Blade real-time fatigue damage assessment method based on modified rain-flow counting

Abstract

The application belongs to the technical field of rotating machinery structure health monitoring and life prediction, and provides a kind of blade real-time fatigue damage evaluation method based on modified rain flow counting, the non-stationary degree of stress data stream is adaptively adjusted sliding time window and lengthens segmentation, discrete stress sequence is obtained by wavelet denoising and extreme compression;The residual error of the last time window is spliced with the current sequence, and the closed cycle is extracted by three-point rain flow counting and the incremental and full-life load spectrum matrix is updated;Through Goodman correction, the asymmetric cycle is equivalent, and the real-time cumulative damage degree is calculated recursively by combining S-N curve and continuous damage mechanics model;Based on the dynamic load evolution spectrum, the damage accumulation rate and the crack propagation acceleration characteristics, the remaining life is inversed, and the damage evolution rate is determined by the theoretical fatigue life and the current damage state.The application can eliminate counting error, improve evaluation accuracy, realize real-time accurate early warning and safe operation.

Description

Technical Field

[0001] This invention belongs to the field of rotating machinery structure health monitoring and life prediction technology, and provides a method for real-time fatigue damage assessment of blades based on modified rainflow counting. Background Technology

[0002] As core power-bearing components in rotating machinery such as steam turbines, aero-engines, and industrial gas turbines, blades operate under extreme conditions of high temperature, high speed, and alternating aerodynamic loads for extended periods. During actual service, blades not only bear enormous centrifugal tensile stress but also experience multiple alternating loads induced by airflow excitation, rotor imbalance, and unsteady flow fields. This complex stress environment easily triggers the initiation and propagation of micro-cracks within the material, ultimately leading to high-cycle fatigue fracture. Statistics show that fatigue failure is one of the main causes of rotating blade accidents. Therefore, real-time assessment of fatigue damage and prediction of remaining life throughout the blade's entire lifespan is of significant engineering importance for ensuring safe unit operation and optimizing maintenance strategies.

[0003] Currently, for fatigue life assessment under variable amplitude loads, the rainflow counting method has become the recognized standard in the field of fatigue analysis because it can accurately extract closed stress cycles in irregular load spectra and conforms to the stress-strain hysteresis loop characteristics of materials. However, traditional rainflow counting methods are usually based on complete, finite-length historical load data for post-processing offline. In online monitoring scenarios in industrial sites, sensors collect infinitely long real-time data streams. Embedded monitoring systems are limited by storage space and computing power and cannot save massive amounts of original time-domain waveforms. Therefore, a "segmented processing" method must be used for real-time calculation.

[0004] Existing real-time fatigue assessment methods primarily employ fixed-length or sliding time window mechanisms to truncate the data stream. However, this simple hard truncation mechanism has significant theoretical flaws: when a low-frequency stress cycle with a large amplitude spans two adjacent time windows, it is forcibly cut into two incomplete half-cycles or directly ignored. This not only disrupts the integrity of the load cycle but also leads to an underestimation of the stress amplitude, resulting in cumulative counting errors that bias the final fatigue damage assessment result towards the "dangerous side."

[0005] Furthermore, most existing online life assessments are based on Miner's linear cumulative damage theory, which assumes that the damage caused by each stress cycle is independent and linearly superimposed, neglecting the nonlinear effect of the load loading sequence on material damage. Under the complex and variable operating conditions of actual blades, this linear assumption often leads to a significant deviation between the predicted results and the actual life. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a real-time fatigue damage assessment method for blades based on corrected rainflow counting, which can eliminate counting errors, improve assessment accuracy, and achieve real-time accurate early warning and safe operation and maintenance.

[0007] The technical solution of the present invention includes: Wavelet denoising and extremum compression are performed on the acquired blade stress data stream to construct a discretized stress sequence that eliminates high-frequency noise interference and retains key load characteristics. Based on the non-stationarity of the real-time acquired blade stress data stream, the time span and sliding step size of the sliding time window are adaptively adjusted to perform variable-length segmentation on the stress data stream.

[0008] The three-point rainflow counting method is used to extract closed stress cycles within a time window and calculate the mean and amplitude of stress. The incremental load spectrum matrix and the total lifetime cumulative load spectrum matrix are updated synchronously. The stress extreme value sequence that is not closed in the current time window is stored in the residual buffer. This sequence is then combined with the discretized stress sequence of the next time window to form a combined calculation sequence. The above steps are repeated.

[0009] A stress correction mechanism is used to transform the asymmetric stress cycle with mean into an equivalent symmetric cycle. Combining the material SN curve and the continuous damage mechanics model, the non-equal weighted damage increment corresponding to each closed stress cycle is calculated, and the real-time cumulative damage degree is obtained through a recursive update strategy.

[0010] A dynamic load evolution spectrum with a time dimension is constructed by using the incremental load spectrum matrix and the full-life cumulative load spectrum matrix. The remaining safe service life of the blade is then inverted by combining the cumulative damage degree, damage accumulation rate and crack propagation acceleration characteristics. The damage evolution rate is calculated based on the theoretical fatigue life corresponding to the cycle and the current damage state.

[0011] Furthermore, the method for adaptively adjusting the sliding time window is as follows: Statistical characteristic indicators of the stress data within the current buffer are calculated in real time, including the coefficient of variation, local standard deviation, and mean; a preset threshold range is established for steady-state and transient operating conditions; when the statistical characteristic indicator exceeds the transient operating condition threshold, it is determined to be a high-dynamic-change zone, and the time window length is shortened and the sliding step size is reduced; when the statistical characteristic indicator is below the steady-state operating condition threshold, it is determined to be a stable zone, and the time window length is extended and the sliding step size is increased.

[0012] Furthermore, the wavelet denoising method is as follows: Symlets wavelet is selected as the basis function to perform multi-scale decomposition on the stress data within the time window to obtain low-frequency approximation coefficients and multi-layer high-frequency detail coefficients; Rigrsure adaptive threshold selection rule is used to calculate the optimal threshold for each layer, high-frequency noise is filtered out by soft threshold function, and a smooth stress signal with high signal-to-noise ratio is reconstructed by inverse wavelet transform.

[0013] Furthermore, the extreme value compression method is as follows: a first-order derivative sign recognition algorithm is used to locate local extreme points in the smooth stress signal, a hysteresis filtering mechanism is introduced, the absolute value of the difference between the current extreme value and the previous retained extreme value is calculated, if the absolute value of the difference is less than a set dead zone threshold, the extreme value is ignored, otherwise it is retained, and finally a simplified discrete sequence composed of timestamps and stress extreme values ​​is output.

[0014] Furthermore, the three-point rainflow counting method is as follows: Let three consecutive points in the combined calculation sequence be the starting point Y, the intermediate point X, and the ending point Z; define the variable range... , When satisfied When the starting point and the ending point are determined to form a closed stress cycle, the mean stress and amplitude of the cycle are calculated, the intermediate point and the ending point are removed, and the pointer is backtracked to continue to check the closure condition of the new adjacent point. After the traversal is completed, the remaining unclosed extreme points are stored as new residuals in the residual buffer.

[0015] Furthermore, the method for updating the load spectrum matrix is ​​as follows: based on the fatigue limit range of the blade material, the mean stress axis and amplitude axis are divided into several equally spaced intervals; the coordinate index of each closed stress cycle in the matrix grid is determined by a linear mapping algorithm; the built-in boundary constraint logic merges abnormally large stresses into the boundary interval or records them as overflow events; and the in-situ accumulation operation is performed to realize the dynamic update of the matrix.

[0016] Furthermore, the stress correction mechanism is the Goodman modified equation, as follows: in, For the first The equivalent total reversal stress amplitude for each cycle, For the first Stress amplitude of each cycle, For the first Average stress over one cycle This represents the tensile strength limit of the material.

[0017] Furthermore, the expression for the continuous damage mechanics model is as follows: in, For the first The total cumulative damage at any given time, ranging from [0,1], where 0 represents intact and 1 represents fracture; For the first The incremental damage generated per cycle The theoretical fatigue life under stress level, The cumulative damage is represented by time t.

[0018] Furthermore, the total lifetime cumulative load spectrum matrix is ​​obtained by merging the incremental load spectrum matrices of each time window, recording all stress cycles of the blade since it was put into operation.

[0019] Furthermore, a graded early warning is triggered when the remaining lifetime is lower than the safety warning line or the damage rate increases exponentially. The remaining lifetime prediction is based on a nonlinear inversion model of damage tolerance and rate gradient, and an acceleration factor is introduced to adapt to the accelerated crack propagation characteristics in the later stage of damage. The estimated time for the blade to reach the failure threshold is calculated by numerical integration.

[0020] The technical solution provided by this invention has the following advantages compared with the prior art: By employing an adaptive dynamic sliding time window to achieve variable-length segmentation of the stress data stream, combined with wavelet denoising and extreme value compression, efficient signal preprocessing is achieved, solving the problems of traditional fixed-window segmentation being unable to adapt to non-stationary stress data streams, data redundancy, and noise interference. Residual sequence propagation and corrected rainflow counting are used to concatenate the unclosed extreme values ​​of the previous time window with the current sequence for counting, completely eliminating the errors caused by traditional hard truncation that leads to the fragmentation of long-cycle stress cycles across windows and undercounting. Stress correction combined with a continuous damage mechanics model is used to calculate unequal-weighted damage increments, replacing Miner's linear accumulation rule, overcoming the evaluation bias caused by ignoring load order and nonlinear damage evolution. Remaining life inversion is performed based on the dynamic load evolution spectrum, damage accumulation rate, and crack propagation acceleration characteristics, achieving accurate full-cycle prediction and graded early warning. Ultimately, this achieves real-time, accurate, and distortion-free assessment of blade fatigue damage, effectively improving the operational safety and maintenance rationality of rotating machinery.

[0021] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart of a real-time fatigue damage assessment method for blades based on modified rainflow counts.

[0024] Figure 2 This is a flowchart of adaptive signal preprocessing based on a dynamic sliding time window.

[0025] Figure 3 This is a flowchart of the modified rainflow counting method based on residual sequence propagation. Detailed Implementation

[0026] The following detailed description of a specific embodiment of the present invention is provided in conjunction with the accompanying drawings. However, it should be understood that the scope of protection of the present invention is not limited to the specific embodiment.

[0027] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "axial," "radial," and "circumferential" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing the technical solution of this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0028] In the description of the embodiments of the present invention, unless otherwise stated, "a plurality of" means two or more.

[0029] like Figures 1 to 3 As shown, this invention provides a real-time fatigue damage assessment method for blades based on modified rainflow counting, comprising: Step 1: Perform wavelet denoising and extreme value compression on the acquired blade stress data stream to construct a discretized stress sequence that eliminates high-frequency noise interference and retains key load characteristics. Based on the non-stationarity of the real-time acquired blade stress data stream, adaptively adjust the time span and sliding step size of the sliding time window to perform variable-length segmentation on the stress data stream.

[0030] Specifically, an adaptive signal preprocessing strategy based on a dynamic sliding time window is used to perform variable length segmentation and peak-valley extreme value compression on the real-time acquired blade stress data stream, thereby constructing a discretized stress sequence that eliminates high-frequency noise interference and retains key load characteristics.

[0031] Unlike the fixed window length in traditional methods, such as Figure 2 As shown, this invention proposes a sliding time window mechanism with dynamic window length to discretize and slice continuous blade stress signals, i.e., the time span of the window. and sliding step size Adjustments are made in real time based on the degree of non-stationarity of the signal within the current buffer.

[0032] Specifically, at each sampling moment or within a preset detection period, the system calculates the statistical characteristic indicators of the data segment in the current buffer in real time, mainly including the coefficient of variation of the signal. Local standard deviation of current buffer data and mean : Threshold ranges corresponding to steady-state and transient operating conditions are preset. When the above statistical characteristic indicators exceed the preset transient threshold, the logic judgment signal is in the high dynamic change region. At this time, the time window length is shortened and the sliding step size is reduced. Conversely, when the statistical characteristic indicators are below the steady-state threshold, the logic judgment signal is in the stable region. The time window length is automatically extended and the sliding step size is increased to cover low-frequency stress fluctuations with longer periods, thereby significantly reducing the data processing frequency while ensuring the integrity of monitoring.

[0033] Adaptive adjustment depends on the relationship between the current system's coefficient of variation and preset steady-state and transient operating thresholds. The steady-state threshold is set as the upper limit of the 95% confidence interval of the statistical index fluctuation range under steady-state operation. If the measured index is lower than this value, the logic judgment signal is in the stable region. The transient threshold needs to refer to the index peak value during non-steady-state processes such as unit start-up and shutdown, peak shaving, or aerodynamic stall. When the index exceeds this threshold, the system immediately determines that it is in a high dynamic change region.

[0034] After determining the optimal time window parameters for the current moment and extracting the corresponding data segment, cascaded signal cleaning and compression operations are then performed within that window.

[0035] First, an adaptive wavelet thresholding denoising algorithm is used to perform multi-scale decomposition on the original data within the window. The Symlets series wavelets are selected as the basis functions to decompose the discrete sequences within the current time window. Decomposed into low-frequency approximation coefficients and High-frequency detail factor : High-frequency noise caused by electromagnetic interference or test noise is filtered out by a soft threshold function, and a true stress waveform with a high signal-to-noise ratio is reconstructed.

[0036] in, For the first The threshold of the layer.

[0037] The system first uses Symlets wavelets as basis functions to perform multi-scale decomposition on the raw stress data within the current dynamic time window, breaking it down into low-frequency approximation coefficients representing the signal backbone and J-layer high-frequency detail coefficients containing a wealth of details. Then, for each layer of high-frequency detail coefficients, the system calls a soft thresholding function for filtering. Based on the Rigrsure adaptive threshold selection rule, the optimal threshold is automatically calculated for each layer of high-frequency coefficients. This formula forces coefficients with moduli less than a threshold (determined as noise) to zero, while coefficients exceeding the threshold are retained and shrunk. This removes white noise while accurately preserving the specific high-frequency characteristics of the blade excited by weak cracks. Finally, the system integrates the unmodified low-frequency approximation coefficients with the high-frequency detail coefficients of each layer after soft thresholding, and performs signal reconstruction through inverse wavelet transform. The resulting "new signal" is the smoothed stress signal after noise reduction. .

[0038] This invention employs the Rigrsure adaptive threshold selection rule, automatically calculating the optimal threshold based on the noise level of each layer. This removes white noise while preserving the specific high-frequency characteristics excited by weak cracks in the blade, utilizing the processed coefficients... Perform inverse wavelet transform (IDWT) to obtain the denoised smooth stress signal. .

[0039] Based on this, in order to meet the input requirements of subsequent rainflow counting, the denoised and smoothed waveform is further processed. Perform peak and trough extremum extraction, locate all local extrema points using a first-order derivative sign recognition algorithm, and apply this to the denoised dynamic stress signal. Perform a first-order difference scan. Define the time step. The condition for a data point to be an extremum is that the sign of its derivative changes (after taking the derivative of the denoised signal, the sign of the derivative changes from positive to negative):

[0040] To further prevent minor fluctuations from causing invalid counting cycles, a hysteresis filtering mechanism is introduced to eliminate counts with amplitude differences less than a set dead zone threshold. The algorithm logic for handling minute fluctuations is as follows: Calculate the current extreme value Compared with the previous retained extreme value The absolute value of the difference; like If the current extreme value is not considered sufficient to constitute a damage cycle, then the current extreme value is ignored. like If so, then the extreme point is retained.

[0041] After the above processing, the original tens of thousands of sampling points per second are compressed into tens to hundreds of key extreme points per second, and finally output a simplified discrete sequence consisting of timestamps and stress extremes, which is also a simplified extreme value sequence (time t + stress v, each stress extreme point has a timestamp corresponding to the time).

[0042] .

[0043] Step 2: Use the three-point rainflow counting method to extract closed stress cycles within the time window and calculate the mean and amplitude of stress. Simultaneously update the incremental load spectrum matrix and the total lifetime cumulative load spectrum matrix. Store the stress extreme value sequence of the current time window that is not closed through the residual buffer. Combine it with the discretized stress sequence of the next time window to form a combined calculation sequence. Repeat the above steps.

[0044] Specifically, the modified rainflow counting algorithm based on residual sequence propagation eliminates counting errors caused by time window truncation and accurately calculates the stress amplitude and mean matrix of long-period loads spanning different time periods in real time.

[0045] During online monitoring, the system cannot acquire the entire lifecycle load history at once, and the data stream must be segmented. However, directly counting individual time windows will forcibly truncate the large-amplitude low-frequency stress cycles, resulting in invalid half-cycle counts or omissions by the algorithm.

[0046] To address this, the present invention constructs a dedicated residual buffer. Its core function is to store the sequence of stress extrema that are not yet closed at the current moment, specifically as follows: Figure 3 As shown. When the system receives the standardized extreme value sequence within the current time window output from step 1. Instead of immediately counting the results, the residual sequence left over from the previous time window is first retrieved from the residual buffer. Use it as a boundary condition to splice it into the current new sequence. At the starting point, a combined computation sequence containing historical unclosed features and current new features is constructed. .

[0047] This invention employs an improved three-point rainflow counting method to scan the combined sequence, assuming the combined sequence... The three consecutive points in the middle are (starting point), (Midpoint) (End point). Define range variation: , .

[0048] If the following geometric conditions are met: Then determine and This constitutes a closed stress cycle. Once closure is determined, the stress parameters of the cycle are immediately calculated:

[0049] Subsequently, from the sequence Middle rejection point and Connect the previous point to the point The pointer then backtracks to check if new adjacent points satisfy the closure condition. When combining sequences... When all points have been traversed and no more closed loops satisfying the above conditions can be extracted, the remaining points constitute the new residual. .

[0050] When the online corrected rainflow counting core algorithm successfully extracts a closed stress hysteresis loop from the combined sequence and calculates its determined mean stress... With stress amplitude Then, the update process of the two-dimensional load spectrum matrix is ​​initiated. Two two-dimensional matrices with the same structure but different functions are constructed: one is the "incremental load spectrum matrix", which is used to temporarily store the stress cycles generated in the current time window and serve as the direct input for the real-time damage calculation in step 3; the other is the "full-life cumulative load spectrum matrix", which is used to record all historical loads experienced by the blade since its service, serving as a database for remaining life prediction.

[0051] The update process begins with a discretization binning operation. Based on the preset fatigue limit range of the blade material, the mean stress axis and amplitude axis are divided into several equally spaced intervals. For each newly extracted stress cycle, its specific coordinate index in the matrix grid is calculated using a linear mapping algorithm.

[0052] First, based on the blade's design limits and historical load data, the physical ranges for two key dimensions are pre-defined: mean stress. [ , — Row index of the corresponding matrix, stress amplitude [ , ] — The column index of the corresponding matrix. Assume the matrix is ​​divided into... The grid, which is the target space for linear mapping, is used to extract stress cycles ( , The system calculates its floating-point position in the matrix using the following linear proportional relationship and rounds down to obtain its integer index:

[0053] Row index calculation: Column index calculation: During this process, a built-in boundary constraint logic is implemented. If the detected abnormally large stress exceeds the preset mesh range, it is forcibly merged into the boundary interval or recorded as an overflow event to prevent the system from crashing due to array out-of-bounds errors. After determining the coordinates, an in-situ accumulation operation is performed to obtain a dynamically updated two-dimensional load spectrum matrix.

[0054] Step 3: The asymmetric stress cycle with mean value is transformed into an equivalent symmetric cycle using a stress correction mechanism. Combining the material SN curve and the continuous damage mechanics model, the non-equal weighted damage increment corresponding to each closed stress cycle is calculated. The real-time cumulative damage degree is obtained through a recursive update strategy.

[0055] Specifically, a nonlinear fatigue accumulation assessment model based on continuous damage mechanics maps the real-time extracted stress cycles into non-weighted damage increments, quantifying the dynamic influence of load order on the cumulative damage of blade materials under complex variable working conditions.

[0056] To address the mean stress experienced by the blades under high-speed rotation, this step introduces an equivalent total reversal stress correction mechanism. The Goodman correction equation is used to transform the asymmetric cycle with mean stress into an equivalent zero-mean symmetric cycle, eliminating the influence of mean stress on the SN curve baseline and ensuring comparability of stress data under different operating conditions.

[0057] in, For the first The equivalent total reversal stress amplitude for each cycle, For the first Stress amplitude of each cycle, For the first Average stress over one cycle This represents the tensile strength limit of the material.

[0058] Based on the material's SN curve, using the Basquin equation, the number of cycles required to withstand the current equivalent stress (symmetric stress cycles are also calculated, and since their mean stress is 0, the equivalent total reversal stress amplitude is their own stress amplitude) until failure is calculated: in, To be at stress level The theoretical fatigue life is as follows. The fatigue strength index of the material. This is the material fatigue constant.

[0059] This embodiment employs a simplified differential model based on Chaboche's continuum damage mechanics. This model treats the blade material as a degenerate continuous medium, whose effective load-bearing area decreases with increasing damage. The rate of damage evolution depends not only on lifetime. It also depends on the current state of damage. The formula is:

[0060] in, For the first The total cumulative damage at any given time, ranging from [0,1], where 0 represents intact and 1 represents fracture; For the first The damage increment generated by each cycle.

[0061] A recursive update strategy is adopted to calculate the increment. By superimposing it onto the total damage variable, the cumulative damage degree can be dynamically updated, and the accurate damage status at the current moment can be output in real time without having to trace back the entire historical load spectrum.

[0062] Step 4: Construct a dynamic load evolution spectrum with time dimension by using the incremental load spectrum matrix and the full-life cumulative load spectrum matrix. Combine the cumulative damage degree, damage accumulation rate and crack propagation acceleration characteristics to invert the remaining safe service life of the blade. The damage evolution rate is calculated based on the theoretical fatigue life corresponding to the cycle and the current damage state.

[0063] Specifically, the dynamic load spectrum evolution and remaining life prediction system based on full life cycle data updates the two-dimensional rainflow load spectrum matrix containing historical accumulated information in real time, and inverts the remaining safe service life of the blade in the current service state.

[0064] First, in terms of load spectrum evolution and update, it does not only record a single total matrix, but establishes a "dynamic evolutionary spectrum" with a time dimension.

[0065] Specifically, a cumulative load spectrum is maintained throughout the entire life cycle. This spectrum is the sum of all stress cycles experienced by the blade since its commissioning, recording every start-up and shutdown, every steady-state operation, and every transient impact. Whenever step 2 completes the processing of a time window, the current incremental load matrix is ​​merged into the full life cycle spectrum.

[0066] in, This is the full-lifetime principal load spectrum matrix, recorded from... All cycles to date; This is the increment matrix for the current time window (from step 2).

[0067] More importantly, in order to capture the changing trend of damage rate, a "sliding window trend spectrum" mechanism was introduced, which independently stores the load distribution within the most recent N running hours (e.g., the most recent 100 hours).

[0068] in, This is the sliding window trend spectrum matrix at the current moment. This is the trend spectrum matrix at the previous sampling time. For an expired data matrix that exceeds the sliding window time span, This is the newly generated incremental load spectrum matrix for the current time window.

[0069] By comparing the statistical characteristics of the "recent trend spectrum" and the "historical baseline spectrum," it is possible to accurately identify whether the blade exhibits aerodynamic performance degradation or stress response deterioration due to changes in structural damping. The centroid coordinates of the load spectrum are calculated. Monitor its drift over time:

[0070] like The significant increase over time indicates that the blade vibration level deteriorates under specific operating conditions.

[0071] Secondly, regarding the prediction of remaining lifetime, this invention employs a nonlinear inversion model based on damage tolerance and rate gradient to monitor the current cumulative damage level in real time. This value is obtained recursively from step 3. The system defines a failure threshold. The calculation of RUL depends not only on the current remaining damage margin. It depends more on the current rate of damage accumulation.

[0072] It is the time interval for calculating cumulative damage; The first formula quantifies how fast blade damage evolves over time, providing a velocity parameter for prediction; The second formula, based on the current level of damage and the rate of accumulation, inverts the estimated time required for the blade to reach the failure threshold.

[0073] Considering the accelerated crack propagation characteristics in the later stages of damage, an acceleration factor is introduced. ( ): In the formula, The corrected remaining lifetime. It is an acceleration factor.

[0074] Finally, the estimated time required for the blade to reach the failure threshold under the current degradation trend is calculated by numerical integration.

[0075] When the predicted remaining life is lower than the set safety warning line (e.g., less than 500 hours remaining life), or when an exponential increase in the damage rate is detected (indicating that the crack may have entered the unstable propagation stage), a graded early warning signal is immediately triggered to guide maintenance personnel to conduct targeted borehole inspections or blade replacement.

[0076] It should be noted that any parts not disclosed or specifically described in this invention are existing technology or conventional configurations, and their specific structures and working principles will not be elaborated further. In this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0077] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. It can be applied to various fields suitable for the present invention. Other modifications can be readily implemented by those skilled in the art. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and examples shown and described herein.

Claims

1. A method for real-time fatigue damage assessment of blades based on modified rainflow counting, characterized in that, Includes the following steps: Wavelet denoising and extreme value compression are performed on the acquired blade stress data stream to construct a discretized stress sequence that eliminates high-frequency noise interference and retains key load characteristics. Based on the non-stationarity of the real-time acquired blade stress data stream, the time span and sliding step size of the sliding time window are adaptively adjusted to perform variable-length segmentation on the stress data stream. The three-point rainflow counting method is used to extract closed stress cycles within a time window and calculate the mean and amplitude of stress. The incremental load spectrum matrix and the total lifetime cumulative load spectrum matrix are updated synchronously. The stress extreme value sequence that is not closed in the current time window is stored in the residual buffer. It is then combined with the discretized stress sequence of the next time window to form a combined calculation sequence. The above steps are repeated. A stress correction mechanism is used to transform the asymmetric stress cycle with mean into an equivalent symmetric cycle. Combining the material SN curve and the continuous damage mechanics model, the non-weighted damage increment corresponding to each closed stress cycle is calculated, and the real-time cumulative damage degree is obtained through a recursive update strategy. A dynamic load evolution spectrum with a time dimension is constructed by using the incremental load spectrum matrix and the full-life cumulative load spectrum matrix. The remaining safe service life of the blade is then inverted by combining the cumulative damage degree, damage accumulation rate and crack propagation acceleration characteristics. The damage evolution rate is calculated based on the theoretical fatigue life corresponding to the cycle and the current damage state.

2. The method for real-time fatigue damage assessment of blades based on modified rainflow counting according to claim 1, characterized in that, The method for adaptively adjusting the sliding time window is as follows: Calculate the statistical characteristic indicators of the stress data in the current buffer in real time. The statistical characteristic indicators include the coefficient of variation, local standard deviation and mean. Preset threshold ranges for steady-state and transient operating conditions; When the statistical characteristic index exceeds the transient operating condition threshold, it is determined to be a high dynamic change zone, and the time window length is shortened and the sliding step size is reduced. When the statistical characteristic index is lower than the steady-state operating condition threshold, it is determined to be in a stable region, and the time window length is extended and the sliding step size is increased.

3. The method for real-time fatigue damage assessment of blades based on modified rainflow counting according to claim 1, characterized in that, The wavelet denoising method is as follows: Symlets wavelet is selected as the basis function to perform multi-scale decomposition on stress data within the time window, and low-frequency approximation coefficients and multi-layer high-frequency detail coefficients are obtained. The Rigrsure adaptive threshold selection rule is used to calculate the optimal threshold for each layer. High-frequency noise is filtered out by a soft threshold function, and a smooth stress signal with high signal-to-noise ratio is reconstructed by inverse wavelet transform.

4. The method for real-time fatigue damage assessment of blades based on modified rainflow counting according to claim 3, characterized in that, The extreme value compression method is as follows: The first derivative sign recognition algorithm is used to locate local extreme points in the smooth stress signal. A hysteresis filtering mechanism is introduced to calculate the absolute value of the difference between the current extreme value and the previous retained extreme value. If the absolute value of the difference is less than the set dead zone threshold, the extreme value is ignored; otherwise, it is retained. The final output is a simplified discrete sequence composed of timestamps and stress extreme values.

5. The method for real-time fatigue damage assessment of blades based on modified rainflow counting according to claim 1, characterized in that, The three-point rainflow counting method is as follows: Let the three consecutive points in the combined calculation sequence be the starting point Y, the intermediate point X, and the ending point Z; Define the range , ; When satisfied When the starting point and the ending point are determined to form a closed stress cycle, the mean stress and amplitude of the cycle are calculated, the intermediate point and the ending point are removed, and the pointer is backtracked to continue to check the closure condition of the new adjacent point. After the traversal is completed, the remaining unclosed extreme points are stored as new residuals in the residual buffer.

6. The method for real-time fatigue damage assessment of blades based on modified rainflow counting according to claim 1, characterized in that, The method for updating the load spectrum matrix is ​​as follows: Based on the fatigue limit range of the blade material, the mean stress axis and amplitude axis are divided into several equally spaced intervals; The coordinate index of each closed stress cycle in the matrix mesh is determined by a linear mapping algorithm. The built-in boundary constraint logic merges abnormally large stresses into the boundary interval or records them as overflow events, and performs in-situ accumulation operation to realize dynamic matrix update.

7. The method for real-time fatigue damage assessment of blades based on modified rainflow counting according to claim 1, characterized in that, The stress correction mechanism is the Goodman correction equation, which is as follows: in, For the first The equivalent total reversal stress amplitude for each cycle, For the first Stress amplitude of each cycle, For the first Average stress over one cycle This represents the tensile strength limit of the material.

8. The method for real-time fatigue damage assessment of blades based on modified rainflow counting according to claim 7, characterized in that, The expression for the continuous damage mechanics model is as follows: Among them, is the first The total cumulative damage at any given time, ranging from [0,1], where 0 represents intact and 1 represents fracture; For the first The incremental damage generated per cycle To be at stress level The theoretical fatigue life is as follows. The cumulative damage is represented by time t.

9. The method for real-time fatigue damage assessment of blades based on modified rainflow counting according to claim 1, characterized in that: The total lifetime cumulative load spectrum matrix is ​​obtained by merging the incremental load spectrum matrices of each time window, recording all stress cycles of the blade since commissioning. By storing the load distribution over the most recent N operating hours using a sliding window trend spectrum, and by comparing the statistical characteristics with historical benchmark spectra and monitoring centroid coordinate drift, the deterioration of blade stress response can be identified.

10. The method for real-time fatigue damage assessment of blades based on modified rainflow counting according to claim 1, characterized in that, A graded warning is triggered when the remaining lifespan is below the safety threshold or the damage rate increases exponentially. The remaining life prediction is based on a nonlinear inversion model of damage tolerance and rate gradient, and introduces an acceleration factor to adapt to the accelerated crack propagation characteristics in the later stage of damage. The estimated time for the blade to reach the failure threshold is calculated by numerical integration.

Images