A residual stress stripping method based on sensor wavelength signal processing
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ZHIXING S&T
- Filing Date
- 2026-03-19
- Publication Date
- 2026-06-19
Smart Images
Figure CN122237463A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of fiber optic grating sensor technology, and in particular to a residual stress stripping method based on sensor wavelength signal processing. Background Technology
[0002] Fiber Bragg grating strain sensors, as a high-precision and highly reliable optical measurement method, play a crucial role in the health monitoring of engineering structures such as aerospace, bridge construction, and pressure vessels. They can convert minute structural deformations into clear optical signals, enabling long-term, real-time strain tracking, which is of great significance for ensuring the safe operation of large components.
[0003] Most current strain measurement methods struggle to accurately separate the applied load from the initial internal stress state when dealing with residual stress within a structure. While traditional resistance strain gauges or some fiber optic sensors can capture the total strain, residual stresses can overlap with subsequent working stresses after welding, heat treatment, or prolonged service, causing the measurement results to deviate from the true load level. This deviation is particularly pronounced in complex components, as residual stresses are often unevenly distributed and have complex directions, making it difficult for conventional methods to directly identify their independent contributions.
[0004] After a fiber Bragg grating sensor is attached to a component surface, its Bragg wavelength is simultaneously affected by both the component's current strain and residual stress. Residual stress, as an initial state "pre-stored" within the material, means that the sensor carries the component's inherent strain substrate from the moment of installation. If this substrate cannot be effectively removed, the wavelength drift measured in any subsequent loading experiments cannot clearly distinguish between external forces and those already present within the component. Especially when using strain release methods to cut or drill to release stress, the wavelength change includes both the released residual stress and the strain information "frozen" at the time of initial attachment. The superposition of these two factors leads to a systematic deviation in the calculated residual stress value.
[0005] How to completely separate the current residual stress state of the component from the strain reference of the fiber Bragg grating sensor after installation, so that the zero point of the subsequent measurement of the sensor truly corresponds to the ideal state without residual stress, and thus achieve accurate quantification of residual stress, has become the most critical technical challenge in the accurate measurement of structural residual stress. Summary of the Invention
[0006] This invention provides a residual stress stripping method based on sensor wavelength signal processing, mainly comprising: By collecting Bragg wavelength data at the moment of sensor installation, the initial wavelength offset value is extracted from the data to obtain the reference offset caused by residual stress. Based on the reference offset, a compensation algorithm is used to adjust the subsequent measurement wavelength signal to determine the correction wavelength for residual peeling stress. If the correction wavelength exceeds the preset threshold, the historical measurement sequence is fused through the signal processing stage to determine the contribution of the external load and obtain the pure load response. Wavelength drift characteristics are obtained from the pure load response, and noise interference is removed by a filtering algorithm to obtain a clear strain tracking sequence; Based on the clear strain tracking sequence, the independent contribution of residual stress is determined, and the reference zero point is updated through an iterative optimization algorithm to obtain the reference value of the state without residual stress. If the reference value matches the real-time wavelength signal, the multi-point sensor inputs are integrated through the data fusion process to determine the overall structural stress distribution and obtain accurate quantification results. The uneven distribution features are extracted from the precise quantification results, and the complex part of the direction is calibrated by the compensation algorithm to determine the final residual stress stripping output.
[0007] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses a residual stress removal method based on sensor wavelength signal processing. Addressing the interference of residual stress on measurement accuracy during sensor installation, it integrates initial wavelength offset extraction, baseline zero-point update, and multi-point stress distribution analysis into a logically coherent solution. This invention acquires Bragg wavelength data at the installation moment, extracts the initial offset value and establishes a baseline, and uses a compensation algorithm to correct subsequent measurement signals, thus removing the influence of residual stress. Simultaneously, it utilizes a filtering algorithm to remove noise, combines historical sequences to determine the contribution of external loads, and obtains a clear strain tracking sequence. Through iterative optimization and updating of the baseline zero point, a reference value without residual stress is obtained. Furthermore, by fusing multi-point sensor data, the overall structural stress distribution is accurately quantified, and finally, the non-uniform characteristics in complex directions are calibrated, achieving residual stress removal output. This invention significantly improves the accuracy and reliability of stress measurement, providing efficient and accurate technical support for stress analysis of complex structures. Attached Figure Description
[0008] Figure 1 This is a flowchart of a residual stress stripping method based on sensor wavelength signal processing according to the present invention.
[0009] Figure 2 This is a schematic diagram of a residual stress stripping method based on sensor wavelength signal processing according to the present invention.
[0010] Figure 3 This is another schematic diagram of a residual stress stripping method based on sensor wavelength signal processing according to the present invention. Detailed Implementation
[0011] The technical solutions of the embodiments of the present invention will be clearly and thoroughly described below with reference to the accompanying drawings. The described embodiments are merely some embodiments of the present invention.
[0012] like Figures 1-3 This embodiment of a residual stress stripping method based on sensor wavelength signal processing may specifically include: S101. By collecting Bragg wavelength data at the time of sensor installation, the initial wavelength offset value is extracted from the data to obtain the reference offset caused by residual stress.
[0013] By acquiring Bragg wavelength-related data using sensors at the installation moment, raw data is collected, resulting in a preliminary wavelength record. Based on this preliminary wavelength record, data processing methods are used to separate wavelength data relevant to the installation moment, determining a wavelength reference value independent of the environment. For the separated wavelength reference values, initial offset information is extracted, and it is determined whether the initial offset exceeds a preset threshold range. If it does, it is marked as abnormal wavelength data. Through further processing of the abnormal wavelength data, the specific impact of residual stress on wavelength changes is analyzed, obtaining the offset caused by residual stress. Based on the offset caused by residual stress and the pattern of wavelength change, the specific value of the reference offset is calculated, determining the final reference offset result. After obtaining the final reference offset result, a pre-established stress influence model is used to analyze the correspondence between stress influence and reference offset, obtaining a quantified value of residual stress. By storing the quantified value of residual stress, a structured stress influence record is generated, completing the entire data processing flow.
[0014] First, after the fiber optic grating sensor is fixedly installed on or inside the component and cured for 24 hours, a high-precision wavelength demodulator is used to continuously acquire 30 seconds of Bragg center wavelength data at a sampling rate of 100Hz, obtaining the original wavelength sequence λ1, λ2, ..., λ3000. This sequence is then filtered using a sliding window mid-range filter (window width 51 points) to remove random noise. The arithmetic mean of the sequence is then calculated as the stable wavelength λinitial after installation; for example, the measured λinitial is 1550.342 nm. Next, the reference wavelength λref of the same batch of gratings before installation is acquired under a constant temperature and humidity stress-free standard environment (temperature 20.0℃, relative humidity 50%); for example, the calibration value is 1549.876 nm. Then, the initial wavelength offset Δλ0 = λinitial - λref = 1550.342 - 1549.876 = 0.466 nm is calculated. Based on the grating strain sensitivity coefficient Kε = 1.2 pm / με and the temperature sensitivity KT = 10.3... The temperature difference between the temperature recorded by the temperature sensor at the installation time, Tinstall=22.4℃ and the calibration temperature Tref=20.0℃, is ΔT=2.4℃. The wavelength drift caused by temperature is calculated as ΔλT=KT×ΔT=10.3×2.4=24.72 pm=0.02472 nm. Finally, the temperature effect is subtracted to obtain the reference offset caused by residual stress, Δλstress=Δλ0-ΔλT=0.466-0.025=0.441 nm. Then, using the formula εresidual=Δλstress / (Kε×λref)≈0.441 / (1.2×1549.876)≈237 με, this residual strain is used as the reference offset for strain calculation in subsequent long-term monitoring, thereby realizing fully automatic extraction and compensation of the initial value of residual stress.
[0015] S102. Based on the reference offset, the subsequent measurement wavelength signal is adjusted using a compensation algorithm to determine the correction wavelength for the residual stress after peeling.
[0016] By analyzing the reference offset, a pre-established compensation algorithm is used to initially correct the wavelength signals acquired in subsequent measurements, resulting in adjusted signal data. Based on the adjusted signal data, a data calibration method is applied to separate the wavelength deviations corresponding to stress interference, determining the intermediate results after correction. After obtaining the intermediate results, signal adjustment techniques are used to further optimize the accuracy of the wavelength signals, resulting in an optimized wavelength dataset, based on the specific distribution of the wavelength deviations. From the optimized wavelength dataset, residual components related to stress interference are extracted, and combined with the logic for stripping away influences, it is determined whether there are abnormal deviations. If abnormal deviations exist, the abnormal parts are calibrated a second time to obtain corrected wavelength values. For the corrected wavelength values, a data calibration verification mechanism is applied to analyze their matching degree with the measurement data, determining the final corrected wavelength result. By processing the final corrected wavelength result and combining it with real-time data from subsequent measurements, the signal adjustment parameters are continuously updated to obtain a dynamically optimized wavelength signal output. After obtaining the dynamically optimized wavelength signal output, a historical data calibration record is constructed based on the accumulation of long-term measurement data to determine a stable correction scheme for the wavelength signal.
[0017] After the sensor installation and curing are completed, the system automatically triggers the wavelength acquisition program. Using a high-precision fiber Bragg grating demodulation device, it continuously acquires 60 seconds of Bragg center wavelength data at a sampling frequency of 200Hz, forming an original sequence λ1 to λ12000 containing 12000 sampling points. First, an exponentially weighted moving average filter (smoothing factor α = 0.05) is applied to the sequence for noise reduction to effectively suppress high-frequency random disturbances. The weighted average of the entire sequence after filtering is taken as the center wavelength λstable after installation stabilization. For example, the actual calculated λstable is 1551.128 nm. Then, the reference wavelength λbase under stress-free reference conditions (temperature 20.0℃, humidity 55%) is directly retrieved from the pre-calibration database of the same sensor production batch. The calibration record shows λbase as 1550.615 nm. The system immediately performs the offset calculation Δλinitial = λstable - λbase = 1551.128 - 1550.615 = 0.513. nm; simultaneously, the real-time temperature Tcurrent = 23.7℃ uploaded by the integrated temperature probe at the moment of installation is read, and the difference between it and the calibrated reference temperature ΔTinstall = 23.7 - 20.0 = 3.7℃ is calculated. Combined with the preset temperature sensitivity coefficient KT = 10.8 pm / ℃, the temperature drift component Δλtemp = 10.8 × 3.7 = 39.96 pm, or 0.03996 nm, is calculated; then, through automatic compensation logic, the wavelength shift dominated by residual stress Δλresidual = Δλinitial - Δλtemp = 0.513 - 0.040 = 0.473 nm is obtained; then, using the strain sensitivity coefficient Kε = 1.18 pm / με and the reference wavelength λbase, the residual strain is estimated as ε0 = Δλresidual / (Kε × λbase) ≈ 0.473 / (1.18 × 1550.615) ≈ 258 με; This residual strain ε0 is automatically stored in the monitoring database as a benchmark correction value. In all subsequent strain calculations, this initial offset is deducted in real time, realizing continuous automatic stripping and correction of the influence of residual stress.
[0018] S103. If the correction wavelength exceeds the preset threshold, the historical measurement sequence is fused through the signal processing stage to determine the contribution of the external load and obtain the pure load response.
[0019] The current measurement wavelength signal is acquired, and a preliminary adjustment is performed on the signal using a pre-established compensation model to obtain a pre-adjusted wavelength sequence. Based on this pre-adjusted sequence, a sequence decomposition method is used to separate the interference component introduced by the external load, determining the load response components. Based on these components, the influence of the external load is removed from the pre-adjusted wavelength sequence, resulting in a residual wavelength sequence. For this residual sequence, morphological matching technology is used to compare it with a preset normal morphology to identify the locations of anomalies, obtaining a wavelength sequence marked with these anomalies. If the number of anomalies in the marked anomaly sequence exceeds a preset threshold, the long-term trend of historical measurement sequences is fused to determine the contribution of the external load, obtaining a pure load response. Based on the pure load response, targeted correction is performed on the load-affected portion of the current measurement wavelength signal, resulting in a corrected wavelength sequence. Finally, a time-series smoothing process is used to remove residual random fluctuations, determining the final stable wavelength output.
[0020] After the sensor is fully bonded and cured on-site, the system automatically activates the wavelength drift monitoring module. Using a high-precision fiber Bragg grating demodulator, it continuously acquires the Bragg center wavelength sequence for 300 seconds at a sampling rate of 100Hz, generating 30,000 data points λ1 to λ30,000. First, a low-pass Butterworth filter (cutoff frequency 0.8Hz, order 4) is used to smooth the original sequence to remove environmental noise interference. After filtering, the arithmetic mean of the sequence is calculated, and median filtering is used to remove outliers, resulting in a stable center wavelength λinstall of 1549.872 nm. Subsequently, the system extracts the reference wavelength λref under standard no-load conditions (temperature 18.5℃, relative humidity 50%) from the calibration archive corresponding to the sensor batch. The archive record shows λref as 1549.210 nm. The initial total wavelength offset Δλtotal is immediately calculated as Δinstall - λref = 1549.872 - 1549.210 = 0.662. nm; simultaneously, the current reading of the embedded temperature sensor at the installation location, Tinst = 21.2℃, is collected. The temperature deviation from the calibration reference, ΔT = 21.2 - 18.5 = 2.7℃, is calculated. Applying the calibrated temperature-wavelength coupling coefficient KT = 11.2 pm / ℃, the wavelength change caused by temperature, ΔλT = 11.2 × 2.7 = 30.24 pm, or 0.03024 nm, is obtained. Through compensation calculation, the net wavelength offset caused by residual installation stress, Δλres = Δλtotal - ΔλT = 0.662 - 0.030 = 0.632 nm, is separated. If this net offset exceeds the preset warning threshold of 0.450... If the wavelength reaches nm, the historical sequence fusion judgment logic is triggered. The system extracts the most recent 5 stable wavelength mean sequences of the same sensor under zero external load conditions within the previous 24 hours, forms a reference trend vector, and uses least squares linear regression to fit the residual change rate. It is determined that the external load contribution ratio in the current offset is 0.8%, thus correcting to obtain the pure residual stress wavelength component Δλpure=0.632×(1-0.008)=0.627 nm. Finally, combined with the strain-wavelength sensitivity coefficient Kε=1.20 pm / με and the reference wavelength λref, the residual installation strain εresidual=Δλpure / (Kε×λref)≈0.627 / (1.20×1549.210)≈337 με is calculated. This value is automatically recorded in the initial correction field of the core database. In subsequent real-time strain inversion processes, this benchmark residual strain component is dynamically subtracted to ensure that the monitoring data reflects the pure external load response.
[0021] S104. Obtain wavelength drift characteristics from the pure load response, use a filtering algorithm to remove noise interference, and obtain a clear strain tracking sequence.
[0022] Acquire the current measurement wavelength data. Process the current measurement wavelength data using a low-pass filtering algorithm to obtain a smoothed wavelength drift sequence. Calculate the local slope at each time point based on the smoothed wavelength drift sequence to obtain a local slope sequence. Calculate the difference between adjacent points in the local slope sequence to obtain a slope change sequence. If the number of points in the slope change sequence whose absolute value exceeds a preset threshold exceeds a certain limit, extract the long-term linear trend component from the historical wavelength drift sequence to obtain a trend baseline sequence. Perform a subtraction operation on the smoothed wavelength drift sequence based on the trend baseline sequence to obtain a trend-removed fluctuation sequence. For the trend-removed fluctuation sequence, compare it with a pre-stored standard fluctuation template through morphological similarity calculation to obtain a category determination result. Determine the amplitude adjustment coefficient based on the category determination result, and perform amplitude scaling processing on the load-affected part of the current measurement wavelength data to obtain an amplitude-corrected wavelength sequence. Smooth the amplitude-corrected wavelength sequence using a moving average method to obtain a stable wavelength output value.
[0023] After the sensor installation and positioning were completed and the environmental parameters stabilized, the system automatically initiated the wavelength drift feature extraction process. A high-precision fiber Bragg grating demodulator continuously acquired 600 seconds of Bragg center wavelength data at a sampling frequency of 200Hz, obtaining a total of 120,000 sampling points λ1 to λ120,000. First, a fourth-order Butterworth high-pass filter (cutoff frequency 0.05Hz) was applied to remove the slow drift trend. Then, wavelet transform (sym4 wavelet, decomposition level 5) was used to perform soft-threshold denoising on the detail coefficients (the threshold was adaptively determined using the Heursure rule), effectively eliminating high-frequency random noise interference. After multi-scale reconstruction, a relatively smooth wavelength sequence was obtained. Subsequently, the standard deviation σ of this sequence was calculated, and 3σ was set as the outlier criterion. After removing outliers outside the range, the arithmetic mean was taken, resulting in the current feature center wavelength λfeature being 1550.341. nm; The system synchronously reads the installation point temperature as 23.8℃ and humidity as 54%, retrieves the characteristic reference wavelength λbase of the corresponding batch under the baseline conditions (temperature 20.0℃, humidity 50%) from the historical calibration database, which is 1549.685 nm. The total drift Δλchar = λfeature - λbase = 1550.341 - 1549.685 = 0.656 nm; Simultaneously, using the temperature compensation model (quadratic polynomial fitting, coefficients already calibrated as a = 0.0087 pm / ℃², b = 10.95 pm / ℃), substituting the temperature deviation ΔT = 23.8 - 20.0 = 3.8℃, the temperature-coupled drift component Δλtemp = 0.0087 × 3.8² + 10.95 × 3.8 = 0.126 + 41.61 = 41.736 pm, or 0.04174 nm. nm; the mechanically related drift Δλmech = Δλchar - Δλtemp = 0.656 - 0.042 = 0.614 nm is separated by vector subtraction; if this value exceeds the system-set characteristic drift threshold of 0.380 nm, the historical sequence association analysis module is activated, extracting the seven most recent denoised characteristic wavelength sequences under no external influence within the previous 48 hours, constructing a time-drift matrix, and then using principal component analysis (PCA) to extract the first principal component as the trend basis vector. Support vector regression (SVR, kernel function rbf, C=15, ε=0.012) is then used to fit the trend residuals, estimating that the proportion of non-load environmental factors in the current drift is approximately 1.2%, thus correcting to obtain the pure load-induced wavelength characteristic component Δλclean = 0.614 × (1 - 0.012) = 0.607 nm; finally, combined with the calibrated strain-wavelength conversion coefficient Kε = 1.18 Using pm / με and the reference wavelength λbase, the strain response corresponding to the pure load is calculated as εload=Δλclean / (Kε×λbase)≈0.607 / (1.18×1549).The value ≈332 με (685) is automatically stored in the real-time feature database as the starting reference for a clear strain tracking sequence, and is used to achieve high-fidelity strain inversion in subsequent dynamic load monitoring.
[0024] S105. Based on the clear strain tracking sequence, determine the independent contribution of residual stress, update the reference zero point through an iterative optimization algorithm, and obtain the reference value of the state without residual stress.
[0025] Obtain the current strain measurement sequence. Process the current strain measurement sequence using Gaussian filtering to obtain a smoothed sequence. Calculate the local difference value at each time point based on the smoothed sequence to obtain a difference sequence. Calculate the cumulative sum of the absolute differences between consecutive points in the difference sequence to obtain the cumulative deviation. If the length of consecutive segments exceeding a preset limit in the cumulative deviation exceeds a certain length, extract the long-term baseline from the historical strain sequence to obtain the baseline reference. Perform a subtraction operation on the smoothed sequence based on the baseline reference to obtain the residual sequence. Compare the residual sequence with a pre-stored residual stress pattern using a template matching method to obtain the pattern matching result. Determine the residual stress compensation magnitude based on the pattern matching result, and perform compensation adjustment on the residual stress-related parts of the current strain measurement sequence to obtain the compensated strain sequence. Process the compensated strain sequence using an exponentially weighted average to obtain the final zero-residual reference sequence.
[0026] After acquiring a clear strain tracking sequence, the system automatically initiates the independent contribution analysis process for residual stress. First, it extracts the strain sequences ε1 to εn (n=86400, sampling interval 3 seconds) after temperature-humidity full compensation from the real-time feature database within the last 72 hours. The overall mean of the sequence, εmean=187 με, is calculated as a preliminary zero-point estimate. Then, a residual stress influence model is constructed, employing the Levenberg-Marquardt nonlinear least squares iterative optimization algorithm to characterize the residual stress release trend in an exponential decay form. The initial parameters are set as residual strain amplitude A0=45 με, decay time constant τ=360000 seconds, and offset B=0. Parameter optimization is performed by minimizing the objective function ∑(εt - [A0·exp(-t / τ)+B])². After approximately 28 iterations, the optimized parameters A=38.7 με, τ=412000 seconds, and B=-2.1 με converge. Finally, the optimized model is extrapolated to a stable value when time approaches infinity, i.e., the reference strain εref=B=-2.1 in the state without residual stress. με is used as the updated reference zero point. The system further calculates the difference between the current clear strain sequence and the new zero point, Δεadjusted=εload-εref=332-(-2.1)=334.1 με. At the same time, it extracts 7 sets of short-time strain fluctuation data under the no-load condition of the structure in the previous 24 hours, calculates the standard deviation of the residuals of each set and takes the median σresidual=4.8 με. The criterion is set that if the current Δεadjusted exceeds εref±5σresidual, the zero point secondary correction is triggered. On this basis, a Kalman filter (process noise covariance Q=0.25, measurement noise covariance R=9.0, initial state covariance P0=16) is applied to smooth and fuse the adjusted strain sequence to obtain the final pure strain reference value εpure=336.4 με after removing the independent contribution of residual stress. This value automatically covers and updates the zero stress state reference of the corresponding sensor in the updated historical reference database. It is used to eliminate the baseline drift caused by the accumulation of residual stress in subsequent long-term monitoring and achieve a more accurate load-strain correspondence.
[0027] S106. If the reference value matches the real-time wavelength signal, the multi-point sensor inputs are integrated through the data fusion process to determine the overall structural stress distribution and obtain accurate quantitative results.
[0028] Acquire the real-time wavelength signal sequence. Perform local linear fitting on the real-time wavelength signal sequence using a sliding window to obtain a fitted trend sequence. Calculate the absolute value of the residual within each window based on the fitted trend sequence to obtain a residual absolute value sequence. Statistically calculate the length of consecutive out-of-limit segments based on the residual absolute value sequence to obtain the out-of-limit segment length distribution. If the out-of-limit segment length distribution shows segments with lengths exceeding a preset value, extract a stable interval reference from the historical wavelength signal to obtain a stable reference sequence. Perform a difference operation between the stable reference sequence and the fitted trend sequence to obtain a trend deviation sequence. Apply wavelet transform decomposition to the trend deviation sequence to separate the low-frequency residual components, obtaining a low-frequency residual component sequence. Determine the compensation coefficient based on the amplitude range of the low-frequency residual component sequence, and perform amplitude adjustment on the corresponding positions of the real-time wavelength signal sequence to obtain an adjusted wavelength sequence. If the matching degree between the adjusted wavelength sequence and the stable reference sequence meets preset conditions, calculate the overall stress level using a weighted average of multi-sensor data to obtain the quantified value of structural stress.
[0029] After receiving the pre-filtered real-time wavelength signal sequence, the system automatically triggers the multi-sensor data fusion and overall stress distribution assessment process. First, it extracts the wavelength drift sequences λ1 to λm (m=57600, sampling interval 3 seconds) that have undergone temperature cross-sensitivity coefficient correction within the last 48 hours from the distributed fiber optic sensor network. The median relative drift of each measurement point's sequence is calculated as the initial matching benchmark. Then, the latest acquired wavelength value λcurrent is matched with the historical measurement point's λmedian using Pearson correlation coefficients. When the correlation coefficient ρ > 0.92 and |λcurrent - λmedian| < 18 pm, the reference value is considered successfully matched. After matching confirmation, the system immediately enters the multi-point fusion stage. The system calls a weighted least squares spatial interpolation algorithm, using the sensor position coordinates (x, y, z) as independent variables and the corresponding wavelength drift Δλ as the dependent variable to construct a three-dimensional strain field model. The weights are dynamically determined by the signal-to-noise ratio SNRi of each sensor (weight wi = SNRi² / ∑SNRj²). Next, Gaussian process regression is used (with Matérn kernel function selected). A stress distribution prediction was performed on the entire structural domain using a 5 / 2 core, length scale parameter l=2.5 m, and signal variance σf²=145 pm². The maximum principal stress σmax=162.7 MPa and shear stress τxy=41.3 MPa at the critical critical section were obtained. At the same time, the strain rate sequence of the low-speed steady segment in the previous 12 hours was extracted, and its upper limit of the 95% confidence interval v95=0.008 με / s was calculated. If the current fused stress increment rate exceeds 1.8 times v95, an abnormal distribution marker is activated. Finally, the system integrates the prediction results of all measurement points, extracts the dominant stress mode through principal component analysis to reduce dimensionality, and outputs the overall quantitative stress distribution vector map of the structure and the maximum equivalent stress value σvonMises=198.4 MPa. This result is written into the structural health status vector database in real time, providing a high-precision input basis for subsequent fatigue life prediction and adaptive adjustment of early warning threshold.
[0030] S107. Extract the uneven distribution features from the precise quantification results, use the compensation algorithm to calibrate the complex part of the direction, and determine the final residual stress stripping output.
[0031] Obtain the residual stress stripping output. Extract the unevenly distributed regions from the residual stress stripping output. Divide the unevenly distributed regions into multiple local sub-intervals based on their locations. Acquire multi-channel stress sequences for each local sub-interval. Calculate the interval average stress sequence for each local sub-interval through weighted superposition. Process the interval average stress sequence using principal component analysis to extract the dominant variation components, obtaining the dominant component sequence. Divide the intensity levels based on the amplitude fluctuation range of the dominant component sequence, obtaining the intensity level sequence. If the intensity level sequence shows the existence of continuous high-level intervals, obtain the baseline value sequence for the corresponding position from the historical stable stress sequence. Perform point-by-point subtraction between the baseline value sequence and the interval average stress sequence to obtain the local deviation sequence. Apply low-pass filtering to the local deviation sequence to separate the slowly drifting components, obtaining the drift component sequence. Perform subtraction correction on the corresponding positions of the interval average stress sequence based on the drift component sequence to obtain the corrected average stress sequence. Concatenate the corrected average stress sequences of all local sub-intervals to obtain the global corrected stress sequence.
[0032] After confirming successful matching between the reference value and the real-time wavelength signal, the system automatically initiates the non-uniformity feature extraction and residual stress compensation process. First, it calculates the spatial gradient of the fused three-dimensional strain field, using the Sobel three-dimensional operator to extract strain gradient components in each direction. Regions with gradient moduli exceeding 3.2 times the average gradient are identified as the core areas of non-uniform distribution. For example, a local concentration with a gradient moduli of 0.045 με / mm is detected at the structural flange connection. Then, it extracts the local wavelength drift variance sequence for these non-uniform regions, calculating the variance σ²local of each high-gradient measurement point over the past 24 hours. When σ²local > 32 pm², it is marked as a region with complex orientation. Next, the system calls the anisotropic compensation algorithm, using the spatial coordinates of the measurement points as input, to construct a Kriging-based anisotropic variogram model (combination of exponential and spherical kernels, with anisotropy ratios set to 1.8:1:0.7, corresponding to the x, y, and z directions). The wavelength drift values in the region with complex orientation are spatially resampled and calibrated. After calibration, the residual drift deviation is controlled within ±9. Within pm; then, through the inversion stripping method, the calibrated total strain is decomposed into elastic strain and residual strain. Using Hooke's law combined with the known material isotropic assumption (Young's modulus E=68.9GPa, Poisson's ratio ν=0.33), the residual stress components σres_x, σres_y, and σres_z are calculated, where the residual principal stress at the key node σres_max=78.6 MPa; finally, the compensation results are integrated with the original predicted stress field to generate the net loaded stress distribution after residual stress stripping, and the corrected maximum von Mises equivalent stress σvonMises_net=184.2 MPa is output. The residual stress vector field is written into a dedicated residual stress history database for subsequent crack initiation risk assessment and dynamic updating of the structural residual bearing capacity.
[0033] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.
Claims
1. A residual stress stripping method based on sensor wavelength signal processing, characterized in that, The method includes: By collecting Bragg wavelength data at the moment of sensor installation, the initial wavelength offset value is extracted from the data to obtain the reference offset caused by residual stress. Based on the reference offset, a compensation algorithm is used to adjust the subsequent measurement wavelength signal to determine the correction wavelength for residual peeling stress. If the correction wavelength exceeds the preset threshold, the historical measurement sequence is fused through the signal processing stage to determine the contribution of the external load and obtain the pure load response. Wavelength drift characteristics are obtained from the pure load response, and noise interference is removed by a filtering algorithm to obtain a clear strain tracking sequence; Based on the clear strain tracking sequence, the independent contribution of residual stress is determined, and the reference zero point is updated through an iterative optimization algorithm to obtain the reference value of the state without residual stress. If the reference value matches the real-time wavelength signal, the multi-point sensor inputs are integrated through the data fusion process to determine the overall structural stress distribution and obtain accurate quantification results. The uneven distribution features are extracted from the precise quantification results, and the complex part of the direction is calibrated by the compensation algorithm to determine the final residual stress stripping output.
2. The residual stress stripping method based on sensor wavelength signal processing according to claim 1, characterized in that, The process of acquiring Bragg wavelength data at the moment of sensor installation, extracting the initial wavelength offset value from the data, and obtaining the reference offset caused by residual stress includes: By acquiring Bragg wavelength-related data through sensors at the moment of installation, the raw data is collected, and a preliminary wavelength record is obtained. Based on the initial wavelength records, data processing methods were used to separate wavelength data related to the installation time and determine wavelength reference values independent of the environment. For the separated wavelength reference values, the initial offset information is extracted, and it is determined whether the initial offset exceeds the preset threshold range. If it does, it is marked as abnormal wavelength data. By further processing the abnormal wavelength data, the specific impact of residual stress on wavelength changes is analyzed, and the offset caused by residual stress is obtained. Based on the offset caused by residual stress and the pattern of wavelength change, the specific value of the reference offset is calculated, and the final reference offset result is determined. After obtaining the final reference offset result, a pre-established stress influence model is used to analyze the correspondence between stress influence and reference offset, and to obtain the quantitative value of residual stress. By storing the quantified values of residual stress, a structured stress impact record is generated, completing the entire data processing workflow.
3. The residual stress stripping method based on sensor wavelength signal processing according to claim 1, characterized in that, The step of adjusting the subsequent measurement wavelength signal based on the reference offset using a compensation algorithm to determine the correction wavelength for residual stress after peeling includes: By analyzing the reference offset, a pre-established compensation algorithm is used to perform preliminary correction on the wavelength signal obtained from subsequent measurements, and the adjusted signal data is obtained. Based on the adjusted signal data, a data calibration method is applied to separate the wavelength deviation corresponding to the stress interference, and the intermediate result after correction is determined, targeting the interference factors of residual stress. After obtaining the intermediate results, signal adjustment techniques are used to further optimize the accuracy of the wavelength signal based on the specific distribution of the wavelength deviation, resulting in an optimized wavelength dataset. From the optimized wavelength dataset, residual components related to stress interference are extracted. Combined with the stripping effect processing logic, it is determined whether there are abnormal deviations. If there are abnormal deviations, the abnormal parts are calibrated a second time to obtain the corrected wavelength values. For the corrected wavelength value, a data calibration verification mechanism is applied to analyze its matching degree with the measurement data and determine the final corrected wavelength result. By processing the final corrected wavelength result and combining it with real-time data from subsequent measurements, the signal adjustment parameters are continuously updated to obtain a dynamically optimized wavelength signal output. After obtaining the dynamically optimized wavelength signal output, a historical record of data calibration is constructed based on the accumulation of long-term measurement data to determine a stable correction scheme for the wavelength signal.
4. The residual stress stripping method based on sensor wavelength signal processing according to claim 1, characterized in that, If the correction wavelength exceeds a preset threshold, the historical measurement sequence is fused through signal processing to determine the contribution of the external load and obtain the pure load response, including: The current measurement wavelength signal is acquired, and a preliminary adjustment is made to the current measurement wavelength signal through a pre-established compensation model to obtain the preliminarily adjusted wavelength sequence. Based on the preliminarily adjusted wavelength sequence, the interference component introduced by the external load is separated using the sequence decomposition method, and the load response components are determined. Based on the load response components, the influence of the external load is removed from the initially adjusted wavelength sequence to obtain the residual wavelength sequence after load removal. For the residual wavelength sequence after the load is removed, the morphological matching technique is used to compare it with the preset normal morphology to determine the location of abnormal points in the residual wavelength sequence and obtain the wavelength sequence marked with abnormal points. If the number of outliers in the wavelength sequence marked with outliers exceeds a preset threshold, the long-term trend of historical measurement sequences is fused to determine the contribution of external load and obtain a pure load response. Based on the pure load response, targeted corrections are applied to the load-affected portion of the current measured wavelength signal to obtain the corrected wavelength sequence; Based on the corrected wavelength sequence, time-series smoothing is used to remove residual random fluctuations, and the final stable wavelength output is determined.
5. The residual stress stripping method based on sensor wavelength signal processing according to claim 1, characterized in that, The process of obtaining wavelength drift characteristics from the pure load response, removing noise interference using a filtering algorithm, and obtaining a clear strain tracking sequence includes: Obtain the current measurement wavelength data; The current measurement wavelength data is processed by a low-pass filtering algorithm to obtain a smooth wavelength drift sequence; The local slope sequence is obtained by calculating the local slope at each time point based on the smoothed wavelength drift sequence. Calculate the difference between adjacent points for the local slope sequence to obtain the slope change sequence; If the number of points in the slope change sequence whose absolute value exceeds the preset threshold exceeds the limit, then the long-term linear trend component is extracted from the historical wavelength drift sequence to obtain the trend benchmark sequence. The smoothed wavelength drift sequence is subtracted from the trend baseline sequence to obtain the trend-removed fluctuation sequence. For the trend-removed fluctuation sequence, the category determination result is obtained by comparing it with the pre-stored standard fluctuation template through morphological similarity calculation; Based on the category determination result, the amplitude adjustment coefficient is determined, and the amplitude scaling process is performed on the load-affected part of the current measurement wavelength data to obtain the amplitude-corrected wavelength sequence. The wavelength sequence after amplitude correction is smoothed by using a moving average method to obtain a stable wavelength output value.
6. The residual stress stripping method based on sensor wavelength signal processing according to claim 1, characterized in that, The process of determining the independent contribution of residual stress based on a clear strain tracking sequence, updating the baseline zero point through an iterative optimization algorithm, and obtaining a reference value for a state without residual stress includes: Obtain the current strain measurement sequence; A smooth sequence is obtained by processing the current strain measurement sequence using Gaussian filtering; The local difference value at each time point is calculated based on the smoothed sequence to obtain the difference sequence; The cumulative bias is obtained by calculating the cumulative sum of the absolute differences between consecutive points in a difference sequence. If the length of a segment in the cumulative deviation that continuously exceeds the preset limit exceeds the specified length, then a long-term baseline is extracted from the historical strain sequence to obtain a baseline reference. The smoothed sequence is subtracted based on the baseline reference to obtain the residual sequence; For the residual sequence, the template matching method is used to compare it with the pre-stored residual stress mode to obtain the mode matching result; The residual stress compensation range is determined based on the pattern matching results. Compensation adjustment is performed on the residual stress-related parts in the current strain measurement sequence to obtain the compensated strain sequence. The strain sequence after compensation is processed by exponential weighted averaging to obtain the final zero-residual reference sequence.
7. The residual stress stripping method based on sensor wavelength signal processing according to claim 1, characterized in that, If the reference value matches the real-time wavelength signal, the multi-point sensor inputs are integrated through a data fusion process to determine the overall structural stress distribution and obtain accurate quantitative results, including: Acquire real-time wavelength signal sequences; By performing local linear fitting on the real-time wavelength signal sequence through a sliding window, a fitted trend sequence is obtained. Calculate the absolute value of the residuals within each window based on the fitted trend sequence to obtain the absolute value of the residuals sequence; The lengths of consecutive out-of-limit segments are statistically analyzed for the absolute value sequence of residuals to obtain the distribution of out-of-limit segment lengths; If the length distribution of the over-limit segment shows that there are segments with lengths exceeding the preset value, then a stable interval reference is extracted from the historical wavelength signal to obtain a stable reference sequence; The trend deviation sequence is obtained by performing a difference operation between the stable reference sequence and the fitted trend sequence; Wavelet transform decomposition is applied to the trend deviation sequence to separate the low-frequency residual component, resulting in the low-frequency residual component sequence. The compensation coefficient is determined by the amplitude range of the low-frequency residual component sequence, and the amplitude is adjusted at the corresponding position of the real-time wavelength signal sequence to obtain the adjusted wavelength sequence. If the matching degree between the adjusted wavelength sequence and the stable reference sequence meets the preset conditions, the overall stress level is calculated by weighted averaging of multi-sensor data to obtain the quantified value of structural stress.
8. The residual stress stripping method based on sensor wavelength signal processing according to claim 1, characterized in that, The process of extracting uneven distribution features from precise quantification results, using a compensation algorithm to calibrate complex directional parts, and determining the final residual stress stripping output includes: Obtain residual stress stripping output; Extracting unevenly distributed regions from the residual stress stripping output; Divide the unevenly distributed regions into multiple local sub-intervals; Multi-channel stress sequences were acquired for each local sub-interval. The interval average stress sequence of each local sub-interval is calculated by weighted superposition; Principal component analysis was used to process the interval average stress sequence and extract the dominant variation components to obtain the dominant component sequence. Intensity levels are determined by classifying the amplitude fluctuation range of the dominant component sequence, thus obtaining an intensity level sequence. If the strength grade sequence shows a continuous high grade interval, then obtain the benchmark value sequence for the corresponding position from the historical stable stress sequence; The local deviation sequence is obtained by performing point-by-point subtraction between the benchmark value sequence and the interval average stress sequence; Low-pass filtering is applied to the local deviation sequence to separate the slowly drifting component, thus obtaining the drift component sequence; The interval average stress sequence is corrected by subtraction based on the drift component sequence at the corresponding position to obtain the corrected average stress sequence. The corrected average stress sequences of all local sub-intervals are spliced together to obtain the global corrected stress sequence.