved by phase-aligned averaging. This allows subsequent RQA analysis to be based on enhanced signals with high signal-to-noise ratios, significantly improving the stability and repeatability of recursion rate and deterministic indicators, and making the evaluation results more reliable. Cubic spline interpolation can fit the continuous trend of vibration signals more accurately while maintaining computational efficiency. Especially in the non-uniform speed range, the waveform fidelity of the angular domain signal is significantly better than that of linear interpolation, providing higher quality input data for subsequent phase space reconstruction. The adaptive method allows the parameters to be determined entirely by the data itself, which avoids the bias caused by subjective experience and achieves long-term stability of the parameters through one-time calibration at the beginning of installation. It can maintain the best phase space reconstruction quality and recursive matrix analysis accuracy across the entire wind speed, speed, and pitch angle range. Simultaneously, two complementary indicators, recursion rate (representing the density of recursion points) and determinism (representing the ratio of diagonal structures), are extracted. The comprehensive deviation is obtained by weighting and summing the indicators with preset real number weights and comparing them with a fixed threshold to output the evaluation result. By integrating two indicators, the recursion rate is used to capture changes in the overall complexity of the system, while the determinism is used to capture the degradation of the system's regular structure. The two complement each other, and the comprehensive deviation can more comprehensively and sensitively reflect the complete evolution process of the leaf from health to damage, significantly reducing the false alarm rate and the missed detection rate. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of an embodiment of the wind turbine blade health status assessment method according to the present invention; Figure 2 This is a schematic diagram of another embodiment of the wind turbine blade health status assessment method in this invention; Figure 3 This is a schematic diagram of one embodiment of the wind turbine blade health status assessment system of the present invention; Figure 4 This is a schematic diagram of one embodiment of the wind turbine blade health status assessment device in this invention. Detailed Implementation

[0018] This invention provides a method and system for assessing the health status of wind turbine blades, enabling intelligent evaluation of blade health. The terms first, second, third, fourth, etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms include or have, and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus comprising a series of steps or units is not necessarily limited to those explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0019] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 One embodiment of the wind turbine blade health status assessment method in this invention includes: 101. Obtain the vibration acceleration signal during the operation of the wind turbine blades, and perform amplitude normalization processing on the vibration acceleration signal to obtain a normalized signal.

[0020] It is understood that the executing entity of this invention can be a wind turbine blade health status assessment system, or it can be a terminal or a server; no specific limitation is made here. This embodiment of the invention will be described using a server as an example.

[0021] It should be noted that Example 1 is applicable to customized high-precision monitoring scenarios with the addition of a high-frequency acquisition terminal and a high-precision encoder. During the data acquisition phase, real-time monitoring data transmitted by an accelerometer installed at the root of the wind turbine blades is received. The wind turbine is currently in stable operation, with a blade rotation speed of 15 revolutions per minute, meaning one rotation takes 4 seconds. A continuous signal segment of 16 seconds is extracted, containing four complete rotation cycles. The sensor operates at a sampling frequency of 1000 Hz, therefore, a total of 16,000 discrete data points are collected within this 16-second period, measured in meters per square second.

[0022] Among the 16,000 raw data points acquired, the signal exhibits continuous fluctuations due to the periodic effects of wind load and centrifugal force on the blades. To ensure the comparability of data from different monitoring periods, the system extracts the 99th percentile of the vibration amplitude based on a historical sliding time window (the most recent three months when the wind turbine was in a healthy state) as a dynamic normalization benchmark. The values ​​of the 16,000 raw data points are divided one by one by this 99th percentile value (which is 6.5 in the current statistical period) to periodically update the benchmark value to adapt to normal equipment aging and prevent signal truncation and distortion.

[0023] Here are some examples of the conversions: 2.5 divided by 5.0 at 0.1 seconds gives a normalized value of 0.5; -2.0 divided by 5.0 at 4.5 seconds gives a normalized value of -0.4; and 5.0 divided by 5.0 at 12.2 seconds gives a normalized value of 1.0.

[0024] After point-by-point division, a pure numerical sequence containing 16,000 data points with amplitudes ranging from -1 to +1 is obtained.

[0025] 102. Based on the rotational speed pulse signal of the wind turbine blades, the normalized signal is resampled from the time domain to the angle domain to obtain the angle domain vibration signal.

[0026] It should be noted that after acquiring the normalized signal with a duration of 16 seconds and a total of 16,000 data points, the server synchronously reads the speed pulse signal transmitted back by the wind turbine main shaft encoder. The encoder is specified to output 360 pulses per revolution, that is, one pulse is generated for every 1 degree of rotation.

[0027] Within the current 16-second sampling window, the fan operates at its nominal speed of 15 revolutions per minute, completing a total of 4 rotations, or 1440 degrees. Therefore, the server receives a total of 1440 pulse signals representing angular positions during this time period.

[0028] Since the rotational speed is not absolutely constant during actual operation, the server discovered slight differences in the rotation time of the spindle in different intervals by recording the timestamp of each pulse. The time interval between the 360th and 361st pulses is 0.0110 seconds, while the time interval between the 720th and 721st pulses is 0.0113 seconds. The server uses these timestamps to establish a non-linear mapping relationship between physical time and rotation angle.

[0029] A resampling operation is then performed. The target angular resolution is set to 0.1 degrees, meaning that one vibration value is extracted for every 0.1 degrees of rotation. For this total rotation range of 1440 degrees, the server needs to re-obtain 14400 data points at equal angular intervals.

[0030] Based on the mapping relationship between time and angle, the server uses an interpolation algorithm to calculate the vibration amplitude corresponding to each 0.1-degree target angle. Taking the blade rotating to a precise 45.0-degree position as an example: the server, through mapping, determines that the time corresponding to this angle is 0.5004 seconds. This time lies between the 500th sampling point (0.500 seconds) and the 501st sampling point (0.501 seconds) of the original time-domain signal. The value of the 500th point is known to be -0.400, and the value of the 501st point is known to be -0.390. The server performs interpolation calculations on these two adjacent values, obtaining a vibration value corresponding to the 45.0-degree position as -0.396.

[0031] Following this logic, all 14,400 target angles are traversed, and the original 16,000 time-domain data points with equal time intervals are transformed into a sequence of 14,400 equally spaced angles.

[0032] 103. Perform phase space reconstruction on the angular domain vibration signal to construct a phase space point set consisting of multiple phase space points.

[0033] It should be noted that the server is set to embedding dimension of 3 (reconstructing a three-dimensional space). To ensure that the reconstructed trajectory can approximate the physical and dynamic characteristics of the system, the delay parameter is pre-calculated using the mutual information method. At the current nominal rotational speed, the optimal delay span determined by the mutual information method is approximately 90 degrees. Since the data resolution is 0.1 degrees, 90 degrees corresponds to an interval of 900 data points.

[0034] The reconstruction process constructs coordinate points in multidimensional space using a sliding window.

[0035] Constructing the first phase space point: Starting from the first point in the sequence, the server sequentially extracts the 1st, 901st, and 1801st points based on a delay parameter spanning 900 points. Assume the values ​​at these three positions are -0.150, -0.080, and 0.020, respectively. These three values ​​are used as the x, y, and y coordinates of the point in three-dimensional space. This yields the first three-dimensional data point.

[0036] Constructing the second phase space point: The server slides the starting position one point backward, starting from the second point of the sequence. The second point, the 902nd point, and the 1802nd point are extracted sequentially and combined to form the second data point in three-dimensional space.

[0037] The server slides backwards one by one on the sequence containing 14,400 data points according to this logic. Since the construction of each 3D point requires traversing a total of 1,800 data points (spanning two intervals of 900 from the first dimension to the third dimension), the total number of phase space points that can be constructed is 14,400 minus 1,800, which is 12,600. After traversal processing, 12,600 3D spatial coordinate points are obtained.

[0038] 104. Calculate the recursion matrix based on the phase space point set, and extract the recursion rate and deterministic index from the recursion matrix. The recursion rate is used to characterize the density of recursion points, and the deterministic index is used to characterize the diagonal structure ratio.

[0039] It should be noted that the server received a set of 12,600 three-dimensional phase space points, resulting in a two-dimensional blank matrix with 12,600 rows and 12,600 columns. The system's preset spatial distance threshold is 0.05.

[0040] When constructing the matrix, the system calculates the straight-line distances between each of the 12,600 spatial points. Taking the 1st point and the 3601st point (separated by a full 360-degree circle) as an example: the system calculates their distance in 3D space to be 0.03. Since 0.03 is less than the threshold of 0.05, recursion is detected, and the coordinate pair is recorded in the system (logically corresponding to a matrix position of 1). Taking the 1st point and the 1801st point (separated by half a circle) as another example: the calculated distance is 0.45, which is greater than 0.05, indicating no recursion has occurred, and the system does not record it (logically, it is 0). To avoid extremely high computational overhead, the system uses spatial indexing technology (KD-tree) for neighborhood search, calculating and sparsely storing only recursive point pairs that satisfy the distance threshold, without needing to construct and traverse the complete global dense matrix.

[0041] The recursion rate, i.e., the probability of state regression in the phase space trajectory, is extracted. Through spatial neighborhood search statistics, there are 7,938,000 recursive point pairs that meet the distance threshold. Dividing 7,938,000 by the theoretical total number of point pairs (12,600 × 12,600 = 158,760,000), the recursion rate is 0.05. This indicates that the wind turbine blades are in a similar state for about 5% of the time during operation.

[0042] Next, a deterministic index was extracted. The system presets a minimum effective diagonal length of 3, representing a stable trajectory that maintains at least 3 consecutive steps. Server scanning revealed that among 7,938,000 recursive points with a value of 1, 6,350,400 points were distributed along diagonals with a length greater than or equal to 3. Dividing 6,350,400 by 7,938,000 yielded a deterministic index of 0.80. This indicates that the vibration signal exhibits a high degree of regularity.

[0043] At this point, the server successfully extracted a recursion rate of 0.05 and a determinism metric of 0.80.

[0044] 105. Compare the recursion rate and deterministic index with the preset health baseline recursion rate and health baseline deterministic index respectively to obtain the comprehensive deviation. Based on the relationship between the comprehensive deviation and the preset threshold, output the evaluation result of the leaf health status.

[0045] It should be noted that the server receives the current operational characteristics: a recursion rate of 0.05 and a deterministic index of 0.80. Simultaneously, it extracts historical benchmark data from the database under normal operating conditions: a healthy benchmark recursion rate of 0.04 and a healthy benchmark deterministic index of 0.90.

[0046] In the calculation phase, the relative deviation is calculated separately. For the recursion rate, the absolute difference between the current value and the benchmark value is 0.01. Dividing this by the benchmark value of 0.04 yields a relative deviation of 0.25. For the deterministic index, the absolute difference is 0.10. Dividing this by the benchmark value of 0.90 yields a relative deviation of approximately 0.11.

[0047] The overall deviation was calculated using a weighted average. The weighting (recursion rate 0.4, determinism 0.6) was derived from principal component analysis (PCA) of the historical failure sample set of this blade model. The deterministic index had a higher variance contribution rate in the early stages of structural fatigue. Multiplying the relative deviation of the recursion rate (0.25) by the weight (0.4), and adding the relative deviation of the deterministic index (0.11) by the weight (0.6), the overall deviation was calculated to be 0.166.

[0048] The correlation between the overall deviation and the leaf health status is shown in Table 1 below: Table 1 Less than 0.10 Normal status Maintain the current sampling and monitoring frequency 0.10 to 0.20 (inclusive) Mild abnormality Increase monitoring frequency to obtain early warning logs 0.20 to 0.35 (inclusive) Moderate injury An alarm has been triggered; it is recommended to schedule a system shutdown for inspection. Greater than 0.35 Serious Fault Trigger emergency stop command and start damage assessment procedure. In the results output stage, the server substitutes the comprehensive deviation value of 0.166 into the table for comparison. The threshold division of comprehensive deviation (0.10, 0.20) is based on the damage mapping data calibration of the full-size fatigue test bench for this type of blade. After verification by non-destructive testing, when the deviation is between 0.10 and 0.20, the corresponding blade skin shows microcracks (length <50mm) or the structural stiffness decreases by about 3%.

[0049] Please see Figure 2 Example 2 is applicable to conventional Supervisory Control and Data Acquisition (SCADA) monitoring scenarios that only have single-cycle pulses and standard vibration sensors. Another embodiment of the wind turbine blade health status assessment method in this invention includes: 201. Obtain the vibration acceleration signal during the operation of the wind turbine blades, and perform amplitude normalization processing on the vibration acceleration signal to obtain a normalized signal.

[0050] Specifically, amplitude normalization processing includes: continuously acquiring vibration acceleration signals using an accelerometer installed at the blade root; extracting the absolute amplitude value at each time point from the vibration acceleration signal; and extracting the vibration acceleration signal and obtaining its amplitude distribution characteristics. Using historical data collected under fault-free, stable operating conditions (first month after a major overhaul), the dynamic quantile (99th percentile) of the vibration amplitude is statistically derived as a health baseline characteristic parameter. The system sets a baseline update cycle; when maintenance verification confirms the blade is in a healthy state, the latest data is used to dynamically update this quantile baseline, effectively avoiding false alarms caused by baseline drift due to natural equipment aging.

[0051] Entering the real-time monitoring and normalization phase, the wind turbine is in a stable daily operation state. Accelerometers at the blade roots continuously collect vibration acceleration signals, with the sampling frequency strictly maintained at 500 Hz. To perform the current health status assessment, the server extracts a 10-second segment of continuous vibration signals from the real-time data stream. This segment contains 5000 raw sampling data points with physical dimensions, measured in meters per square second. The data at this point includes both forward and reverse vibrations, with complete physical phase information.

[0052] The amplitude of this segment is normalized. Unlike the calibration stage, which only extracts absolute values, in order to preserve the complete dynamic topology of the system and the positive and negative directional characteristics of the signal, the server extracts the original vibration acceleration values ​​corresponding to these 5000 time points one by one (preserving the original positive and negative signs), and divides each original value directly by the dynamic reference value that is effective in the current update cycle (the reference value is updated to 6.5 during this monitoring period).

[0053] The data transformation process is illustrated using data from three specific time points within this 10-second clip as an example: At the first selected time point, the sensor's original vibration acceleration measurement was 2.6 m / s². The system extracted the value 2.6, divided it by the dynamic baseline value of 6.5, and calculated the dimensionless normalized value of 0.4.

[0054] At the second time point, the sensor's original vibration acceleration measurement was -3.9 m / s². The system extracted the directional value of -3.9, divided it by the dynamic baseline value of 6.5, and calculated the dimensionless normalized value of -0.6.

[0055] At the third time point, the sensor's original vibration acceleration measurement was -1.3 m / s². The system extracted the directional value of -1.3, divided it by the dynamic baseline value of 6.5, and calculated the dimensionless normalized value of -0.2.

[0056] Following the unified division logic described above, all 5000 original data points within the 10-second data segment were converted one by one. After processing, the 5000 vibration acceleration data points, originally with different positive and negative characteristics and physical units, were scaled proportionally and converted into a pure numerical sequence with values ​​distributed between -1 and +1, preserving the alternating properties of the original signal.

[0057] 202. Based on the rotational speed pulse signal of the wind turbine blades, the normalized signal is resampled from the time domain to the angle domain to obtain the angle domain vibration signal.

[0058] Specifically, resampling to the angle domain includes: acquiring a wind turbine blade rotation speed pulse signal that is synchronously acquired with the vibration acceleration signal; calculating the duration of each rotation of the blade based on the time interval between adjacent pulses in the rotation speed pulse signal; resampling the vibration acceleration signal within the duration using linear interpolation based on a preset number of sampling points per rotation to obtain an angle domain vibration signal with a fixed number of sampling points per rotation; the blade rotation angle intervals corresponding to adjacent sampling points in the angle domain vibration signal are equal, and the number of sampling points remains unchanged throughout the entire operating cycle of the blade.

[0059] Furthermore, the linear interpolation employs cubic spline interpolation, specifically including: within the duration interval corresponding to each rotational speed pulse, using the vibration values ​​at the start and end points of the interval and the vibration values ​​of each original sampling point within the interval as interpolation nodes, and using a cubic spline function to fit a continuous interpolation curve; according to the preset number of sampling points per revolution, extracting values ​​from the interpolation curve at equal angular intervals to form the angular domain vibration signal for that revolution.

[0060] It should be noted that during the speed pulse extraction and period definition stage, the server receives the 10-second dimensionless normalized vibration signal (a total of 5000 data points, with a fixed time interval of 0.002 seconds) stored in the queue in step 201, and simultaneously reads the pulse signal transmitted back by the speed sensor installed on the wind turbine main shaft or gearbox input shaft. The hardware of this speed sensor is set to output a reference pulse for every full rotation (360 degrees) of the blade main shaft.

[0061] The system scans the rotational speed pulse time series data within this 10-second time window, identifies adjacent pulse timestamps, and uses this to define each complete mechanical rotation cycle of the blade. The system is set to identify three consecutive pulses within the current signal segment: the first pulse occurs at 0.800 seconds of the 10-second time period, the second pulse occurs at 4.850 seconds, and the third pulse occurs at 8.850 seconds.

[0062] Taking the first complete rotation cycle defined by these three pulses as an example, the system calculates the time difference between the second pulse timestamp and the first pulse timestamp, determining that the duration of this rotation is 4.050 seconds. Similarly, the system calculates the time difference between the third pulse and the second pulse, determining that the duration of the next rotation cycle is 4.000 seconds. The variation in time difference objectively reflects the minute fluctuations in the fan speed at the microscopic level.

[0063] Entering the interpolation curve fitting stage, the system performs numerical processing for the first rotation cycle, which lasts 4.050 seconds. Since the sampling frequency of the time-domain signal is 500 Hz, 500 data points are generated per second. Within this 4.050-second time interval, the system extracts the corresponding 2025 original discrete data points from the normalized signal sequence.

[0064] The system employs a cubic spline interpolation algorithm to process this batch of discrete data. Specifically, the system uses the start time (0.800 seconds) and end time (4.850 seconds) of the rotation interval, along with the specific time labels corresponding to the 2025 original sampling points within the interval, as interpolation nodes. Using these time nodes and their corresponding dimensionless normalized values ​​between -1 and +1, the system solves a system of piecewise polynomial equations to fit a smooth and continuous cubic spline interpolation curve. This continuous curve mathematically characterizes the vibration response at any extremely short time point within the rotation period.

[0065] Upon entering the equal-angle resampling stage, the system reads the preset number of sampling points per revolution in the system parameter configuration. In this embodiment, based on the principle of moderate sampling and the low-frequency structural response characteristics of the blade, the system sets the target resampling angle resolution to 1.0 degree (i.e., extracting 360 sampling points at equal angles within a complete rotation cycle), which preserves the dynamic characteristics while avoiding data redundancy and excessive interpolation noise.

[0066] The system divides the 4.050-second period into proportional segments based on the 2000 nodes. The 1000th sampling point within this period corresponds to the mechanical position of the blade rotating half a circle (180 degrees). The corresponding time point is calculated as 0.800 seconds plus 4.050 seconds divided by 2, yielding a target time point of 2.825 seconds. The system directly extracts the ordinate value corresponding to the x-axis at 2.825 seconds from the cubic spline interpolation continuous curve obtained in the previous stage. Assuming the extracted value is 0.45, this value is taken as the vibration amplitude of the angle domain signal at the 180-degree position of that circle.

[0067] Following this logic, the corresponding values ​​of 2000 time nodes are extracted sequentially and at equal intervals from the interpolation curve to form the angular domain vibration signal for that specific rotation. For the next rotation cycle with a duration of 4.000 seconds, the system resets the nodes and repeats the above curve fitting and proportional resampling extraction process, similarly extracting a sequence of 2000 data points.

[0068] After processing each defined complete rotation cycle in turn, the time-domain signal sequence, which was originally affected by speed fluctuations and contained a different number of original data points per revolution, was uniformly reconstructed into an angle-domain vibration signal sequence containing a fixed 2000 sampling points per revolution and with strictly equal angular intervals between adjacent data points (all 0.18 degrees).

[0069] 203. Perform phase space reconstruction on the angular domain vibration signal to construct a phase space point set consisting of multiple phase space points.

[0070] Specifically, phase space reconstruction includes: extracting a continuous sequence of vibration values ​​from the angular domain vibration signal; combining each data point in the extracted vibration value sequence with data points at a specified interval thereafter to form a phase space vector containing a fixed number of data points; traversing the vibration value sequence to obtain a series of phase space vectors, each of which contains a vibration value at a time point and other vibration values ​​at a specified interval thereafter, and the phase space vectors together form a phase space point set.

[0071] Furthermore, the subsequent specified interval is determined as follows: the statistical dependence of the angular domain vibration signal under different delay amounts is calculated, and the delay amount corresponding to the first occurrence of a local minimum in the curve of the statistical dependence as the delay increases is taken as the subsequent specified interval; the number of data points contained in the phase space vector is determined as follows: as the number of data points is gradually increased, the proportion of false nearest neighbors in the phase space points under each number of data points is calculated, and the number of data points corresponding to the first time this proportion is lower than a preset real number threshold is selected as the number of data points contained in each phase space vector.

[0072] It should be noted that during the data preparation phase, the server receives the equiangularly spaced vibration signals output in step 202. To ensure the sufficiency of the phase space trajectory unfolding and the completeness of the dynamic characteristic analysis, the system continuously extracts angular domain signal sequences of 5 complete rotation cycles as the reconstruction processing object. Since each rotation after resampling contains a fixed 2000 sampling points, the currently extracted one-dimensional vibration numerical sequence contains a total of 10000 consecutive data points indexed by angle.

[0073] The system enters the phase space reconstruction parameter determination stage, which requires determining the subsequent specified interval (i.e., delay) in the phase space vector. The system gradually increases the delay step size of the data points and uses the mutual information algorithm to calculate the statistical dependence of the sequence under different delay values. In the calculation program, the system recorded that: when the delay is 1 sampling point, the redundancy of adjacent data is extremely high, and the statistical dependence value is close to the peak; as the delay gradually increases, the dependence value generally shows a non-linear decreasing trend; when the delay step reaches 15 sampling points, the curve of statistical dependence first shows a local minimum; when the delay continues to increase to 16 or more, the value rebounds. Based on this, the system determines the delay value of 15 corresponding to the first local minimum as the optimal subsequent specified interval, which ensures the independence of the coordinates of each dimension in the phase space.

[0074] The system determines the number of data points (i.e., the embedding dimension) contained in each phase space vector, and pre-sets a real threshold of 5% for the proportion of false nearest neighbors in the control parameters. The system employs a false nearest neighbor algorithm, gradually increasing the dimension of the vector starting with two data points, and simultaneously calculating the proportion of false nearest neighbors caused by projection overlap in the corresponding dimension: when the phase space vector contains two data points (two-dimensional space), the calculated proportion of false nearest neighbors is 42%; when increasing to three data points (three-dimensional space), the projection trajectory expands, and this proportion decreases to 12%; when further increasing to four data points (four-dimensional space), the proportion of false nearest neighbors drops significantly to 2.8%, which is the first time in the calculation sequence that it falls below the preset real threshold of 5%. Therefore, the system determines that each phase space vector should always contain four data points.

[0075] After specifying the key parameters of 15 intervals and 4 data points, the system executes a sliding window mechanism to traverse the vibration sequence of these 10,000 data points and construct a specific phase space point set.

[0076] Taking the construction of the first phase space vector as an example: the system takes the first data point in the sequence as the starting point (assuming its value is 0.45), and then extracts the 16th data point (assuming its value is 0.38), the 31st data point (assuming its value is 0.22), and the 46th data point (assuming its value is 0.15) in strict intervals of 15. Arranging these four values ​​in the extraction order forms a multi-dimensional coordinate system, which constitutes the first phase space vector in the four-dimensional phase space.

[0077] When constructing the second phase space vector, the system slides the starting index position backward by one data point, takes the second data point as the starting point, and extracts the 17th, 32nd and 47th data points in sequence according to the interval parameter of 15. These four values ​​are combined to form the second phase space vector in space.

[0078] The system traverses the entire one-dimensional sequence sequentially according to a fixed sliding rule. Since each vector requires four coordinate dimensions, it needs to traverse three specified intervals of length 15, resulting in a total index span of 45 data points for a single vector. Therefore, the last valid starting point for a vector in the original sequence is the 9955th data point. Using the 9955th data point as the final starting point, the system sequentially extracts the 9970th, 9985th, and 10000th data points, combining them to form the last phase space vector within the system set.

[0079] After a complete traversal of the entire sequence, the system transforms the original one-dimensional angular domain scalar sequence containing 10,000 data points into 9,955 phase space vectors containing four dimensions, while preserving the original dynamic characteristics. These 9,955 phase space vectors together constitute the phase space point set in the high-dimensional space.

[0080] 204. Calculate the recursion matrix based on the phase space point set, and extract the recursion rate and deterministic index from the recursion matrix. The recursion rate is used to characterize the density of recursion points, and the deterministic index is used to characterize the diagonal structure ratio.

[0081] Specifically, calculating the recursion matrix and extracting the recursion rate and deterministic index includes: calculating the Euclidean distance between every two phase space points in the phase space point set; setting a fixed distance threshold and comparing this distance threshold with the Euclidean distance between each pair of phase space points; if the Euclidean distance is less than the distance threshold, the recursion matrix position corresponding to the pair of phase space points is determined to be a recursive point, otherwise it is a non-recursive point, thus obtaining the recursion matrix; counting the number of all recursive points in the recursion matrix and dividing this number by the total number of positions in the recursion matrix to obtain the recursion rate; identifying all diagonals composed of consecutive recursive points from the recursion matrix, recording diagonals with a length reaching or exceeding a preset minimum length as valid diagonals, counting the total number of recursive points on all valid diagonals, and dividing this total number by the total number of all recursive points in the recursion matrix to obtain the deterministic index.

[0082] Furthermore, the distance threshold is determined as follows: the Euclidean distance between all pairs of phase space points in the phase space point set is calculated to obtain a distance set; the values ​​in this distance set are arranged in ascending order, and the distance value corresponding to the preset real percentile position after the arrangement is taken as the distance threshold. The calibration principle of this preset real percentile is: exhaustively optimizing using healthy samples from the first month after blade overhaul, selecting the real percentile (calibrated to 8% in this embodiment) that makes the baseline recursion rate stable within a reasonable dynamic range of 3% to 5%, and this percentile threshold is not updated after calibration.

[0083] It should be noted that during the Euclidean distance calculation and dynamic threshold determination stage, the system operation module receives the phase space point set output in step 203. This set contains 9955 phase space vectors, each containing 4 independent data points (i.e., in a four-dimensional state space). The system activates the distance calculation engine to calculate the spatial Euclidean distance between each pair of these 9955 phase space vectors. Due to the symmetry of matrix distances and the need to exclude zero-distance comparisons between vectors themselves, the system calculates a total of 49,546,035 unique Euclidean distance values ​​during the traversal process. These values ​​constitute the system state distance set.

[0084] The system uses a sorting algorithm to sort all 49,546,035 values ​​in the distance set in ascending order. It then reads the preset real percentile parameter (8%), determined during the initial calibration phase of the wind turbine blades, from the system's non-volatile memory. Within this large, sorted distance set, the system directly locates and extracts the specific distance value corresponding to the 8th percentile position using an index. This value, found to be 0.085, is used as a fixed distance threshold to determine whether recursion has occurred.

[0085] During the recursion matrix construction and recursion rate extraction phase, the system establishes a sparse matrix structure based on a coordinate list in memory (theoretically 9955 rows and 9955 columns), containing a total of 99,102,025 coordinate positions. The Euclidean distance between each pair of phase space points is logically compared with a distance threshold of 0.085. If the Euclidean distance between a pair of phase space points is less than 0.085, the system determines that the dynamic evolution of the system has undergone state regression at the physical time corresponding to these two points, and stores this coordinate pair as a recursive point in the sparse matrix; if the distance is greater than or equal to 0.085, it is determined to be a non-recursive point and skipped from storage.

[0086] After Boolean judgment and traversal of the entire matrix data, the system uses an accumulator to count the total number of recursive points (values ​​of 1) in the recursion matrix, which is 7,928,162. Dividing this number of recursive points by the total number of positions in the recursion matrix, 99,102,025, the system calculates the recursion rate (RR) to be 0.080. This value macroscopically characterizes the probability density of similar repetitions of blade vibration states in the phase space evolution trajectory.

[0087] In the deterministic index extraction stage, the system further identifies the diagonal topology of the obtained sparse recursive matrix. The system presets a minimum length parameter of 4 for the effective diagonal, which physically means that the system determines that a stable dynamic trajectory in phase space must maintain at least 4 consecutive sampling intervals in space. The system executes a scanning algorithm to identify all line segments in the matrix that consist of consecutive recursive points of 1, are parallel to the main diagonal, and have a length of 4 or more, and these are recorded as effective diagonals.

[0088] Matrix topological scanning revealed a total of 6,818,219 recursive points along all valid diagonals. Dividing this total number of valid recursive points by the total number of recursive points in the matrix (7,928,162) yielded a deterministic metric (DET) of 0.860. This metric represents the predictable deterministic components with a diagonal structure ratio during system operation.

[0089] To further analyze and attempt to correlate the aforementioned diagonal structure with the actual damage state, the system computation module introduces a structured recursive energy accumulation index (CEI) at this stage. This index characterizes the relative concentration of periodic kinetic energy induced by potential damage in the recursive state of a sparse system. Its internal computational model is expressed as follows: ; Where DET is the extracted deterministic index (0.860) and RR is the recursion rate (0.080). The total number of valid diagonals identified by the system (in this embodiment, the total number of valid diagonals is 12,000, based on the system's topology scan statistics). This represents the total number of recursive points (7,928,162) mentioned above. This index is simplified from the Shannon entropy model based on the recursive diagonal length distribution. Since blade fatigue cracks trigger specific periodic impacts, leading to an increase in determinism (DET), by introducing a natural logarithm term and dividing by the global density (RR), multiple bench tests have demonstrated that this combination effectively improves the signal-to-noise ratio of the feature response to micro-cracks. The specific derivation logic is as follows: DET 2 The weights of deterministic rule structures are amplified, and dividing by RR eliminates the global influence of the overall system state repetition density, making the index more sensitive to abnormal changes in local trajectories, especially for several terms. It can effectively compress the dynamic range caused by fluctuations in the number of effective diagonals (corresponding to stable periodic motion components). As a dimensionless empirical parameter built by the system, all its constituent variables are dimensionless ratios or pure numerical statistics. It is not only strictly self-consistent in mathematical dimensions, but also directly maps the relative relationship between local energy accumulation and global energy dissipation in a physical sense.

[0090] The system substitutes the specific statistical data from this embodiment into the model for calculation and solution: ; The system precisely derives the recursion rate (0.080), deterministic index (0.860), and energy accumulation index (0.014) from the recursive matrix operation. The recursion rate and deterministic index serve as the data base for the deviation fusion in step 205.

[0091] 205. Compare the recursion rate and deterministic index with the preset health baseline recursion rate and health baseline deterministic index respectively to obtain the comprehensive deviation. Based on the relationship between the comprehensive deviation and the preset threshold, output the evaluation result of the leaf health status.

[0092] Specifically, obtaining the comprehensive deviation and outputting the evaluation result includes: pre-storing the health baseline recursion rate and health baseline deterministic index calculated from stable operating data continuously collected during the initial stage of blade installation or after major overhaul. The health baseline recursion rate and health baseline deterministic index are dynamic baseline values ​​and remain locked within the current major overhaul maintenance cycle to truly reflect the deviation caused by accumulated damage; calculating the deviation of the current recursion rate relative to the health baseline recursion rate, and simultaneously calculating the deviation of the current deterministic index relative to the health baseline deterministic index; weighting and summing the two deviations using preset real number weights to obtain the comprehensive deviation; comparing the comprehensive deviation with a preset real number deviation threshold. If the comprehensive deviation is greater than the deviation threshold, an evaluation result indicating blade damage is output; otherwise, an evaluation result indicating blade health is output.

[0093] It should be noted that during the historical baseline data reading phase, the server main control program connects to the parameter configuration table in the local database and retrieves the health baseline parameters corresponding to the wind turbine blade. According to the preset conditions of the steps, these baseline parameters are dynamic baseline values ​​calculated by continuously collecting stable operating data and performing the same phase space reconstruction algorithm during the initial installation of the wind turbine blade or after a major overhaul. These values ​​represent the intrinsic dynamic characteristics of the blade in a physically intact state and are locked in the database, no longer automatically updated with daily operation. In this embodiment, the health baseline recursion rate read by the system from the database is set to 0.050, and the health baseline deterministic index is set to 0.900. Simultaneously, the system register receives the current blade operating characteristic data output in step 204: the currently calculated recursion rate is 0.080, and the currently calculated deterministic index is 0.860.

[0094] Entering the deviation calculation phase, the system's processing unit calculates the absolute mathematical difference between the currently extracted operational indicators and the health baseline indicators, using this as a measure of the relative deviation of the system state from its normal trajectory. For the recursion rate, which reflects the state repetition density, the system performs subtraction and absolute value operations, calculating the difference between the current value of 0.080 and the baseline value of 0.050, resulting in a recursion rate deviation of 0.030. For the deterministic indicator, which reflects trajectory stability, the system calculates the difference between the baseline value of 0.900 and the current value of 0.860, resulting in a deterministic indicator deviation of 0.040.

[0095] The system parameter configuration table pre-sets real-valued weight coefficients for these two characteristic indicators to balance the influence of different dynamic dimensions in the comprehensive health assessment. In this embodiment, the weight coefficient for the recursion rate deviation is set to 0.4, and the weight coefficient for the deterministic indicator deviation is set to 0.6. The system calculation unit multiplies the recursion rate deviation of 0.030 by its corresponding weight of 0.4, resulting in a product of 0.012; and multiplies the deterministic indicator deviation of 0.040 by its corresponding weight of 0.6, resulting in a product of 0.024. The system arithmetically adds these two calculated values ​​to obtain a comprehensive deviation of 0.036, representing the overall deviation of the current blade operating state.

[0096] During the evaluation and result output phase, the system calls the preset real-valued deviation threshold in the alarm logic module. In this embodiment, the health deviation safety threshold set for this type of blade is 0.025. The system compares the real-time calculated comprehensive deviation value with this safety threshold for judgment. The system's internal judgment logic framework and result mapping correspondence are shown in Table 2 below: Table 2 Less than or equal to the preset threshold 0 to 0.025 healthy Output blade health status codes and record routine operation logs. Greater than the preset threshold Greater than 0.025 Damage exists Output a damage alarm code for the blades, triggering the issuance of a diagnostic work order. The system logically checks the currently calculated overall deviation of 0.036 against the threshold conditions in the table above. Since the value of 0.036 is mathematically significantly greater than the preset safety threshold of 0.025, this condition is directly triggered and meets the second judgment logic in the table.

[0097] Based on this comparison result, the decision module of the assessment system arrives at a conclusion and outputs the assessment result indicating blade damage to the terminal interface of the wind turbine monitoring station via the network interface. Simultaneously, the underlying system triggers relevant maintenance alarm records and automatically generates a preliminary diagnostic work order, which is then sent to the maintenance team. At this point, the complete routine processing and assessment workflow, from underlying sensor signal acquisition, data normalization, and phase space feature extraction to status determination output, is completed.

[0098] 206. After obtaining the angular domain vibration signal, the angular domain vibration signal is phase aligned and averaged, that is, the angular domain vibration signals of multiple consecutive rotations are superimposed according to the same rotation angle position and the average value is taken to obtain the enhanced angular domain vibration signal, and the enhanced angular domain vibration signal is used for phase space reconstruction.

[0099] Furthermore, the phase alignment averaging process includes: selecting several rotation signals that meet preset operating conditions from the continuously acquired angular domain vibration signals. The preset operating conditions include wind speed, rotational speed, and pitch angle all being within the rated operating range, and the maximum amplitude change within each rotation being less than a preset amplitude threshold; aligning the selected rotation signals according to the same rotation angle index, and performing an arithmetic average on multiple values ​​at each rotation angle position to obtain the enhanced angular domain vibration signal; each rotation angle position in the enhanced angular domain vibration signal corresponds to an averaged value, and the averaged value is used for subsequent phase space reconstruction.

[0100] It should be noted that during the signal selection and rigorous operating condition determination phase, the server buffer receives the continuous multi-rotation angular domain vibration signal sequence obtained from step 202. As previously stated, each rotation in this sequence contains a fixed 2000 sampling points, and the rotation angle interval between adjacent sampling points is fixed at 0.18 degrees. Simultaneously, the system control interface synchronously reads the real-time operating environment and status parameters of the wind turbine main control system's SCADA system.

[0101] The system pre-sets stringent rated filtration conditions for this enhanced processing: requiring wind speed to be stable between 10 and 11 meters per second, spindle speed to be maintained between 14.5 and 15.5 revolutions per minute, and blade pitch angle to remain at 0 degrees without any movement. The system scanned the 10-second operation log for the current period, confirming that the current ambient wind speed was 10.5 meters per second, the spindle speed was stable at 15.0 revolutions per minute, and the blade pitch angle was locked at 0 degrees. The data fully met the preset rated steady-state operating range.

[0102] Based on the confirmation of compliance with operating conditions, the system continuously extracts five full-rotation signal segments from the continuous angular domain vibration signal stream for processing. To ensure that these five signals were not severely interfered with by occasional gusts or sudden aerodynamic loads during acquisition, the system extraction program performs a secondary verification of the maximum amplitude variation within each signal. The system calculates and extracts the maximum vibration amplitudes in the data sequence from the first to the fifth rotation, which are 0.88, 0.89, 0.87, 0.88, and 0.86, respectively. The comparator calculates that the range variation of the largest amplitude in these five signals is 0.03. The system sets a safety threshold of 0.05 for amplitude range variation; the current variation of 0.03 is less than the preset threshold, thus confirming that the five-rotation signal sequence meets all the steady-state preconditions for performing phase alignment processing.

[0103] During the phase alignment and superposition averaging stage, the system matrix operation unit performs point-by-point matrix alignment on the selected 5 one-dimensional angle domain signals. Since each signal has been strictly resampled at a rotation interval of 0.18 degrees and standardized to 2000 data points in the early stage, the system directly arranges and aligns these 5 sequences vertically in memory according to the rotation position index (i.e., from the position of the 1st data point to the position of the 2000th data point).

[0104] After alignment, the system calculates the arithmetic mean for each of the 2000 fixed corner positions, column by column. The specific calculation logic is as follows: the five independent vibration values ​​at the same corner position index are summed, and then the sum is divided by the accumulated number of revolutions, 5, to obtain the average expected value for that specific corner position.

[0105] Taking four representative corner points in the data sequence as examples, the specific alignment and arithmetic mean calculation process is shown in Table 3 below: Table 3 Point 1 (0.18 degrees) 0.45 0.42 0.48 0.44 0.46 0.45 Point 500 (90.0 degrees) 0.85 0.88 0.82 0.86 0.84 0.85 Point 1000 (180.0 degrees) 0.20 0.22 0.18 0.21 0.19 0.20 Point 1500 (270.0 degrees) 0.60 0.63 0.58 0.61 0.58 0.60 Following the vertical summation and averaging logic shown in the table above, the addition and division operations are performed one by one by iterating through all 2000 corner position indices using a loop.

[0106] After the above full-sequence traversal processing, the system merged, compressed, and converted the original 10,000 data points, which contained random dynamic fluctuations, into a new single enhanced sequence with a length of only 2,000 data points. Due to the use of the in-phase superposition mechanism, the signal-to-noise ratio of the periodic vibration characteristics related to the blade rotation frequency was significantly improved, and random background noise was effectively smoothed in the time-domain superposition.

[0107] The new sequence of 2000 data points obtained after arithmetic averaging is the enhanced angular domain vibration signal output in step 206. When the system detects that the wind turbine is in continuous steady-state operation, it automatically uses this high signal-to-noise ratio sequence as the standard input and seamlessly connects it to the subsequent step 203 to perform phase space reconstruction (because the input sequence is compressed to single-rotation features, the total number of phase space vectors obtained by reconstruction in step 203 is adaptively adjusted to 1955). This processing mechanism ensures that the evaluation process has higher noise robustness and result uniqueness under steady-state conditions.

[0108] The method for assessing the health status of wind turbine blades in this embodiment of the invention has been described above. The system for assessing the health status of wind turbine blades in this embodiment of the invention is described below. Please refer to [link / reference]. Figure 3An embodiment of the wind turbine blade health status assessment system of the present invention includes: an amplitude processing module 301, used to acquire the vibration signal of the wind turbine blade, and perform amplitude processing on the vibration signal to obtain a normalized signal; a resampling module 302, used to acquire the rotational speed signal of the wind turbine blade, and based on the rotational speed signal, resample the normalized signal from the time domain to the angle domain to obtain an angle domain vibration signal; a phase space reconstruction module 303, used to perform phase space reconstruction on the angle domain vibration signal to construct a phase space point set; a feature extraction module 304, used to calculate a recursion matrix based on the phase space point set, and extract the recursion rate and deterministic index from the recursion matrix; and a status assessment module 305, used to determine assessment feature values ​​based on the recursion rate, the deterministic index, and preset benchmark feature parameters, and output the health status assessment result of the wind turbine blade based on the assessment feature values.

[0109] above Figure 3 The wind turbine blade health status assessment system in this embodiment of the invention is described in detail from the perspective of modular functional entities. The wind turbine blade health status assessment device in this embodiment of the invention is described in detail from the perspective of hardware processing.

[0110] Figure 4 This is a schematic diagram of a wind turbine blade health status assessment device provided in an embodiment of the present invention. The device 400 may include: a processor 401, a receiver 402, a transmitter 403, and a memory 404. The receiver 402, transmitter 403, and memory 404 are respectively connected to the processor 401 via a bus. It should be noted that in some possible implementations, the processor 401 and the memory 404 may be integrated together.

[0111] The processor 401 includes one or more processing cores. The processor 401 executes the methods performed by the base station in the random access method provided in this application embodiment by running software programs and modules. The memory 404 can be used to store software programs and modules. Specifically, the memory 404 can store an operating system 4041 and at least one application module 4042 required for a function. The receiver 402 is used to receive communication data sent by other devices, and the transmitter 403 is used to send communication data to other devices.

[0112] The present invention also provides a wind turbine blade health status assessment device, which includes a memory and a processor. The memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor performs the steps of the wind turbine blade health status assessment method in the above embodiments.

[0113] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the steps of the wind turbine blade health status assessment method.

[0114] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0115] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0116] The above-described 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 the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for assessing the health status of wind turbine blades, characterized in that, Includes the following steps: The vibration signal of the wind turbine blade is acquired, and the amplitude of the vibration signal is processed to obtain a normalized signal. The rotational speed signal of the wind turbine blades is obtained, and based on the rotational speed signal, the normalized signal is resampled from the time domain to the angle domain to obtain the angle domain vibration signal; The phase space of the angular domain vibration signal is reconstructed to construct a phase space point set; The recursion matrix is ​​calculated based on the phase space point set, and the recursion rate and deterministic index are extracted from the recursion matrix, including: calculating the distance value between each pair of phase space points in the phase space point set; comparing the distance value with a distance threshold to determine the recursive points and non-recursive points in the recursion matrix; calculating the recursion rate based on the proportion of the number of recursive points in the recursion matrix; identifying effective diagonals with a length greater than or equal to a preset length in the recursion matrix, and obtaining the deterministic index based on the ratio of the number of recursive points on the effective diagonals to the total number of recursive points. It also includes calculating the recursive accumulated energy index (CEI): ； Where DET is the deterministic indicator and RR is the recursion rate. The total number of valid diagonals identified. This represents the total number of all recursive points. The evaluation feature values ​​are determined based on the recursion rate, the deterministic index, and the preset benchmark feature parameters, and the health status evaluation results of the wind turbine blades are output based on the evaluation feature values.

2. The method for assessing the health status of wind turbine blades according to claim 1, characterized in that, include: The cycle time information of the wind turbine blades is extracted based on the rotation speed signal; Based on a preset number of sampling points per cycle, the normalized signal corresponding to the cycle time information is interpolated and resampled to obtain an angular domain vibration signal with equal rotation intervals between adjacent sampling points.

3. The method for assessing the health status of wind turbine blades according to claim 1, characterized in that, Also includes: Select the angular domain vibration signals from multiple consecutive operating cycles that meet the preset operating conditions; The selected angular domain vibration signals are aligned according to the same rotation position and averaged to obtain the enhanced angular domain vibration signal, which is then used as the input for phase space reconstruction.

4. The method for assessing the health status of wind turbine blades according to claim 1, characterized in that, include: Extract time-series data from the angular domain vibration signal; Based on a preset delay time and embedding dimension, the time series data is reconstructed into multiple phase space vectors containing a fixed number of data points, and all the phase space vectors form a phase space point set.

5. The method for assessing the health status of wind turbine blades according to claim 4, characterized in that, The delay time is determined by calculating the local minimum of the statistical dependence as a function of the delay amount; The embedding dimension is determined by calculating the number of data points corresponding to a false nearest neighbor ratio that is lower than a preset threshold.

6. The method for assessing the health status of wind turbine blades according to claim 1, characterized in that, The distance threshold is determined in the following ways: Obtain the set of distances between all pairs of phase space points in the phase space point set; Extract the distance value corresponding to the preset real percentile position from the distance set as the distance threshold.

7. The method for assessing the health status of wind turbine blades according to claim 1, characterized in that, include: Calculate the first deviation of the recursion rate from the benchmark recursion rate, and the second deviation of the deterministic index from the benchmark deterministic index. The benchmark recursion rate and the benchmark deterministic index together constitute the benchmark feature parameter. The first deviation and the second deviation are weighted and summed to obtain the evaluation feature value; The evaluation feature value is compared with a preset deviation threshold. If the evaluation feature value is greater than the preset deviation threshold, an evaluation result indicating that the wind turbine blades are damaged is output.

8. A wind turbine blade health status assessment system, characterized in that, The wind turbine blade health status assessment system includes: An amplitude processing module is used to acquire the vibration signal of the wind turbine blades, perform amplitude processing on the vibration signal, and obtain a normalized signal. The resampling module is used to acquire the rotational speed signal of the wind turbine blades, and based on the rotational speed signal, resample the normalized signal from the time domain to the angle domain to obtain the angle domain vibration signal. The phase space reconstruction module is used to reconstruct the phase space of the angular domain vibration signal and construct a phase space point set. The feature extraction module is used to calculate a recursion matrix based on the phase space point set, and extract the recursion rate and deterministic index from the recursion matrix, including: calculating the distance value between each pair of phase space points in the phase space point set; comparing the distance value with a distance threshold to determine the recursive points and non-recursive points in the recursion matrix; calculating the recursion rate based on the proportion of the number of recursive points in the recursion matrix; identifying effective diagonals with a length greater than or equal to a preset length in the recursion matrix, and obtaining the deterministic index based on the ratio of the number of recursive points on the effective diagonals to the total number of recursive points. It also includes calculating the recursive accumulated energy index (CEI): ； Where DET is the deterministic indicator and RR is the recursion rate. The total number of valid diagonals identified. This represents the total number of all recursive points. The status assessment module is used to determine assessment feature values ​​based on the recursion rate, the deterministic index, and preset benchmark feature parameters, and output the health status assessment results of the wind turbine blades based on the assessment feature values.

Images