CFRP wing skin sensor signal optimization method based on double-index screening
Patent Information
- Application Number
- CN202511777740.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-28
- Publication Date
- 2026-03-06
Smart Images
Figure CN121612985A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of CFRP wing skin structure health monitoring technology, and relates to a method for optimizing CFRP wing skin sensor signals based on dual-index screening. Background Technology
[0002] Carbon fiber reinforced polymer (CFRP) composites, with their lightweight, high strength, excellent fatigue resistance, and designable anisotropy, have become the mainstream material for UAV wing skins. Globally, CFRP accounts for 60% of UAV structures. 80%. As a critical outer structural component of the UAV wing, the wing skin not only needs to maintain its aerodynamic shape and transmit aerodynamic loads, but also bears the important responsibility of protecting internal equipment. Its structural integrity directly determines the UAV's flight performance and safety. In practical testing scenarios, Lamb wave monitoring technology, with its millisecond-level response speed, long-distance propagation capability, and high sensitivity to internal latent damage, has become an ideal technical solution for detecting internal crack damage in CFRP wing skin. Its A0 / S0 mode conversion characteristics can also accurately characterize the damage depth, providing a physical basis for damage localization.
[0003] However, during long-term service, CFRP wing skin is highly susceptible to developing internal microcracks due to cyclic loading and impacts, which gradually accumulate and propagate. If these cracks are not detected promptly and accurately, they can lead to macroscopic damage such as delamination and fiber breakage, ultimately causing structural failure and seriously threatening flight safety. Simultaneously, when Lamb waves propagate in curved CFRP wing skin, modal coupling and dispersion effects are caused by the structural curvature, resulting in severe modal aliasing noise interference in signals collected by multiple sensors. Existing sensor signal screening methods often rely on single indicators, such as correlation coefficient methods and energy threshold methods, which struggle to simultaneously capture the random abrupt changes and self-similar destructive characteristics of signals. They cannot effectively distinguish between damaged signals and noise, leading to high false positive and false negative rates during screening, which in turn affects the accuracy of subsequent damage localization. This has become a key bottleneck restricting the implementation of CFRP wing skin structural health monitoring technology.
[0004] To address the aforementioned issues, existing technologies have proposed various sensor signal optimization or screening schemes. For example, patent CN115310208A discloses a sensor optimization arrangement method for wing aerodynamic load reconstruction. This method acquires holographic aerodynamic data of the wing through CFD calculations, establishes a mapping relationship between coordinate parameters and pressure coefficients using a Gaussian process regression model, and selects the optimal sensor placement method based on the reconstruction error. Its core advantage lies in improving the aerodynamic load reconstruction accuracy with a limited number of sensors and reducing experimental costs. However, this method focuses on optimizing the sensor placement position and does not involve noise screening and quality improvement of the signal itself, failing to solve the signal distortion problem caused by mode aliasing. Other studies have used time-frequency analysis methods such as Short-Time Fourier Transform (STFT) and Empirical Mode Decomposition (EMD) to process sensor signals. However, STFT is difficult to capture non-stationary state transitions caused by microcracks due to its fixed time window, and EMD is prone to generating spurious components in curved structures, neither of which can effectively extract damage features. Furthermore, traditional dimensionality reduction methods such as Principal Component Analysis (PCA) have poor adaptability to nonlinear time-series signals, easily leading to the loss of damage features. These methods either focus solely on sensor placement optimization or suffer from inaccurate feature extraction and weak anti-interference capabilities, making it difficult to meet the practical requirements for efficient screening of sensor signals in CFRP wing skin. Therefore, there is an urgent need for a sensor signal optimization method that can simultaneously achieve noise suppression, accurate feature capture, and high stability to address the shortcomings of existing technologies and provide high-quality data support for subsequent damage localization. Summary of the Invention
[0005] This invention provides a method for optimizing CFRP wing skin sensor signals based on dual-index screening, which solves the technical problems of difficult-to-distinguish modal aliasing noise, high false positive rate and missed selection rate caused by relying on a single index in the existing CFRP wing skin sensor signal screening.
[0006] To solve the above problems, the technical solution adopted by the invention is as follows: The method for optimizing CFRP wing skin sensor signals based on dual-index screening includes the following steps: S01 Construct a multi-sensor network for CFRP wing skin, the sensor network including transmitting sensors and receiving sensors; collect Lamb wave signals of the CFRP wing skin in a damage-free state under the sensor network to form a damage dataset; collect Lamb wave signals of at least 45 different internal crack damage locations of the CFRP wing skin under the sensor network to form a damage dataset. S02 sets two calculation parameters: embedding dimension and maximum number of segments; based on the lossless dataset, the baseline value of permutation entropy PE is calculated. Baseline value of Higuchi fractal dimension HFD Simultaneously calculate the standard deviation of all PE values in the damage-free dataset. Standard deviation of all HFD values in the damage-free dataset Combined with the preset statistical significance threshold, the PE screening threshold and HFD screening threshold are determined respectively; the maximum number of segments refers to the maximum number of segments when dividing the Lamb wave time domain signal into sub-segments, and the value range is 5~20. S03 calculates the PE value and HFD value of each of the 26 received sensor signals corresponding to each damage location in the damage dataset; a dynamic threshold strategy is used to filter sensor signals: if the PE value of a certain sensor signal meets the following conditions... Or the HFD value of the sensor signal satisfies If the sensor signal is not found, then the sensor signal is retained; where, , All are statistical significance coefficients, ranging from 1 to 3. ; S04 Statistical damage dataset: Sensors retained in all damage location cases; Sensors retained in at least 35 damage location cases are selected to form a common sensitive sensor set; Lamb wave signal corresponding to this common sensitive sensor set is output.
[0007] The principle and advantages of this scheme are as follows: A multi-sensor network consisting of one transmitting sensor and 26 receiving sensors was constructed to collect Lamb wave signals from both undamaged states and at least 45 different internal crack damage locations, establishing undamaged and damaged datasets to provide a complete data foundation for subsequent analysis. Secondly, key parameters such as embedding dimension (3-7) and maximum number of segments (5-20) were set. Based on the undamaged dataset, baseline values and standard deviations of permutation entropy (PE) and Higuchi fractal dimension (HFD) were calculated. Combined with statistical significance coefficients α and β (1-3, with α=β), a dynamic screening threshold was determined. PE quantifies the randomness of signal variation, while HFD captures the self-similarity disruption of the signal. These two indicators characterize the impact of damage on the signal from different dimensions. Subsequently, for the sensor signal at each damage location, its PE and HFD values were calculated. If either indicator significantly deviates from the baseline, the signal was retained, achieving preliminary screening of valid signals. Finally, by statistically analyzing the sensors retained in all damage cases, sensors that showed sensitivity at at least 35 damage locations were selected to form a common sensitive sensor set, ensuring the stability and reliability of the output signal and providing high-quality data input for subsequent damage localization.
[0008] Compared with existing technologies, this scheme utilizes the synergistic effect of PE and HFD dual indicators to simultaneously capture random abrupt changes and self-similarity disruptions in signals, effectively distinguishing modal mixing noise from real damage signals and significantly reducing false positive and false negative rates. For example, the traditional correlation coefficient method can only reflect the linear correlation of signals and is difficult to cope with the complex signal interference in curved CFRP structures, while the complementarity of the two indicators in this scheme can improve the screening accuracy by more than 30%. The dynamic threshold strategy design realizes an adaptive response to signal changes. Compared with the fixed threshold method, it can better adapt to signal fluctuations under different damage locations and different propagation paths, avoiding the loss of effective signals or noise residue caused by threshold rigidity. A common sensitive sensor set is constructed through large-sample validation of at least 35 damage cases, ensuring the stability and generalization ability of the screening results and overcoming the problem of sensor sensitivity changing with damage location in existing technologies. When the optimized signal output by this scheme is used for subsequent damage localization, the localization error can be reduced to about 2mm, which is far superior to the localization accuracy of traditional screening methods. This solution requires no complex hardware upgrades; signal quality can be improved simply through algorithm optimization. It balances practicality and cost-effectiveness, providing more reliable technical support for online health monitoring of UAV CFRP wing skin.
[0009] Furthermore, the specific process for calculating the permutation entropy PE in S02 is as follows: Set the embedding dimension as The value ranges from 3 to 7, and the delay time is 1; the Lamb wave time-domain signal in the damage-free dataset is denoted as... ,in The value can be 1 or 2. , The number of sampling points for the Lamb wave time-domain signal is determined by the sampling frequency of 100kHz and the analysis step duration to ensure complete acquisition of the Lamb wave propagation waveform; according to In the form of Divided into A length of The subsequence, where (i.e., delay time) The value can be 1 or 2. ; For each subsequence Sort the elements within the subsequence in ascending order of their numerical values to obtain the permutation pattern of the subsequence. , The expression is ,in Representing a subsequence The Middle The index of the small element in the atomic sequence The value can be 1 or 2. ; Count the occurrences of all subsequence permutations to obtain the total number of different permutations. ( express (factorial), and calculate the probability of occurrence of each permutation pattern. , The value can be 1 or 2. , The calculation method is to divide the number of occurrences of the permutation pattern by the total number of all subsequences. ; According to the formula Calculate the permutation entropy; the value of PE is maximized when the probability of all permutation patterns occurring is equal. When the Lamb wave time-domain signal is completely regular, the value of PE is the minimum value of 0.
[0010] Furthermore, the specific process for calculating the Higuchi fractal dimension HFD in S02 is as follows: Set the maximum number of segments to The value ranges from 5 to 20; the time-domain signal of the Lamb wave in the non-destructive dataset is denoted as... ( , The number of sampling points; the number of segments. ( The value can be 1 or 2. That is, from 1 to the maximum number of segments. Take values in sequence, Divided into Section, No. The expression for the segment signal is: ,in The value can be 1 or 2. , Indicates to Take the integer part; According to the formula ; Calculate the curve length of each signal segment. ;in, This indicates the integer operation. This indicates that the number of segments is k At that time, the first i The total number of sampling points contained in a segment signal. The sampling interval for the Lamb wave signal is determined by the sampling frequency of 100kHz. ); by x-axis Using the y-axis, for all The data points are linearly fitted; the slope of the fitted line is the Higuchi fractal dimension HFD; the value of HFD ranges from 1 to 2.
[0011] Furthermore, the embedding dimension in S02 has a value range of 37, and its specific value is determined based on the number of sampling points N of the Lamb wave signal: when When the embedding dimension is 3~5, when When the embedding dimension is 5 to 7, the total number of subsequences is... .
[0012] Furthermore, when acquiring the Lamb wave signal in S01, the sampling frequency is set to 100kHz; the propagation speed of the Lamb wave in the CFRP wing skin curved panel is calculated as follows: First, follow the formula Calculate the low-frequency velocity of the A0 mode of the Lamb wave in a CFRP flat plate. ;in, The tensile modulus of CFRP material along the fiber direction. The density of CFRP material Poisson's ratio of CFRP material along the 1-2 direction. Poisson's ratio of CFRP material along the 2-1 direction; Then follow the formula Calculate the propagation speed of the Lamb wave in the CFRP curved panel. ;in, This is the curvature correction factor, determined based on the actual curvature of the CFRP wing skin: when the radius of curvature is the first specification, When the radius of curvature is the second specification, When the radius of curvature is the third specification, When the radius of curvature is the fourth specification, ; Measuring the straight-line distance from the transmitting sensor to the farthest receiving sensor in a sensor network According to the formula Determine the analysis step duration of the Lamb wave signal This ensures that the acquired Lamb wave signal completely includes the waveform that has propagated to the farthest receiving sensor.
[0013] Furthermore, the dynamic threshold strategy in S03 also includes a signal validity verification step: calculating the signal-to-noise ratio of the initially retained sensor signals. , The calculation formula is: in, The effective frequency band of the sensor signal is 80kHz~120kHz, corresponding to the power of the main frequency band of the Lamb wave signal. The power in the noise frequency band of the sensor signal <50kHz or >200kHz; if the calculated If so, then the sensor signal will be retained; if If so, then the sensor signal will be discarded.
[0014] Furthermore, step S04 also includes a step for quantifying the sensitivity of a common sensitive sensor: according to the formula... Calculate the overall sensitivity coefficient of each common sensing sensor. ;in, This represents the number of damage cases that are retained for this sensor in the damage dataset. , For the sensor in the first PE values in individual damage cases For the sensor in the first HFD values in individual damage cases, The baseline PE value for the damage-free dataset. The baseline values of HFD for the damage-free dataset. The standard deviation of the PE values in the damage-free dataset. The standard deviation of the HFD values in the damage-free dataset; according to Sort all commonly sensitive sensors by value from largest to smallest, and select the one with priority. The top 5 sensors are used to locate damage to the CFRP wing skin.
[0015] Furthermore, in step S03, when calculating the Higuchi fractal dimension, a weighted linear fitting method is used to optimize the calculation accuracy. The specific process is as follows: For each segment number ( , (Maximum number of segments), according to the formula Calculate the weight corresponding to the number of segments. ;in, The length of the curve under this number of segments. The smaller the value, the smoother the signal and the higher the data reliability. The larger; According to the formula Calculate the slope of the linear fit ;in, Number of segments The natural logarithm, Curve length The natural logarithm; The slope This is the optimized Higuchi fractal dimension, whose calculation error is reduced by 10% compared to traditional equal-weight linear fitting. 15%.
[0016] Furthermore, the model parameters of the CFRP wing skin in S01 are as follows: Dimensions: Length, width, and thickness are designed according to the actual application dimensions of CFRP wing skin; the radius of curvature includes four different specifications, namely the first specification. radius of curvature Second specification radius of curvature Third specification radius of curvature Fourth specification radius of curvature This corresponds to the root, middle, wingtip transition area, and wingtip area of the CFRP wing skin. Ply structure: The ply angle sequence is as follows It consists of 6 layers; each layer has a uniform thickness, with a single layer thickness of 0.125mm and a total thickness of 0.75mm. Material parameters: density is Tensile modulus , , Poisson's ratio , , shear modulus , , .
[0017] Furthermore, it also includes a quality assessment step for the optimized signal: According to the formula Calculate the damage feature identification degree of the optimized signal ;in, The number of sensors in the common sensitive sensor set ; For the first The average signal of each sensitive sensor under damaged conditions. For the first The standard deviation of the signal from a sensitive sensor under damaged conditions; For the first The average value of the signal from each sensitive sensor in a non-damaged state. For the first The standard deviation of the signal from a sensitive sensor in a non-damaged state; If the calculation yields If the optimized signal is deemed to be of acceptable quality and can be used for subsequent damage localization; if Then the embedding dimension in S02 is readjusted. Adjust within a range of 37, maximum number of segments Adjust within the range of 520, or the statistical significance coefficient in S03. , Adjust within the range of 13, and repeat step 24 until a qualified optimized signal is obtained. Attached Figure Description
[0018] Figure 1 This is a flowchart of the present invention; Figure 2 This is a sensor network diagram; Figure 3 Waveforms of Lamb signals received at different center frequencies for CFRP wing skin. Detailed Implementation
[0019] Example 1 As attached Figure 1 As shown, the CFRP wing skin sensor signal optimization method based on dual-index screening includes the following steps: S01 Construct a multi-sensor network for CFRP wing skin, which includes one transmitting sensor and 26 receiving sensors; collect Lamb wave signals of the CFRP wing skin in a damage-free state under this sensor network to form a damage-free dataset; collect Lamb wave signals of at least 45 different internal crack damage locations of the CFRP wing skin under this sensor network to form a damage dataset. S02 sets two calculation parameters: embedding dimension and maximum number of segments; based on the lossless dataset, the baseline value of permutation entropy PE is calculated. Baseline value of Higuchi fractal dimension HFD Simultaneously calculate the standard deviation of all PE values in the damage-free dataset. Standard deviation of all HFD values in the damage-free dataset Combined with the preset statistical significance threshold, the PE screening threshold and HFD screening threshold are determined respectively; the maximum number of segments refers to the maximum number of segments when dividing the Lamb wave time domain signal into sub-segments, and the value range is 5~20. S03 calculates the PE value and HFD value of each of the 26 received sensor signals corresponding to each damage location in the damage dataset; a dynamic threshold strategy is used to filter sensor signals: if the PE value of a certain sensor signal meets the following conditions... Or the HFD value of the sensor signal satisfies If the sensor signal is not found, then the sensor signal is retained; where, , All are statistical significance coefficients, ranging from 1 to 3. ; S04 Statistical damage dataset: Sensors retained in all damage location cases; Sensors retained in at least 35 damage location cases are selected to form a common sensitive sensor set; Lamb wave signal corresponding to this common sensitive sensor set is output.
[0020] A multi-sensor network consisting of one transmitting sensor and 26 receiving sensors was constructed to collect Lamb wave signals from both undamaged states and at least 45 different internal crack damage locations, establishing undamaged and damaged datasets to provide a complete data foundation for subsequent analysis. Secondly, key parameters such as embedding dimension (3-7) and maximum number of segments (5-20) were set. Based on the undamaged dataset, baseline values and standard deviations of permutation entropy (PE) and Higuchi fractal dimension (HFD) were calculated. Combined with statistical significance coefficients α and β (1-3, with α=β), a dynamic screening threshold was determined. PE quantifies the randomness of signal variation, while HFD captures the self-similarity disruption of the signal. These two indicators characterize the impact of damage on the signal from different dimensions. Subsequently, for the sensor signal at each damage location, its PE and HFD values were calculated. If either indicator significantly deviates from the baseline, the signal was retained, achieving preliminary screening of valid signals. Finally, by statistically analyzing the sensors retained in all damage cases, sensors that showed sensitivity at at least 35 damage locations were selected to form a common sensitive sensor set, ensuring the stability and reliability of the output signal and providing high-quality data input for subsequent damage localization.
[0021] Compared with existing technologies, this scheme utilizes the synergistic effect of PE and HFD dual indicators to simultaneously capture random abrupt changes and self-similarity disruptions in signals, effectively distinguishing modal mixing noise from real damage signals and significantly reducing false positive and false negative rates. For example, the traditional correlation coefficient method can only reflect the linear correlation of signals and is difficult to cope with the complex signal interference in curved CFRP structures, while the complementarity of the two indicators in this scheme can improve the screening accuracy by more than 30%. The dynamic threshold strategy design realizes an adaptive response to signal changes. Compared with the fixed threshold method, it can better adapt to signal fluctuations under different damage locations and different propagation paths, avoiding the loss of effective signals or noise residue caused by threshold rigidity. A common sensitive sensor set is constructed through large-sample validation of at least 35 damage cases, ensuring the stability and generalization ability of the screening results and overcoming the problem of sensor sensitivity changing with damage location in existing technologies. When the optimized signal output by this scheme is used for subsequent damage localization, the localization error can be reduced to about 2mm, which is far superior to the localization accuracy of traditional screening methods. This solution requires no complex hardware upgrades; signal quality can be improved simply through algorithm optimization. It balances practicality and cost-effectiveness, providing more reliable technical support for online health monitoring of UAV CFRP wing skin.
[0022] The specific process for calculating the permutation entropy PE in S02 is as follows: Set the embedding dimension as The value ranges from 3 to 7, and the delay time is 1; the Lamb wave time-domain signal in the damage-free dataset is denoted as... ,in The value can be 1 or 2. , The number of sampling points for the Lamb wave time-domain signal is determined by the sampling frequency of 100kHz and the analysis step duration to ensure complete acquisition of the Lamb wave propagation waveform; according to In the form of Divided into A length of The subsequence, where (i.e., delay time) The value can be 1 or 2. ; For each subsequence Sort the elements within the subsequence in ascending order of their numerical values to obtain the permutation pattern of the subsequence. , The expression is ,in Representing a subsequence The Middle The index of the small element in the atomic sequence The value can be 1 or 2. ; Count the occurrences of all subsequence permutations to obtain the total number of different permutations. ( express (factorial), and calculate the probability of occurrence of each permutation pattern. , The value can be 1 or 2. , The calculation method is to divide the number of occurrences of the permutation pattern by the total number of all subsequences. ; According to the formula Calculate the permutation entropy; the value of PE is maximized when the probability of all permutation patterns occurring is equal. When the Lamb wave time-domain signal is completely regular, the PE value is at its minimum of 0. By setting an embedding dimension of 3 to 7 to adapt to the signal characteristics of different sampling points and a delay time of 1, the original signal is divided into several subsequences. This preserves the temporal correlation of the signal while avoiding information redundancy through reasonable control of the subsequence length. Each subsequence is sorted in ascending order and the arrangement pattern is extracted. Essentially, this transforms the nonlinear time-domain signal into a statistically significant symbol sequence, thereby capturing the hidden changes in order within the signal. When there are internal cracks in the CFRP wing skin, the damage causes a change in the Lamb wave propagation path, making the signal arrangement pattern more dispersed and its probability of occurrence more uniform, resulting in a significant increase in the PE value. While mode mixing noise interferes with the signal, it is difficult to cause a global change in the arrangement pattern, and the PE value fluctuates less. The range of PE values can both anchor the extreme state of the signal through the maximum and minimum values and intuitively reflect the differences in signal complexity caused by damage through the magnitude of specific values, providing a quantifiable and easily comparable feature benchmark for subsequent dynamic threshold selection.
[0023] Furthermore, the specific process for calculating the Higuchi fractal dimension HFD in S02 is as follows: Set the maximum number of segments to The value ranges from 5 to 20; the time-domain signal of the Lamb wave in the non-destructive dataset is denoted as... ( , The number of sampling points; the number of segments. ( The value can be 1 or 2. That is, from 1 to the maximum number of segments. Take values in sequence, Divided into Section, No. The expression for the segment signal is: ,in The value can be 1 or 2. , Indicates to Take the integer part; According to the formula ; Calculate the curve length of each signal segment. ;in, This represents the integer operation, where k is the number of sampling points contained in the i-th sub-segment of the Lamb wave time-domain signal. The sampling interval for the Lamb wave signal is determined by the sampling frequency of 100kHz. ); by x-axis Using the y-axis, for all Linear fitting is performed on the data points; the slope of the fitted line is the Higuchi fractal dimension (HFD). The HFD value ranges from 1 to 2, and a maximum number of segments K is set from 5 to 20. By taking values from 1 to K sequentially, the signal is segmented into multiple scales, which can comprehensively cover the overall trend and local details of the signal, while avoiding feature loss due to overly coarse segmentation or noise interference introduced by overly fine segmentation. When calculating the length of each curve segment according to a specific formula, the sampling interval is taken into consideration, ensuring the accuracy and dimensional consistency of the length calculation, and effectively characterizing the joint variation characteristics of the signal in the time and amplitude dimensions. When there are internal cracks in CFRP, the damage will disrupt the continuity of Lamb wave propagation, leading to a decrease in signal self-similarity and a significant drop in the HFD value; while mode mixing noise usually manifests as local random fluctuations, which is difficult to change the overall self-similarity characteristics of the signal, and the HFD value is relatively stable. In addition, the HFD value is limited to the range of 1 to 2. The value can intuitively reflect the complexity of the signal structure. It complements the PE from the perspective of randomness and together they form a solid foundation for dual index screening. This provides important support for the subsequent accurate differentiation of damage signals from noise and improves the reliability of sensor signal screening. It effectively makes up for the shortcomings of traditional single feature indexes in comprehensively describing nonlinear signal changes.
[0024] Furthermore, the embedding dimension in S02 has a value range of 37, and its specific value is determined based on the number of sampling points N of the Lamb wave signal: when When the embedding dimension is 3~5, when When the embedding dimension is 5 to 7, the total number of subsequences is... When the number of sampling points is small When choosing a smaller embedding dimension, 3 to 5, select that. This avoids insufficient effective subsequences due to excessively long factor sequences, preventing the loss of critical timing information in the signal; when the number of sampling points is large... At this time, appropriately increase the embedding dimension by 5~7. This approach can more comprehensively depict the complex changing patterns of signals, avoiding the omission of subtle features caused by insufficient dimensionality. Simultaneously, limiting the total number of subsequences to no less than 100 ensures a sufficient sample size for subsequent permutation pattern statistics, reducing the interference of random factors on the permutation entropy (PE) calculation results and ensuring that the PE value stably and accurately reflects the random changes of the signal. This dynamic adaptation strategy effectively solves the problem that a fixed embedding dimension is difficult to adapt to signals of different scales, avoiding both feature underfitting due to excessively small dimensions and computational redundancy caused by excessively large dimensions.
[0025] Furthermore, when acquiring the Lamb wave signal in S01, the sampling frequency is set to 100kHz; the propagation speed of the Lamb wave in the CFRP wing skin curved panel is calculated as follows: First, follow the formula Calculate the low-frequency velocity of the A0 mode of the Lamb wave in a CFRP flat plate. ;in, The tensile modulus of CFRP material along the fiber direction. The density of CFRP material Poisson's ratio of CFRP material along the 1-2 direction. Poisson's ratio of CFRP material along the 2-1 direction; Then follow the formula Calculate the propagation speed of the Lamb wave in the CFRP curved panel. ;in, This is the curvature correction factor, determined based on the actual curvature of the CFRP wing skin: when the radius of curvature is the first specification, When the radius of curvature is the second specification, When the radius of curvature is the third specification, When the radius of curvature is the fourth specification, ; Measuring the straight-line distance from the transmitting sensor to the farthest receiving sensor in a sensor network According to the formula Determine the analysis step duration of the Lamb wave signal This ensures that the acquired Lamb wave signal completely contains the waveform that propagates to the farthest receiving sensor. A sampling frequency of 100kHz accurately captures subtle fluctuations in Lamb waves, avoiding signal distortion caused by insufficient sampling. Simultaneously, matching the center frequency of the excitation wave effectively preserves transient response characteristics induced by damage. By first calculating the low-frequency velocity of the A0 mode in the flat panel, and then combining this with correction coefficients corresponding to different radii of curvature, the actual propagation velocity in the curved panel is obtained. This fully considers the influence of the curved surface characteristics of the CFRP wing skin on wave propagation, solving the problem of mismatch between traditional flat panel velocity calculations and actual curved surface scenarios. Determining the analysis step duration based on the distance from the transmitting sensor to the farthest receiving sensor and the corrected propagation velocity ensures complete acquisition of the entire Lamb wave propagation waveform, avoiding effective signal truncation due to excessively short durations or redundant noise introduced by excessively long durations. This ensures data validity while reducing the computational burden of subsequent signal processing.
[0026] Furthermore, the dynamic threshold strategy in S03 also includes a signal validity verification step: calculating the signal-to-noise ratio of the initially retained sensor signals. , The calculation formula is: in, The effective frequency band of the sensor signal is 80kHz~120kHz, corresponding to the power of the main frequency band of the Lamb wave signal. The power in the noise frequency band of the sensor signal <50kHz or >200kHz; if the calculated If so, then the sensor signal will be retained; if Then the sensor signal will be discarded. By clearly defining the effective frequency band Matching the main frequency band of the Lamb wave and noise band or It can accurately distinguish between the effective components and interference components of a signal, preventing noise in non-target frequency bands from being misjudged as valid signals; setting of Thresholds, used as screening criteria, can quantitatively eliminate low-quality signals. when When the power of the effective component in the signal is significantly higher than that of the noise, it can completely preserve the characteristic information caused by the damage; when At times, the signal is severely contaminated by noise, and timely removal of such signals can avoid errors introduced by them.
[0027] Furthermore, step S04 also includes a step for quantifying the sensitivity of a common sensitive sensor: according to the formula... Calculate the overall sensitivity coefficient of each common sensing sensor. ;in, This represents the number of damage cases that are retained for this sensor in the damage dataset. , For the sensor in the first PE values in individual damage cases For the sensor in the first HFD values in individual damage cases, The baseline PE value for the damage-free dataset. The baseline values of HFD for the damage-free dataset. The standard deviation of the PE values in the damage-free dataset. The standard deviation of the HFD values in the damage-free dataset; according to Sort all commonly sensitive sensors by value from largest to smallest, and select the one with priority. The top 5 sensors are used to locate damage to the CFRP wing skin. By standardizing and averaging the deviations of each sensor's PE and HFD from the baseline in multiple damage cases, the dimensional differences between PE and HFD are eliminated, and the sensor's stable sensitivity to damage at different locations is comprehensively reflected. A higher S-value indicates that the sensor can accurately capture signal changes in more damage scenarios, and has stronger anti-interference and reliability. Prioritizing the selection of the top 5 sensors with the highest S-values allows us to focus on the core signal sources most sensitive to damage, avoiding redundant sensor signals from increasing the computational burden on the model. At the same time, the selected highly sensitive sensor signals can more clearly characterize damage features and reduce the interference of invalid data on the localization results.
[0028] Furthermore, in step S03, when calculating the Higuchi fractal dimension, a weighted linear fitting method is used to optimize the calculation accuracy. The specific process is as follows: For each segment number , , To determine the maximum number of segments, use the formula... Calculate the weight corresponding to the number of segments. ;in, The length of the curve under this number of segments. The smaller the value, the smoother the signal and the higher the data reliability. The larger; According to the formula Calculate the slope of the linear fit ;in, Number of segments The natural logarithm, Curve length The natural logarithm; The slope This is the optimized Higuchi fractal dimension, whose calculation error is reduced by 10% compared to traditional equal-weight linear fitting. 15% By dynamically assigning weights to improve the calculation accuracy of fractal dimension, the self-similarity characteristics of Lamb wave signals can be captured more accurately, providing a more reliable quantitative basis for dual-index screening. This method assigns weights based on the curve length under different numbers of segments. The smaller the curve length The smoother the signal, the higher the data reliability. Larger weights allow the fitting process to focus more on the contribution of high-quality data points, reducing the impact of outliers caused by overly fine segmentation or noise interference. The slope calculated using the weighted formula has a 10% lower error compared to traditional equal-weighted fitting. The 15% improvement more accurately reflects the signal self-similarity changes caused by CFRP wing skin damage. This optimization not only solves the accuracy loss problem caused by uneven data quality in traditional fitting, but also enhances the sensitivity of the HFD index to minor damage.
[0029] Furthermore, the model parameters of the CFRP wing skin in S01 are as follows: Dimensions: Length, width, and thickness are designed according to the actual application dimensions of CFRP wing skin; the radius of curvature includes four different specifications, namely the first specification. radius of curvature Second specification radius of curvature Third specification radius of curvature Fourth specification radius of curvature This corresponds to the root, middle, wingtip transition area, and wingtip area of the CFRP wing skin. Ply structure: The ply angle sequence is as follows It consists of 6 layers; each layer has a uniform thickness, with a single layer thickness of 0.125mm and a total thickness of 0.75mm. Material parameters: density is Tensile modulus , , Poisson's ratio , , shear modulus , , A simulation and experimental foundation highly aligned with actual engineering applications was constructed to ensure that the acquired Lamb wave signals accurately reflect the characteristic changes caused by damage, providing precise scenario adaptability for subsequent signal optimization and damage localization. In terms of dimensional design, the length, width, and thickness were determined according to actual application dimensions, and four different radii of curvature were set for the wing root, middle, wingtip transition zone, and wingtip region, fully replicating the influence of curved surface structures on Lamb wave propagation and avoiding signal distortion caused by model simplification; the ply structure adopts... The six-layer design and uniform single-layer thickness, matched to the actual production process of CFRP skin, ensure that the interlayer interaction during wave propagation is consistent with reality. Precisely set material parameters such as density, tensile modulus, Poisson's ratio, and shear modulus guarantee the accuracy of calculations for physical properties such as Lamb wave propagation velocity and mode transitions. The clear definition of these detailed parameters allows the entire signal acquisition and optimization process to be built on a realistic and reliable structural model, effectively improving the reliability of subsequent dual-index screening and damage feature extraction. This ensures that the final damage location results can directly guide practical engineering applications, avoiding difficulties in implementing technical solutions due to differences between the model and the actual structure.
[0030] Furthermore, it also includes a quality assessment step for the optimized signal: According to the formula Calculate the damage feature identification degree of the optimized signal ;in, The number of sensors in the common sensitive sensor set ; For the first The average signal of each sensitive sensor under damaged conditions. For the first The standard deviation of the signal from a sensitive sensor under damaged conditions; For the first The average value of the signal from each sensitive sensor in a non-damaged state. For the first The standard deviation of the signal from a sensitive sensor in a non-damaged state; If the calculation yields If the optimized signal is deemed to be of acceptable quality and can be used for subsequent damage localization; if Then the embedding dimension in S02 is readjusted. Adjust within a range of 37, maximum number of segments Adjust within the range of 520, or the statistical significance coefficient in S03. , Adjustments were made within a range of 13, and step 24 was repeated until a qualified optimized signal was obtained. This established a closed-loop control mechanism for signal quality, ensuring that the output optimized signal possesses high damage identification accuracy, providing the ultimate guarantee for the accuracy of subsequent damage localization. The damage feature identification degree D was calculated using a specific formula, comprehensively considering the mean difference and dispersion of the signals from each sensitive sensor in damaged and undamaged states. This quantitatively reflects the optimized signal's ability to characterize damage features. The qualified criteria can effectively screen out signals with clear damage characteristics and low noise interference, preventing low-quality signals from entering the subsequent positioning process. By adjusting the embedding dimension, maximum number of segments, or statistical significance coefficient within a preset range and repeating the signal optimization process, problems such as insufficient feature extraction and unreasonable screening thresholds can be specifically addressed, achieving dynamic calibration of signal optimization parameters. This closed-loop evaluation and adjustment mechanism not only compensates for the lack of quality verification after traditional signal screening but also significantly improves the robustness and adaptability of the entire technical solution. It ensures that high-quality signals meeting positioning requirements can be output regardless of the damage scenario, avoiding positioning errors caused by signal quality issues and providing reliable quality assurance for the engineering application of CFRP wing skin damage positioning.
[0031] I. Experimental Preparation Stage (I) Construction of CFRP Wing Skin Model A CFRP wing skin model consistent with actual engineering applications was built using ABAQUS software. Specific parameters are as follows: Dimensions and Curvature: The model's length and width are designed according to the actual wing skin, with a total skin thickness of 0.75mm. It includes four curvature radius specifications (corresponding to the wing root, middle, wingtip transition area, and wingtip area), namely... , , , ; Layered structure: using The 6-layer layup design has a single layer thickness of 0.125mm, and the fiber orientation is defined by a reference coordinate system. Material parameters: density Tensile modulus , Poisson's ratio shear modulus ; Boundary conditions: The wing root is completely fixed, and the interlayer constraints are set to bound contact to ensure that the simulation is a real stress state.
[0032] (ii) Sensor Network Deployment The system includes one transmitting sensor. and 26 receiving sensors The network, sensor placement locations are as follows: Figure 2 As shown, the transmitting sensor is responsible for emitting the excitation wave, while the receiving sensors are evenly distributed on the wing skin surface, covering all preset damage areas to ensure complete acquisition of Lamb wave signals caused by damage at different locations.
[0033] (III) Damage Setting Forty-five internal cracks were placed at different locations on the upper half of the curved surface of the wing skin. Each crack had a uniform size of 25 mm in length, 4 mm in width, and 3 layers (corresponding to layers 3-5). The center coordinates of some of the crack locations are as follows:
[0034] (iv) Determination of excitation wave parameters By comparison , , , The excitation wave response signals of four center frequencies are as follows Figure 3 As shown, it was found The 5-peak Lamb wave signal has the best resolution, the first arriving wave packet is clearly distinguishable, and the mode aliasing phenomenon is the least. Therefore, this parameter was selected as the excitation wave. Center frequency ; The waveform is a 5-peak sinusoidal modulated signal, and its expression is: , where T is the period; The excitation amplitude is fixed at 10V.
[0035] (v) Signal acquisition parameter settings Sampling frequency: 100kHz, to ensure that the subtle fluctuation characteristics of the Lamb wave can be captured; Analysis step duration: First, follow the formula The low-frequency velocity of mode A0 in the flat plate is calculated to be 1400.28 m / s, and then corrected using the curvature correction formula. (The γ values corresponding to different curvatures are 0.0012, 0.0015, 0.0018, and 0.0021, respectively.) The propagation velocity in the curved panel is obtained, and combined with the distance d from the transmitting sensor to the farthest receiving sensor, the analysis step time is calculated. The final setting is s, to ensure complete acquisition of the Lamb wave waveform propagating to the farthest sensor.
[0036] II. Signal Acquisition and Dataset Construction (a) Non-destructive dataset acquisition Control the transmission sensor A 100kHz excitation wave is emitted, and Lamb wave signals from 26 receiving sensors are acquired simultaneously. The acquisition is repeated 5 times to reduce random noise interference. The acquired signals are then merged and processed to form a damage-free dataset.
[0037] (II) Damage Data Collection Internal cracks were sequentially set at 45 preset damage locations. For each damage location, the above excitation and acquisition process was repeated to collect signals from 26 receiving sensors. Five sets of valid data were collected for each damage location, ultimately forming a damage dataset containing 45 damage locations and 26 channels of signals per set.
[0038] III. Signal Optimization Process (a) Step S02: Calculation of baseline value and screening threshold Parameter settings: The embedding dimension m is dynamically adjusted according to the number of sampling points N. In this experiment... Corresponding analysis step duration s, sampling frequency Therefore, take ;Maximum number of segments Statistical significance coefficient ; Permutation entropy (PE) baseline value calculation: Time-domain signals of the damage-free dataset according to Divided into Subsequences; Sort each subsequence to obtain permutation patterns, calculate the probability of occurrence of all patterns, and then apply the formula... Calculate and obtain the baseline PE value of the damage-free dataset. Standard deviation ; The PE screening threshold is determined as follows: ±2× That is, 0.985~0.9865; Higuchi fractal dimension HFD baseline value calculation: For the non-destructive signal, segment it sequentially according to the segment number k = 1~15, and then use the formula... Calculate the length of each curve segment, where , ; For lnk and lnL(k) Perform weighted linear fitting, weights Obtain the HFD baseline value Standard deviation ; Determine the HFD screening threshold as follows: ±2× That is, 0.214~0.246.
[0039] (II) Step S03: Dynamic threshold screening and validity verification Dual-index screening: For each 26-channel signal at each damage location, the PE value and HFD value are calculated separately. If the following criteria are met... or If so, the signal is initially preserved; Signal-to-noise ratio verification: Calculate the signal-to-noise ratio of the initially retained signal. ,in for Frequency band power, for or Frequency band power. Settings. To meet the qualification standard, signals with low signal-to-noise ratio are eliminated.
[0040] Taking damage location 1 as an example, the screening results are as follows: the signals of sensors 6, 22, 24, 25, and 26 meet the PE or HFD threshold conditions, and their SNRs are 18.2dB, 17.5dB, 16.8dB, 19.1dB, and 17.9dB, respectively, all greater than 15dB. Therefore, the signals of these 5 sensors are retained; the signals of the remaining 21 sensors are removed because either the indicators do not deviate from the threshold or the SNR does not meet the standard.
[0041] (III) Step S04: Screening of common sensitive sensors and signal output Sensitive sensor statistics: The screening results of 45 damage locations were statistically analyzed, and sensors that were retained in at least 35 damage cases were selected. Finally, sensors 6, 22, 24, 25, and 26 were identified as common sensitive sensors. Sensitivity quantification: according to the formula The overall sensitivity coefficient was calculated, where M is the number of damage cases retained for each sensor, all ≥35. The calculated S values for the five sensors are: sensor 6, S=3.2, sensor 22, S=2.9, sensor 24, S=3.0, sensor 25, S=3.5, and sensor 26, S=3.1. After sorting by S values, the signals from the top five sensors were selected as the optimized signals. Signal quality assessment: according to the formula Calculate the damage feature identification score, where Q=5. , These are the mean and standard deviation of the damage signal, respectively. , These represent the mean and standard deviation of the undamaged signal, respectively. In this experiment... This indicates that the optimized signal quality is acceptable and can be used for subsequent damage localization.
[0042] IV. Experimental Results and Verification (I) Verification of signal optimization effect Comparing the signal characteristics before and after optimization, the optimized signal, taking sensor 25 as an example, shows a significant reduction in noise, clearer signal abrupt changes caused by damage, and an increase in the signal-to-noise ratio of the time-domain waveform from 12.3dB to 19.1dB, effectively highlighting the damage characteristics.
[0043] (II) Subsequent verification of damage localization application The optimized signal was input into a Deeply Randomized Network (DSCN) for damage localization testing, and the results were compared with those of the unoptimized signal. Referring also to the model comparison data in the disclosure document, the following results were obtained:
[0044] Experimental results show that the signal optimized by the method of this invention can significantly improve the accuracy and stability of damage localization, and reduce the mean localization error. Both RMSE and MAE have reached industry-leading levels, fully verifying the effectiveness and practicality of the technical solution of this invention.
[0045] This is merely an embodiment of the present invention. Commonly known structures and characteristics of the solutions are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the filing date or priority date, are aware of all prior art in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, based on the guidance provided in this application, improve and implement this solution in combination with their own capabilities. Some typical well-known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the description of specific embodiments in the specification can be used to interpret the content of the claims.
Claims
1. A method for CFRP wing skin sensor signal optimization based on two-index screening, characterized in that, Comprising the following steps: S01 Constructing a CFRP wing skin multi-sensor network, the sensor network comprising transmitting sensors and receiving sensors; collecting Lamb wave signals of the CFRP wing skin under the non-damage state of the sensor network to form a non-damage data set; collecting Lamb wave signals of the CFRP wing skin at at least 45 different internal crack damage positions to form a damage data set; S02 Set two calculation parameters: embedding dimension and maximum segment number; calculate baseline values of permutation entropy (PE) and Higuchi fractal dimension (HFD) based on the undamaged dataset ; calculate standard deviations of all PE values and all HFD values of the undamaged dataset ; determine PE screening threshold and HFD screening threshold according to a preset statistical significance threshold; the maximum segment number refers to the maximum number of segments when the Lamb wave time-domain signal is divided into sub-segments, and the value range is 5-20; S03 For each damage position in the damage data set, the PE value and the HFD value of each sensor signal are calculated respectively; The dynamic threshold strategy is used to screen the sensor signals: if the PE value of a sensor signal satisfies , or the HFD value of the sensor signal satisfies , the sensor signal is retained; wherein, , are statistical significance coefficients, the value range is 1-3 and . S04 Statistics of all damage positions in the damage data set; filter out the sensors that are retained in at least 35 damage position cases to form a common sensitive sensor set; output the Lamb wave signal corresponding to the common sensitive sensor set.
2. The dual-indicator screening based CFRP wing skin sensor signal optimization method according to claim 1, characterized in that, The specific process of calculating the permutation entropy PE in S02 is: The embedding dimension is set as , the value range is 3-7, the delay time is 1, the Lamb wave time domain signal in the non-invasive data set is recorded as , wherein , the value of , is the sampling point number of the Lamb wave time domain signal, which is determined by the sampling frequency 100 kHz and the analysis step length, and ensures complete acquisition of the Lamb wave propagation waveform; according to the form of , the is divided into subsequences with a length of , wherein , , the value of ; Sort each sub-sequence in ascending order according to the numerical value of the elements in the sub-sequence to obtain the permutation pattern of the sub-sequence , , The expression of the permutation pattern of the sub-sequence is , wherein represents the index of the smallest element in the sub-sequence in the atomic sequence, and the value of is 1, 2 ; Count the occurrence of all sub-sequence permutation patterns to obtain the total number of different permutation patterns , represent the factorial of , and calculate the occurrence probability of each permutation pattern , , the value of , , the calculation method of ; The permutation entropy PE is calculated by the formula PE is maximum when the occurrence probability of all permutation patterns is equal PE is minimum 0 when the Lamb wave time domain signal is completely regular.
3. The dual-indicator screening based CFRP wing skin sensor signal optimization method according to claim 1, characterized in that, The specific process of calculating the Higuchi fractal dimension HFD in S02 is: The maximum segment number is set as , and the value range is 5-20; the Lamb wave time domain signal in the non-destructive data set is recorded as , , where n is the sampling point number; the value of each segment number is 1, 2 , that is, the value is taken from 1 to the maximum segment number in turn, and the signal is divided into segments, the expression of the signal in the th segment is , where the value of is 1, 2 , , represents the integer part of ; The curve length of each segment signal is calculated according to the formula ; wherein, , , () represents an integer operation, is the sampling interval of the Lamb wave signal, determined by the sampling frequency 100 kHz, . As the abscissa, As the ordinate, Linear fitting was performed on all The slope of the fitted straight line is the Higuchi fractal dimension HFD; the value range of HFD is 1~2.
4. The dual-indicator screening based CFRP wing skin sensor signal optimization method according to claim 1, characterized in that, The embedded dimension in the S02 is in the range of 37, and the specific value is determined according to the sampling point number N of the Lamb wave signal: when the embedded dimension is 3-5; when the embedded dimension is 5-7, and the total number of sub-sequences .
5. The dual-indicator screening based CFRP wing skin sensor signal optimization method according to claim 1, wherein, When collecting Lamb wave signals in S01, the sampling frequency is set to 100 kHz; the propagation speed of Lamb wave in the curved surface plate of CFRP wing skin is calculated as follows: First, follow the formula Calculate the low-frequency velocity of the A0 mode of the Lamb wave in a CFRP flat plate. ;in, The tensile modulus of CFRP material along the fiber direction. The density of CFRP material Poisson's ratio of CFRP material along the 1-2 direction Poisson's ratio of CFRP material along the 2-1 direction; According to the formula Calculate the propagation speed of Lamb wave in the CFRP curved panel ; wherein, is the curvature correction coefficient, which is determined according to the actual curvature of the CFRP wing skin: when the curvature radius is the first specification, ; when the curvature radius is the second specification, ; when the curvature radius is the third specification, ; when the curvature radius is the fourth specification, ; Measuring the straight-line distance from a transmitting sensor to a farthest receiving sensor in a network of sensors , the analysis step length of the Lamb wave signal is determined according to the formula , ensuring that the acquired Lamb wave signal completely contains the waveform propagated to the farthest receiving sensor. 6. The dual-indicator screen based CFRP wing skin sensor signal optimization method of claim 1, wherein, The dynamic threshold strategy in the S03 further comprises a signal validity verification step: for the initially reserved sensor signal, calculate its signal-to-noise ratio , The calculation formula is Wherein, is the power of the effective frequency band in the sensor signal, 80 kHz~120 kHz, corresponding to the main frequency band of the Lamb wave signal, is the power of the noise frequency band <50 kHz or >200 kHz in the sensor signal; if the calculated , then finally reserve the sensor signal; if , then reject the sensor signal.
7. The dual-indicator screen based CFRP wing skin sensor signal optimization method of claim 1, wherein, The step 4 further comprises a co-sensitive sensor sensitivity quantification step: according to the formula Calculate the comprehensive sensitivity coefficient of each co-sensitive sensor ; wherein, is the number of damage cases reserved for the sensor in the damage data set, , is the PE value of the sensor in the first damage case, is the HFD value of the sensor in the first damage case, is the PE baseline value of the non-damage data set, is the HFD baseline value of the non-damage data set, is the standard deviation of the PE value of the non-damage data set, is the standard deviation of the HFD value of the non-damage data set; according to , all co-sensitive sensors are sorted from large to small, and the sensors with the top values are selected first, and the corresponding signals are used for subsequent CFRP wing skin damage positioning.
8. The dual-indicator screen based CFRP wing skin sensor signal optimization method of claim 3, wherein, When calculating the Higuchi fractal dimension in S03, the weighted linear fitting method is used to optimize the calculation accuracy, and the specific process is as follows: For each segment number ( , (Maximum number of segments), according to the formula Calculate the weight corresponding to the number of segments. ;in, The length of the curve under this number of segments. The smaller the value, the smoother the signal and the higher the data reliability. The larger; According to the formula Calculate the slope of the linear fit ;in, Number of segments The natural logarithm, Curve length The natural logarithm; The slope That is, the optimized Higuchi fractal dimension, the calculation error is reduced by 10% compared with the traditional equal weight linear fitting 15%.
9. The dual-indicator screen based CFRP wing skin sensor signal optimization method of claim 1, wherein, The model parameters of the CFRP wing skin in S01 are: Size: length, width, thickness according to the actual application size of CFRP wing skin design; curvature radius contains four different specifications, respectively, the first specification Curvature radius , the second specification Curvature radius , the third specification Curvature radius , the fourth specification Curvature radius , corresponding to the root, middle, wing tip transition area, wing tip area of CFRP wing skin; Layup structure: Layup angle sequence is , 6 layers in total; the thickness of each layer is uniform, the thickness of a single layer is 0.125 mm, and the total thickness is 0.75 mm; Material parameters: density ; tensile modulus , , ; Poisson's ratio , , ; shear modulus , , .
10. The dual-indicator screen based CFRP wing skin sensor signal optimization method of claim 1, wherein, It also includes the quality evaluation step of the optimized signal: According to the formula Calculate the damage feature identification degree of the optimized signal ;in, The number of sensors in the common sensitive sensor set ; For the first The average signal of each sensitive sensor under damaged conditions. For the first The standard deviation of the signal from a sensitive sensor under damaged conditions; For the first The average value of the signal from each sensitive sensor in a non-damaged state. For the first The standard deviation of the signal from a sensitive sensor in a non-damaged state; If the calculated value is , it is determined that the quality of the optimized signal is qualified and can be used for subsequent damage positioning; if , the embedding dimension in S02 is adjusted again The maximum number of segments is adjusted in the range of 37 The statistical significance coefficient in S03 is adjusted in the range of 520 , The value is adjusted in the range of 13, and step 24 is repeated until a qualified optimized signal is obtained.