Environment noise adaptive multi-spectrum stress wave-acoustic print damage detection method and system

Abstract

This invention discloses an environmental noise-adaptive multi-spectral stress wave-acoustic fingerprint damage detection method and system, belonging to the field of non-destructive testing technology. The method constructs a passive-active hybrid excitation, using environmental noise as the passive source combined with an active signal, and adaptively weights this to form an enhanced excitation; it simultaneously acquires signals from 20-150kHz across multiple frequency bands and filters out interference; it constructs a three-dimensional acoustic fingerprint map including a benchmark and calibration; and it achieves defect identification and quantification through acoustic impedance tensor inversion, difference algorithms, and a Bayesian framework. The system includes acquisition, acoustic fingerprint construction, intelligent analysis, and monitoring and early warning modules, which collaboratively complete the entire detection process. This invention solves the limitations of single excitation, environmental interference, and the challenges of detecting complex structures, improving detection sensitivity, accuracy, and environmental adaptability. It is applicable to damage detection of metal and composite material components and has significant engineering value.

Description

Technical Field

[0001] This invention relates to the field of nondestructive testing technology, and in particular to an environmental noise-adaptive multi-spectral stress wave-acoustic print damage detection method and system. Background Technology

[0002] In fields such as industrial non-destructive testing, aerospace, petrochemicals, rail transportation, nuclear power equipment, and intelligent manufacturing, equipment structural health monitoring and defect detection are key links to ensure production safety and extend equipment life. Currently, the mainstream testing technology is based on traditional ultrasonic testing, but this type of technology and its derivatives have significant limitations in practical applications and are difficult to adapt to the complex needs of industrial scenarios.

[0003] From the perspective of technological application characteristics, existing detection technologies generally treat noise in industrial environments as interference signals. Traditional ultrasonic testing requires the construction of a quiet testing environment (such as an isolated workshop or soundproofing devices) to operate, which severely limits the testing scenarios. At the same time, most technologies adopt a single-band detection mode. Due to the limitations of the physical characteristics of sound waves, single-band signals cannot simultaneously achieve both penetration depth and detection resolution. Low-frequency signals can penetrate thicker materials but are difficult to identify microscopic defects, while high-frequency signals have high resolution but weak penetration ability, directly causing the loss of information on multi-scale defects (such as macroscopic structural cracks and surface micro-damage). In addition, existing technologies mostly rely on shutdown testing modes, which require interrupting the production process to create stable testing conditions, making it impossible to achieve real-time monitoring while the equipment is running. In the defect judgment stage, traditional methods mainly rely on manual experience to set fixed thresholds, which have poor adaptability to detection signals under different materials and operating conditions. False alarms or false alarms are easily caused by threshold deviations, further reducing the reliability of the detection.

[0004] From the perspective of practical application shortcomings, the limitations of existing technologies have created multiple constraints: First, stringent environmental requirements significantly increase testing costs, requiring additional investment in the purchase, installation, and maintenance of soundproofing equipment, and making it unsuitable for special industrial scenarios such as outdoor, high-temperature, and high-noise environments; Second, insufficient information utilization leads to inadequate testing coverage, as single-band signals can only capture defects at a certain scale, making it difficult to comprehensively assess the overall health status of equipment. For example, in the inspection of metal components, macroscopic cracks in weld areas and micro-corrosion on the surface may be missed; Third, the shutdown testing mode severely impacts production efficiency, especially in the inspection of large equipment such as aero-engines and nuclear power reactors, where a single shutdown can cause economic losses ranging from tens of thousands to millions of yuan; Fourth, low levels of intelligence lead to strong reliance on operator experience, and differences in judgment standards among different personnel further amplify testing errors; Fifth, insufficient quantitative capabilities, with existing technologies mostly only able to qualitatively determine the existence of defects, making it difficult to accurately assess key parameters such as defect size, depth, and propagation rate, thus failing to provide precise data support for equipment maintenance.

[0005] In summary, the shortcomings of existing non-destructive testing technologies in terms of environmental adaptability, testing efficiency, intelligence level, and quantitative capabilities can no longer meet the demands for "online, real-time, accurate, and intelligent" equipment testing in the context of Industry 4.0. There is an urgent need for a new testing solution that can overcome the limitations of traditional technologies. Summary of the Invention

[0006] The purpose of this invention is to provide an environmental noise-adaptive multi-spectral stress wave-acoustic damage detection method and system, which overcomes the limitations of traditional detection methods such as environmental interference, lack of single-frequency band information, insufficient quantitative capability, and reliance on downtime. This enables online damage identification, quantitative assessment, and life prediction of metal and composite material components, meeting the real-time, accurate, and intelligent requirements of industrial equipment health monitoring.

[0007] To achieve the above objectives, this invention provides an environmental noise-adaptive multi-spectral stress wave-acoustic print damage detection method, the steps of which are as follows:

[0008] S1: Construct a passive-active hybrid excitation, using ambient noise as a passive broadband excitation source, combined with an active excitation signal swept by 20-150kHz standard white noise, and adjust the signal ratio of ambient noise and active excitation through an adaptive weighting function to form an enhanced excitation signal;

[0009] S2: Multi-band signal acquisition, using a multi-channel synchronous acquisition device with FPGA as the clock synchronization controller, to acquire multi-spectral acoustic emission stress wave signals under enhanced excitation, synchronously acquire multi-band acoustic emission stress wave signals under enhanced excitation, and filter out fixed frequency and harmonic interference.

[0010] S3: Voiceprint map construction. First, the factory baseline voiceprint of the object to be tested is collected in a standard environment. The frequency response and acoustic impedance baseline characteristics are obtained simultaneously. Then, the voiceprint calibration is completed by compensation model in combination with the actual working conditions after installation. Based on the calibrated voiceprint data, a three-dimensional voiceprint map is constructed by radial basis function interpolation.

[0011] S4: Defect and Anomaly Identification: Tensor invariants are calculated by inverting the acoustic impedance tensor field. After data validity verification and filtering, multi-resolution analysis is used to extract signal features and construct high-dimensional feature vectors. Defects and anomalies are identified and anomaly levels are output through the vector distribution difference algorithm.

[0012] S5. Defect Type Judgment and Quantitative Assessment: Based on the statistical framework, the defect type is determined for the abnormal feature vector. The defect size is quantitatively assessed by combining acoustic empirical formulas. The assessment results are optimized by allocating weights according to the spatial stress distribution of the object being inspected.

[0013] Preferably, in S1, the enhanced excitation signal Represented as:

[0014]

[0015] in, For environmental noise signals, As an active incentive signal, For adaptive weighting functions, when the ambient noise level is ≥60dB When the ambient noise intensity is <60dB .

[0016] Preferably, in S2, the multi-channel synchronous acquisition device adopts an orthogonal arrangement strategy, with sensors arranged perpendicular to the surface of the object to capture longitudinal waves, and sensors arranged tangentially to the surface to capture transverse waves. A triaxial sensor group is used at the stress concentration points of the object. The sensor spacing... Based on the spatial sampling theorem c represents the sound velocity of the material being tested. The highest detection frequency is set to 150kHz, and the sensor spacing for steel objects is set to 15mm.

[0017] Multi-band acoustic emission stress wave signals include low-frequency (20-40kHz), mid-frequency (40-80kHz), and high-frequency (80-150kHz) signals.

[0018] Specifically, filtering out fixed interference frequency band signals involves filtering out 50Hz and its integer multiples of harmonics, and collecting the host power switching frequency and its integer multiples of harmonics.

[0019] Preferably, S3 specifically includes;

[0020] S3.1. Collect the factory baseline acoustic signature of the test object under standard conditions:

[0021] Standard environment refers to a noise-free environment with a temperature of 20±2°C, relative humidity of 50±10%, and background noise of <40dB.

[0022] During active excitation, the response is measured by stepping 1kHz or 500Hz and synchronizing the frequency. Represented as:

[0023]

[0024] For the input excitation spectrum, To output the response spectrum, the synchronously measured acoustic impedance reference value is expressed as:

[0025]

[0026] in, The density of the material being tested is given by c, the speed of sound is given by R(f), and the reflection coefficient is given by R(f).

[0027] The impedance mean, impedance standard deviation, skewness, and kurtosis are extracted as baseline features, among which:

[0028] The mean impedance is expressed as: ;

[0029] The standard deviation of impedance is expressed as: ;

[0030] Skewness is expressed as: ;

[0031] Kurtosis is expressed as: ;

[0032] S3.2. Based on the actual working conditions after installation, the voiceprint calibration is completed using a compensation model:

[0033] The compensation model is expressed as:

[0034]

[0035] in, The measured acoustic impedance after installation and the installation stress compensation factor are given. , Installation stress value; boundary condition compensation coefficient. f is a function of boundary impedance; temperature compensation coefficient T represents the actual operating temperature. Reference temperature; The value range is 0.001. 0.005 MPa; The range of values ​​is 0.002 0.001 / ℃;

[0036] S3.3. Construct a three-dimensional voiceprint map based on the calibrated voiceprint data using radial basis function interpolation:

[0037] The formula for radial basis function interpolation is:

[0038]

[0039] in, For thin plate spline functions, The coefficients are undetermined and are solved using measurement point data. For spatial points With measurement point The Euclidean distance;

[0040] The interpolation process satisfies the boundary conditions: , , .

[0041] Preferably, S4 specifically includes:

[0042] S4.1. Tensor invariants are calculated by inverting the acoustic impedance tensor field. The tensor model is expressed as:

[0043]

[0044] in, The diagonal element represents the resistance of pressure in direction i to velocity in direction j. , , Main impedance components, off-diagonal elements , , , , , This is the coupling impedance component;

[0045] Isotropic Defect-Free Materials ,and hour ;

[0046] The impedance change of the defective material is: ;

[0047] The defect characteristic parameters satisfy: At that time, it was determined that there might be a defect;

[0048] At that time, it was determined to be a void-type defect;

[0049] When this occurs, it is determined to be an anisotropic defect;

[0050] The tensor invariants include:

[0051]

[0052]

[0053]

[0054] in, As the first invariant, It is the second invariant; It is the third invariant;

[0055] The criterion is expressed as follows:

[0056]

[0057] in, It is the first invariant of the current tensor. is the first invariant of the baseline tensor; threshold is the defect judgment threshold determined by actual measurement, and the threshold for metallic materials is set to 0.08-0.12;

[0058] S4.2 Perform data validity verification on tensor invariants and related characteristic parameters:

[0059] Set the confirmation window to N=10 consecutive sampling points, and calculate the consistency criterion as follows:

[0060]

[0061] Where MAD is the median absolute deviation, expressed as: ;

[0062] The confirmation rules are as follows: data is considered valid when C ≥ 0.8, data is considered pending and requires additional sampling when 0.5 ≤ C < 0.8, and data is considered invalid and discarded when C < 0.5.

[0063] S4.3. Perform Kalman filtering on the confirmed valid data to optimize the stability of the defective data. The state equation is as follows:

[0064]

[0065] The observation equation is:

[0066]

[0067] Wherein, the state transition matrix , The sampling interval;

[0068] Observation matrix ;

[0069] Process noise covariance ;

[0070] The measurement noise covariance R = 1.0;

[0071] Filter gain P is the state covariance matrix;

[0072] S4.4. Wavelet packet decomposition is used to perform time-frequency analysis on the filtered acoustic emission stress wave. The formula is as follows:

[0073]

[0074] Where j is the number of decomposition levels, ranging from 1 to 5; k is the frequency subband index; and h is the wavelet filter coefficient, using db4, sym4, or coif4 wavelet types.

[0075] The formula for calculating the energy distribution of each sub-band is: This forms an energy feature vector containing the energy of each sub-band, represented as: ,

[0076] The energy feature vector is fused with time-domain features, frequency-domain features, and acoustic impedance features to form a high-dimensional feature vector.

[0077] S4.5, based on the high-dimensional feature vector constructed in S4.4, identifies defects and anomalies using a vector distribution difference algorithm, and outputs the anomaly level:

[0078] Vector distribution difference algorithms include KL divergence, Wasserstein distance, or Mahalanobis distance:

[0079] (1) The criterion for judging KL divergence is: when D KL When the value is less than the preset normal threshold, it is considered to be in a normal state, indicating that the current feature distribution is not significantly different from the defect-free baseline distribution; when the preset normal threshold is less than or equal to D... KL When the value is less than the preset minor anomaly threshold, it is considered a minor anomaly, indicating a slight deviation between the current feature distribution and the baseline distribution; when the preset minor anomaly threshold is less than or equal to D... KL When the value is less than the preset moderate anomaly threshold, it is judged as moderately abnormal, indicating that the current feature distribution deviates significantly from the baseline distribution; when D... KL When the value is greater than or equal to the preset moderate anomaly threshold, it is judged as a severe anomaly, indicating that the current feature distribution differs greatly from the baseline distribution;

[0080] (2) The criteria for judging Wasserstein distance are as follows: when W < the preset normal distance threshold, it is judged as normal, indicating that the optimal matching cost between the current feature and the benchmark feature is extremely low; when the preset normal distance threshold ≤ W < the preset slight anomaly distance threshold, it is judged as slightly anomaly, indicating that there is a small matching deviation between the current feature and the benchmark feature; when the preset slight anomaly distance threshold ≤ W, it is judged as slightly anomaly. W When W < the preset moderate anomaly distance threshold, it is judged as moderate anomaly, indicating that the current feature deviates significantly from the baseline feature; when W ≥ the preset moderate anomaly distance threshold, it is judged as severe anomaly, indicating that the current feature deviates significantly from the baseline feature.

[0081] (3) The criterion for judging the Mahalanobis distance is: D M The condition is considered normal when the chi-square distribution quantile corresponding to feature dimension p is less than or equal to the preset confidence level. M >It is judged as an anomaly when the chi-square distribution quantile of the corresponding feature dimension p is set at the pre-set confidence level.

[0082] Preferably, S5 specifically includes:

[0083] S5.1. For feature vectors identified as abnormal in S4.5, the defect type is determined based on a multidimensional Bayesian framework:

[0084] The prior probabilities of the multidimensional Bayesian framework are set based on historical detection data: P(crack) = 0.45, P(corrosion) = 0.30, P(delamination) = 0.15, P(void) = 0.10;

[0085] The likelihood function is used as follows:

[0086]

[0087] Parameters are estimated using the maximum likelihood method. , This is a mean estimate; , For variance estimation;

[0088] Posterior probability The formula for determining the defect type is: ;

[0089] S5.2. Quantitatively assess the size of the defect types determined in S5.1 using acoustic empirical formulas:

[0090] The acoustic empirical formulas include:

[0091] Crack length formula: , For the corresponding frequency band detection wavelength, The rate of change of impedance;

[0092] Crack depth formula: , For defect echo time difference;

[0093] The formula for crack width is: AR is an empirical aspect ratio, ranging from 0.1 to 0.3, and is adjusted according to the material hardness.

[0094] S5.3. Optimize the quantitative evaluation results of S5.2 by allocating weights according to the spatial stress distribution of the tested object:

[0095] Stress weighting is assigned using a weighting function:

[0096]

[0097] in, The spatial stress distribution function of the object being tested. This is the maximum stress value. This is the weighting coefficient, with a value ranging from 0.5 to 2.0.

[0098] Preferably, in S5.2, the defect size assessment error range is: length error ±10%, depth error ±15%, and width error ±20%.

[0099] Preferably, in S5.3, the typical region weights for stress weight allocation are: weld zone weight w=2.5, bending zone weight w=2.0, support point weight w=1.8, and general region weight w=1.0; the weight coefficients The value is adjusted according to the strength of the material being tested, for high-strength alloy materials. Values ​​range from 1.5 to 2.0 for common metallic materials. Values ​​range from 0.5 to 1.0 for composite materials. Values ​​range from 1.0 to 1.5.

[0100] An environmental noise-adaptive multi-spectral stress wave-acoustic print damage detection system includes:

[0101] The system includes a multi-spectral acoustic emission stress wave acquisition module, an acoustic pattern map construction module, an intelligent analysis and diagnosis module, and a real-time monitoring and early warning module, which are sequentially connected by signals.

[0102] The multi-spectral acoustic emission stress wave acquisition module is used to acquire passive and active environmental noise excitation signals. By adjusting the proportion of the two through an adaptive weighting function, a mixed excitation signal is formed. Multi-band acoustic emission stress wave signals under mixed excitation are acquired simultaneously and interference signals are filtered out.

[0103] The acoustic signature map construction module is used to receive the signal output by the multi-spectral acoustic emission stress wave acquisition module, acquire the factory baseline acoustic signature of the detection object and complete the acoustic signature calibration in combination with the actual working conditions, construct a three-dimensional acoustic signature map based on the calibrated signal and calculate the acoustic impedance tensor field.

[0104] The intelligent analysis and diagnosis module is used to extract feature parameters from the 3D acoustic map, identify defects and anomalies through the vector distribution difference algorithm, determine the defect type based on the statistical framework, and quantitatively evaluate the defect size by combining acoustic empirical formulas.

[0105] The real-time monitoring and early warning module receives defect data output by the intelligent analysis and diagnosis module, filters and verifies the validity of the data, analyzes the development trend of defects and predicts the remaining lifespan, calculates the overall confidence level based on preset weight allocation, and triggers graded early warnings.

[0106] Preferably, the multi-spectral acoustic emission stress wave acquisition module includes a multi-channel synchronous acquisition unit, a frequency band separation unit, and a noise filtering unit, specifically:

[0107] Multi-channel synchronous acquisition unit: It adopts a 12-channel hardware architecture, uses FPGA as the clock synchronization controller, with a clock accuracy of ±10ns, and is equipped with a 24-bit ADC chip with a dynamic range of >120dB to realize time synchronous acquisition of multi-channel signals at a sampling rate of 1MHz.

[0108] Frequency band separation unit: Through real-time FFT frequency division technology, it synchronously separates the acoustic emission stress wave signals of low frequency 20-40kHz, mid frequency 40-80kHz, and high frequency 80-150kHz, and stores and transmits each frequency band independently;

[0109] Noise filtering unit: Employs an IIR notch filter to filter out 50Hz and its harmonics, and collects the host power switching frequency and its harmonics, with a filter cutoff attenuation ≥40dB;

[0110] The voiceprint map construction module includes a multi-level spectrum mapping unit, an acoustic impedance tensor calculation unit, and a spatial field reconstruction unit, specifically:

[0111] Multi-level spectrum mapping unit: receives multi-band acoustic emission stress wave signals, collects factory reference spectrum characteristics under standard environment, and performs calibration in combination with actual working conditions after installation. Actual working conditions include temperature, stress, and boundary conditions, eliminating the influence of environmental differences on spectrum characteristics, and establishing a multi-level mapping relationship between frequency band, spatial location, and spectrum parameters.

[0112] Acoustic impedance tensor calculation unit: Based on the calibrated acoustic impedance data output by the multi-level spectrum mapping unit, an acoustic impedance tensor model is constructed at each point in space. The model includes principal impedance components and coupled impedance components. The principal impedance components are diagonal elements, and the coupled impedance components are off-diagonal elements. The first, second, and third invariants of the tensor are calculated as core feature parameters.

[0113] Spatial field reconstruction unit: Based on acoustic impedance tensor invariants as the basic data, spatial field reconstruction is performed using radial basis function interpolation. By setting boundary conditions, the interpolation accuracy is ensured, and finally a three-dimensional acoustic texture map covering the entire space of the detected object is constructed to realize the spatial continuous distribution representation of acoustic impedance characteristics.

[0114] The intelligent analysis and diagnosis module includes a feature extraction unit, an anomaly identification unit, and a quantitative evaluation unit, specifically:

[0115] Feature extraction unit: Extracts time-domain features, frequency-domain features, and acoustic impedance features from the voiceprint map. The time-domain features include eight items: root mean square value, peak value, peak factor, kurtosis, skewness, waveform factor, impulse factor, and margin factor. The frequency-domain features include spectral centroid, spectral width, spectral entropy, spectral peak frequency, spectral peak amplitude, spectral roll-off point, spectral flatness, spectral irregularity, dominant frequency ratio, and harmonic distortion. The acoustic impedance features include six items: impedance mean, impedance gradient, impedance anisotropy, impedance nonlinearity, impedance phase delay, and impedance quality factor, forming a high-dimensional feature vector.

[0116] The feature extraction unit also includes a multi-resolution analysis module, which uses wavelet packet decomposition to perform time-frequency decomposition on the acoustic emission stress wave signal to obtain an energy feature vector containing the energy of each sub-band. The energy feature vector, together with the time domain features, frequency domain features, and acoustic impedance features, constitutes a high-dimensional feature vector.

[0117] Anomaly identification unit: Built-in KL divergence, Wasserstein distance, and Mahalanobis distance calculation models, automatically selects the appropriate vector distribution difference algorithm to identify defects and anomalies, and outputs the anomaly level, which includes normal, slight, moderate, and severe.

[0118] Quantitative evaluation unit: Integrates a multidimensional Bayesian statistical framework and an acoustic empirical formula library. After inputting feature vectors, it automatically calculates the probability of defect type and size parameters, and optimizes the evaluation results by combining stress weight allocation.

[0119] The real-time monitoring and early warning module includes a data stream processing unit, a trend analysis unit, and an alarm decision unit, specifically:

[0120] Data stream processing unit: integrates Kalman filtering algorithm and delay confirmation mechanism to filter and denoise diagnostic data and confirm its validity, and output stable defect data;

[0121] Trend analysis unit: Built-in exponential degradation model, power-law degradation model, and linear-exponential hybrid model.

[0122] The expression for the exponential degradation model is: :

[0123] The expression for the power-law degeneracy model is: ;

[0124] The expression for the linear-exponential mixed model is: ;

[0125] The optimal model is selected based on the AIC criterion, and the expression for the AIC criterion is as follows: , where k is the number of model parameters and L is the likelihood value;

[0126] Predicting Defect Development Trends and Remaining Life , The failure threshold Output the 95% confidence interval for the current time. , Obtained through Monte Carlo simulation;

[0127] Alarm decision unit: Preset weight allocation, with weights of defect probability 0.3, defect size 0.3, growth trend 0.25, and location importance 0.15, based on overall confidence level. Triggering a tiered warning, Let be the confidence level of the i-th criterion. For the corresponding weights, When Conf≥0.8, a Level 1 alarm is triggered, which is for emergency handling. When 0.5≤Conf<0.8, a Level 2 alarm is triggered, which is for planned maintenance. When Conf<0.5, a Level 3 alarm is triggered, which is for continuous monitoring.

[0128] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0129] 1. Enhanced Detection Sensitivity and Early Defect Identification Capability: By employing a passive-active hybrid excitation mode, using environmental noise as a natural broadband excitation source, and combining it with an active frequency sweep signal, the proportion of both is dynamically adjusted through an adaptive weighting function, solving the problem of insufficient energy or incomplete frequency band coverage of a single excitation signal. Simultaneous acquisition of 20-150kHz signals across multiple frequency bands can capture characteristic signals of defects of different sizes. In particular, the high-frequency band of 80-150kHz significantly improves the response sensitivity to micron-level cracks, enabling effective identification of early-stage, minute defects.

[0130] 2. Improved detection accuracy and anti-interference capability: The multi-channel synchronous acquisition device ensures the spatiotemporal consistency of the signal, and the IIR notch filter specifically filters out 50Hz and its harmonics, as well as power switching frequency interference, effectively improving the signal-to-noise ratio. The acoustic texture map construction process incorporates temperature, stress, and boundary condition compensation models, significantly reducing errors caused by differences between installation conditions and standard environments, providing a stable benchmark for defect judgment.

[0131] 3. Enhanced ability to identify complex structures and various types of defects: The 3D acoustic pattern map, combined with acoustic impedance tensor field inversion, characterizes the anisotropic features of the material through tensor invariants and coupled components, effectively distinguishing different types of defects such as cracks, corrosion, and voids. Compared to traditional scalar impedance analysis, the accuracy of defect location for complex structural components is significantly improved.

[0132] 4. Achieving high precision and reliability in quantitative defect assessment: Based on a multi-dimensional Bayesian framework that integrates multi-band features and combines KL divergence, Mahalanobis distance, and other difference algorithms, accurate quantitative determination of defect levels is achieved. A stress weight allocation mechanism is introduced to reasonably increase the assessment weight of defects in high-stress areas, matching actual failure risks. The combination of acoustic empirical formulas and statistical models improves the accuracy of defect size assessment, meeting the quantitative detection needs of engineering projects.

[0133] In summary, this invention significantly improves the sensitivity, accuracy, and adaptability of acoustic emission stress wave detection in complex industrial environments through technological innovations such as multi-excitation fusion, multi-spectral analysis, tensor field inversion, and intelligent decision-making. It can be widely applied to damage detection of metal structures, composite material components, etc., and has important engineering practical value.

[0134] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0135] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0136] Figure 1 This is a flowchart illustrating an embodiment of the environmental noise-adaptive multi-spectral stress wave-acoustic print damage detection method of the present invention.

[0137] Figure 2 This is a schematic diagram of the structure of the environmental noise-adaptive multi-spectral stress wave-acoustic damage detection system according to an embodiment of the present invention. Detailed Implementation

[0138] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0139] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0140] Example

[0141] like Figure 1 As shown, the environmental noise-adaptive multi-spectral stress wave-acoustic print damage detection method comprises the following steps:

[0142] S1: Construct a passive-active hybrid excitation, using ambient noise as a passive broadband excitation source, combined with an active excitation signal swept by 20-150kHz standard white noise, and adjust the signal ratio of ambient noise and active excitation through an adaptive weighting function to form an enhanced excitation signal;

[0143] Enhance incentive signals Represented as:

[0144]

[0145] in, For environmental noise signals, As an active incentive signal, For adaptive weighting functions, when the ambient noise level is ≥60dB When the ambient noise intensity is <60dB .

[0146] S2: Multi-band signal acquisition, using a multi-channel synchronous acquisition device with an FPGA as the clock synchronization controller, to acquire multi-spectral acoustic emission stress wave signals under enhanced excitation, synchronously separate low-frequency (20-40kHz), mid-frequency (40-80kHz), and high-frequency (80-150kHz) signals, and filter out fixed interference frequency band signals;

[0147] The multi-channel synchronous acquisition device adopts an orthogonal arrangement strategy. Sensors are arranged perpendicular to the surface of the object to capture longitudinal waves, and sensors are arranged tangentially to the horizontal surface to capture transverse waves. A triaxial sensor group is used at the stress concentration point of the object.

[0148] Sensor spacing Based on the spatial sampling theorem c represents the sound velocity of the material being tested. The highest detection frequency is set to 150kHz, and the sensor spacing for steel objects is set to 15mm.

[0149] Filtering out fixed interference frequency band signals specifically involves: filtering out 50Hz and its integer multiples of harmonics, and collecting the host power switching frequency and its integer multiples of harmonics.

[0150] S3: Voiceprint map construction. First, the factory baseline voiceprint of the object to be tested is collected under standard conditions. Then, the voiceprint is calibrated using a compensation model based on the actual working conditions after installation. A three-dimensional voiceprint map is constructed based on the calibrated voiceprint data using radial basis function interpolation. Specifically, this includes:

[0151] S3.1 Collect the factory baseline acoustic print of the test object under standard environment.

[0152] Standard environment refers to a noise-free environment with a temperature of 20±2°C, relative humidity of 50±10%, and background noise of <40dB.

[0153] During active excitation, the response is measured by stepping 1kHz or 500Hz and synchronizing the frequency. Represented as:

[0154]

[0155] For the input excitation spectrum, To output the response spectrum, the synchronously measured acoustic impedance reference value is expressed as:

[0156]

[0157] in, The density of the material being tested is given by c, the speed of sound is given by R(f), and the reflection coefficient is given by R(f).

[0158] The impedance mean, impedance standard deviation, skewness, and kurtosis are extracted as baseline features, among which:

[0159] The mean impedance is expressed as: ;

[0160] The standard deviation of impedance is expressed as: ;

[0161] Skewness is expressed as: ;

[0162] Kurtosis is expressed as:

[0163] S3.2. Based on the actual working conditions after installation, the voiceprint calibration is completed through the compensation model.

[0164] The compensation model is expressed as:

[0165]

[0166] in, The measured acoustic impedance after installation and the installation stress compensation factor are given. , Installation stress value; boundary condition compensation coefficient. f is a function of boundary impedance; temperature compensation coefficient T represents the actual operating temperature. Reference temperature; The value range is 0.001. 0.005 MPa; The range of values ​​is 0.002 0.001 / ℃;

[0167] S3.3. A three-dimensional voiceprint map is constructed based on the calibrated voiceprint data using radial basis function interpolation.

[0168] The formula for radial basis function interpolation is:

[0169]

[0170] in, For thin plate spline functions, The coefficients are undetermined and are solved using measurement point data. For spatial points With measurement point The Euclidean distance;

[0171] The interpolation process satisfies the boundary conditions: , , .

[0172] S4: Defect and Anomaly Identification: Tensor invariants are calculated through acoustic impedance tensor field inversion. After data validity verification and filtering, multi-resolution analysis is used to extract signal features and construct high-dimensional feature vectors. Defects and anomalies are identified using a vector distribution difference algorithm, and the anomaly level is output. Specifically, this includes:

[0173] S4.1. Tensor invariants are calculated by inverting the acoustic impedance tensor field. The tensor model is expressed as:

[0174]

[0175] in, The diagonal element represents the resistance of pressure in direction i to velocity in direction j. , , Main impedance components, off-diagonal elements , , , , , This is the coupling impedance component;

[0176] Isotropic Defect-Free Materials ,and hour ;

[0177] The impedance change of the defective material is: ;

[0178] The defect characteristic parameters satisfy: At that time, it was determined that there might be a defect;

[0179] At that time, it was determined to be a void-type defect;

[0180] When this occurs, it is determined to be an anisotropic defect;

[0181] The tensor invariants include:

[0182]

[0183]

[0184]

[0185] in, As the first invariant, It is the second invariant; It is the third invariant;

[0186] The criterion is expressed as follows:

[0187]

[0188] in, It is the first invariant of the current tensor. is the first invariant of the baseline tensor; threshold is the defect judgment threshold determined by actual measurement, and the threshold for metallic materials is set to 0.08-0.12.

[0189] S4.2 Perform data validity verification on tensor invariants and related characteristic parameters:

[0190] Set the confirmation window to N=10 consecutive sampling points, and calculate the consistency criterion as follows:

[0191]

[0192] Where MAD is the median absolute deviation, expressed as: ;

[0193] The confirmation rules are as follows: data is considered valid when C ≥ 0.8, data is considered pending and requires additional sampling when 0.5 ≤ C < 0.8, and data is considered invalid and discarded when C < 0.5.

[0194] S4.3. Perform Kalman filtering on the confirmed valid data to optimize the stability of the defective data. The state equation is as follows:

[0195]

[0196] The observation equation is:

[0197]

[0198] Wherein, the state transition matrix , The sampling interval;

[0199] Observation matrix ;

[0200] Process noise covariance ;

[0201] The measurement noise covariance R = 1.0;

[0202] Filter gain P is the state covariance matrix;

[0203] S4.4. Wavelet packet decomposition is used to perform time-frequency analysis on the filtered acoustic emission stress wave. The formula is as follows:

[0204]

[0205] Where j is the number of decomposition levels, ranging from 1 to 5; k is the frequency subband index; and h is the wavelet filter coefficient, using db4, sym4, or coif4 wavelet types.

[0206] The formula for calculating the energy distribution of each sub-band is: This forms an energy feature vector containing the energy of each sub-band, represented as: ,

[0207] The energy feature vector is fused with time-domain features, frequency-domain features, and acoustic impedance features to form a high-dimensional feature vector.

[0208] S4.5, based on the high-dimensional feature vector constructed in S4.4, identifies defects and anomalies using a vector distribution difference algorithm, and outputs the anomaly level:

[0209] (1) The formula for calculating the KL divergence is:

[0210]

[0211] in, Let P represent the probability value of distribution P in the i-th discrete dimension, satisfying , Let Q represent the probability value of distribution Q in the i-th discrete dimension, satisfying ;

[0212] The judgment criteria are as follows: when DKL < 0.1, it is judged as normal, and the object under inspection has no defect risk; when 0.1 ≤ DKL < 0.5, it is judged as slightly abnormal, indicating that there is a small matching deviation between the current feature and the reference feature, and there may be initial micro-damage; when 0.5 ≤ DKL < 1.0, it is judged as moderately abnormal, indicating that there is a significant matching deviation between the current feature and the reference feature, and there are defects that need attention (such as cracks, delamination, etc.); when DKL ≥ 1.0, it is judged as severely abnormal, indicating that there is a huge matching deviation between the current feature and the reference feature, and there are serious defects that threaten the structural safety.

[0213] (2) The formula for calculating the Wasserstein distance is:

[0214]

[0215] Where X is the single high-dimensional feature vector of the current detection object, which is the data of the multi-spectral acoustic emission signal collected in real time after processing by S4.1-S4.4, and Y is the single high-dimensional feature vector of the defect-free reference object, that is, the feature vector of the factory reference acoustic texture after processing by S3.1-S3.3. Let P represent the probability distribution of the random variable X. Let Q represent the probability distribution of the random variable Y;

[0216] The judgment criteria are as follows: W < 0.2 is considered normal, indicating no defect risk in the inspected object; 0.2 ≤ W < 0.5 is considered a minor abnormality, indicating potential initial micro-damage to the inspected object; 0.5 ≤ W < 0.5 is considered a minor abnormality. W When W < 0.8, it is judged as a moderate anomaly, and the object under inspection has defects that require attention (such as cracks, delamination, etc.); when W ≥ 0.8, it is judged as a severe anomaly, and the object under inspection has serious defects that threaten structural safety.

[0217] (3) The formula for calculating Mahalanobis distance is:

[0218]

[0219] Where x is the high-dimensional feature vector to be evaluated. The mean vector of the baseline feature set. Difference vector transpose, The covariance matrix of the baseline feature set,

[0220] The judgment criteria are: The time frame is determined to be normal, and the inspected object has no defects; If an anomaly is detected, the object being tested is deemed to have a defect risk. p is the feature dimension of the high-dimensional feature vector. The value is the 95th quantile of the chi-square distribution with p degrees of freedom.

[0221] S5. Defect Type Judgment and Quantitative Assessment: Based on the abnormal feature vector, the defect type is determined using a statistical framework. A quantitative assessment of the defect size is then performed using acoustic empirical formulas. Furthermore, the assessment results are optimized by assigning weights based on the spatial stress distribution of the inspected object. Specifically, this includes:

[0222] S5.1. For feature vectors identified as abnormal in S4.5, the defect type is determined based on a multidimensional Bayesian framework:

[0223] The prior probabilities of the multidimensional Bayesian framework are set based on historical detection data: P(crack) = 0.45, P(corrosion) = 0.30, P(delamination) = 0.15, P(void) = 0.10;

[0224] The likelihood function is used as follows:

[0225]

[0226] Parameters are estimated using the maximum likelihood method. , This is a mean estimate; , For variance estimation;

[0227] Posterior probability The formula for determining the defect type is: ;

[0228] S5.2. Quantitatively assess the size of the defect types determined in S5.1 using acoustic empirical formulas:

[0229] The acoustic empirical formulas include:

[0230] Crack length formula: , For the corresponding frequency band detection wavelength, The rate of change of impedance;

[0231] Crack depth formula: , For defect echo time difference;

[0232] The formula for crack width is: AR is an empirical aspect ratio, ranging from 0.1 to 0.3, and is adjusted according to the material hardness.

[0233] The defect size assessment error range is: length error ±10%, depth error ±15%, and width error ±20%.

[0234] S5.3. Optimize the quantitative evaluation results of S5.2 by allocating weights according to the spatial stress distribution of the tested object:

[0235] Stress weighting is assigned using a weighting function:

[0236]

[0237] in, The spatial stress distribution function of the object being tested. This is the maximum stress value. This is the weighting coefficient, with a value ranging from 0.5 to 2.0.

[0238] Typical stress weighting values ​​for different regions are: weld zone weight w=2.5, bending zone weight w=2.0, support point weight w=1.8, and general region weight w=1.0; weighting coefficients. The value is adjusted according to the strength of the material being tested, for high-strength alloy materials. Values ​​range from 1.5 to 2.0 for common metallic materials. Values ​​range from 0.5 to 1.0 for composite materials. Values ​​range from 1.0 to 1.5.

[0239] Environmental noise-adaptive multi-spectral stress wave-acoustic print damage detection system, such as Figure 2 As shown, it includes:

[0240] The system includes a multi-spectral acoustic emission stress wave acquisition module, an acoustic texture map construction module, an intelligent analysis and diagnosis module, and a real-time monitoring and early warning module, which are sequentially connected by signals.

[0241] (1) Multi-spectral acoustic emission stress wave acquisition module

[0242] This method is used to acquire passive and active excitation signals of environmental noise, adjusts the ratio of the two through an adaptive weighting function to form a hybrid excitation signal, and simultaneously acquires multi-band acoustic emission stress wave signals under hybrid excitation and filters out interference signals; specifically including:

[0243] Multi-channel synchronous acquisition unit: It adopts a 12-channel hardware architecture, uses FPGA as the clock synchronization controller, with a clock accuracy of ±10ns, and is equipped with a 24-bit ADC chip with a dynamic range of >120dB to realize time synchronous acquisition of multi-channel signals at a sampling rate of 1MHz.

[0244] Frequency band separation unit: Through real-time FFT frequency division technology, it synchronously separates the acoustic emission stress wave signals of low frequency 20-40kHz, mid frequency 40-80kHz, and high frequency 80-150kHz, and stores and transmits each frequency band independently;

[0245] Noise filtering unit: It adopts an IIR notch filter to filter out 50Hz and its harmonics, and collects the host power switching frequency and its harmonics. The filter cutoff attenuation is ≥40dB.

[0246] (2) Voiceprint Map Construction Module

[0247] This is used to receive the signal output from the multi-spectral acoustic emission stress wave acquisition module, acquire the factory-standard acoustic signature of the object being tested, and perform acoustic signature calibration based on actual working conditions. Based on the calibrated signal, a three-dimensional acoustic signature map is constructed, and the acoustic impedance tensor field is calculated. Specifically, this includes:

[0248] Multi-level spectrum mapping unit: receives multi-band acoustic emission stress wave signals, collects factory reference spectrum characteristics under standard environment, and performs calibration in combination with actual working conditions after installation. Actual working conditions include temperature, stress, and boundary conditions, eliminating the influence of environmental differences on spectrum characteristics, and establishing a multi-level mapping relationship between frequency band, spatial location, and spectrum parameters.

[0249] Acoustic impedance tensor calculation unit: Based on the calibrated acoustic impedance data output by the multi-level spectrum mapping unit, an acoustic impedance tensor model is constructed at each point in space. The model includes principal impedance components and coupled impedance components. The principal impedance components are diagonal elements, and the coupled impedance components are off-diagonal elements. The first, second, and third invariants of the tensor are calculated as core feature parameters.

[0250] Spatial field reconstruction unit: Based on acoustic impedance tensor invariants as the basic data, spatial field reconstruction is performed using radial basis function interpolation. By setting boundary conditions, the interpolation accuracy is ensured, and finally a three-dimensional acoustic texture map covering the entire space of the detected object is constructed to realize the spatial continuous distribution representation of acoustic impedance characteristics.

[0251] (3) Intelligent analysis and diagnosis module

[0252] This method is used to extract feature parameters from a 3D acoustic signature map, identify defects and anomalies using a vector distribution difference algorithm, determine defect types based on a statistical framework, and quantitatively assess defect size using acoustic empirical formulas; specifically, it includes:

[0253] Feature extraction unit: Extracts time-domain features, frequency-domain features, and acoustic impedance features from the voiceprint map. The time-domain features include eight items: root mean square value, peak value, peak factor, kurtosis, skewness, waveform factor, impulse factor, and margin factor. The frequency-domain features include spectral centroid, spectral width, spectral entropy, spectral peak frequency, spectral peak amplitude, spectral roll-off point, spectral flatness, spectral irregularity, dominant frequency ratio, and harmonic distortion. The acoustic impedance features include six items: impedance mean, impedance gradient, impedance anisotropy, impedance nonlinearity, impedance phase delay, and impedance quality factor, forming a high-dimensional feature vector.

[0254] The feature extraction unit also includes a multi-resolution analysis module, which uses wavelet packet decomposition to perform time-frequency decomposition on the acoustic emission stress wave signal to obtain an energy feature vector containing the energy of each sub-band. The energy feature vector, together with the time domain features, frequency domain features, and acoustic impedance features, constitutes a high-dimensional feature vector.

[0255] Anomaly identification unit: Built-in KL divergence, Wasserstein distance, and Mahalanobis distance calculation models, automatically selects the appropriate vector distribution difference algorithm to identify defects and anomalies, and outputs the anomaly level, which includes normal, slight, moderate, and severe.

[0256] Quantitative evaluation unit: Integrates a multidimensional Bayesian statistical framework and an acoustic empirical formula library. After inputting feature vectors, it automatically calculates the probability of defect type and size parameters, and optimizes the evaluation results by combining stress weight allocation.

[0257] (4) Real-time monitoring and early warning module

[0258] This system receives defect data output from the intelligent analysis and diagnosis module, filters and verifies the validity of the data, analyzes defect development trends and predicts remaining lifetime, calculates overall confidence based on preset weight allocation, and triggers tiered early warnings. Specifically, it includes:

[0259] Data stream processing unit: integrates Kalman filtering algorithm and delay confirmation mechanism to filter and denoise diagnostic data and confirm its validity, and output stable defect data;

[0260] Trend analysis unit: Built-in exponential degradation model, power-law degradation model, and linear-exponential hybrid model.

[0261] The expression for the exponential degradation model is: :

[0262] The expression for the power-law degeneracy model is: ;

[0263] The expression for the linear-exponential mixed model is: ;

[0264] The optimal model is selected based on the AIC criterion, and the expression for the AIC criterion is as follows: , where k is the number of model parameters and L is the likelihood value;

[0265] Predicting Defect Development Trends and Remaining Life , The failure threshold Output the 95% confidence interval for the current time. , Obtained through Monte Carlo simulation;

[0266] Alarm decision unit: Preset weight allocation, with weights of defect probability 0.3, defect size 0.3, growth trend 0.25, and location importance 0.15, based on overall confidence level. Triggering a tiered warning, Let be the confidence level of the i-th criterion. For the corresponding weights, When Conf≥0.8, a Level 1 alarm is triggered, which is for emergency handling. When 0.5≤Conf<0.8, a Level 2 alarm is triggered, which is for planned maintenance. When Conf<0.5, a Level 3 alarm is triggered, which is for continuous monitoring.

[0267] The remaining technical features in the above embodiments can be flexibly selected by those skilled in the art to meet different specific practical needs according to actual circumstances. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims. In the above description, numerous specific details have been set forth to provide a thorough understanding of the present invention. However, it will be apparent to those skilled in the art that these specific details are not necessary to implement the present invention. In other instances, to avoid obscuring the present invention, well-known techniques, such as specific construction details, operating conditions, and other technical conditions, have not been specifically described.

[0268] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. An environmental noise-adaptive multi-spectral stress wave-acoustic print damage detection method, characterized in that, The steps are as follows: S1: Construct a passive-active hybrid excitation, using ambient noise as a passive broadband excitation source, combined with an active excitation signal swept by 20-150kHz standard white noise, and adjust the signal ratio of ambient noise and active excitation through an adaptive weighting function to form an enhanced excitation signal; S2: Multi-band signal acquisition, using a multi-channel synchronous acquisition device with FPGA as the clock synchronization controller, to acquire multi-spectral acoustic emission stress wave signals under enhanced excitation, synchronously acquire multi-band acoustic emission stress wave signals under enhanced excitation, and filter out fixed frequency and harmonic interference. S3: Voiceprint map construction. First, the factory baseline voiceprint of the object to be tested is collected under a standard environment. The frequency response and acoustic impedance baseline characteristics are acquired simultaneously. Then, the voiceprint is calibrated by a compensation model in combination with the actual working conditions after installation. Based on the calibrated voiceprint data, a three-dimensional voiceprint map is constructed using radial basis function interpolation. Specifically, it includes: S3.

1. Collect the factory baseline acoustic signature of the test object under standard conditions: Standard environment refers to a noise-free environment with a temperature of 20±2°C, relative humidity of 50±10%, and background noise of <40dB. During active excitation, the response is measured by stepping 1kHz or 500Hz and synchronizing the frequency. Represented as: For the input excitation spectrum, To output the response spectrum, the synchronously measured acoustic impedance reference value is expressed as: in, The density of the material being tested is given by c, the speed of sound is given by R(f), and the reflection coefficient is given by R(f). The impedance mean, impedance standard deviation, skewness, and kurtosis are extracted as baseline features, among which: The mean impedance is expressed as: ; The standard deviation of impedance is expressed as: ; Skewness is expressed as: ; Kurtosis is expressed as: ; S3.

2. Based on the actual working conditions after installation, the voiceprint calibration is completed using a compensation model: The compensation model is expressed as: in, The measured acoustic impedance after installation and the installation stress compensation factor are given. , Installation stress value; boundary condition compensation coefficient. f is a function of boundary impedance; temperature compensation coefficient T represents the actual operating temperature. Reference temperature; The value range is 0.

001. 0.005 MPa; The range of values ​​is 0.002 0.001 / ℃; S3.

3. Construct a three-dimensional voiceprint map based on the calibrated voiceprint data using radial basis function interpolation: The formula for radial basis function interpolation is: in, For thin plate spline functions, The coefficients are undetermined and are solved using measurement point data. For spatial points With measurement point The Euclidean distance; The interpolation process satisfies the boundary conditions: , , ; S4: Defect and Anomaly Identification: Tensor invariants are calculated through acoustic impedance tensor field inversion. After data validity verification and filtering, multi-resolution analysis is used to extract signal features and construct high-dimensional feature vectors. Defects and anomalies are identified using a vector distribution difference algorithm, and the anomaly level is output. Specifically, this includes: S4.

1. Tensor invariants are calculated by inverting the acoustic impedance tensor field. The tensor model is expressed as: in, The diagonal element represents the resistance of pressure in direction i to velocity in direction j. , , Main impedance components, off-diagonal elements , , , , , This is the coupling impedance component; Isotropic Defect-Free Materials ,and hour ; The impedance change of the defective material is: ; The defect characteristic parameters satisfy: At that time, it was determined that there might be a defect; At that time, it was determined to be a void-type defect; When this occurs, it is determined to be an anisotropic defect; The tensor invariants include: in, As the first invariant, It is the second invariant; It is the third invariant; The criterion is expressed as follows: in, It is the first invariant of the current tensor. is the first invariant of the baseline tensor; threshold is the defect judgment threshold determined by actual measurement, and the threshold for metallic materials is set to 0.08-0.12; S4.2 Perform data validity verification on tensor invariants and related characteristic parameters: Set the confirmation window to N=10 consecutive sampling points, and calculate the consistency criterion as follows: Where MAD is the median absolute deviation, expressed as: ; The confirmation rules are as follows: data is considered valid when C ≥ 0.8, data is considered pending and requires additional sampling when 0.5 ≤ C < 0.8, and data is considered invalid and discarded when C < 0.

5. S4.

3. Perform Kalman filtering on the confirmed valid data to optimize the stability of the defective data. The state equation is as follows: The observation equation is: Wherein, the state transition matrix , The sampling interval; Observation matrix ; Process noise covariance ; The measurement noise covariance R = 1.0; Filter gain P is the state covariance matrix; S4.

4. Wavelet packet decomposition is used to perform time-frequency analysis on the filtered acoustic emission stress wave. The formula is as follows: Where j is the number of decomposition levels, ranging from 1 to 5; k is the frequency subband index; and h is the wavelet filter coefficient, using db4, sym4, or coif4 wavelet types. The formula for calculating the energy distribution of each sub-band is: This forms an energy feature vector containing the energy of each sub-band, represented as: , The energy feature vector is fused with time-domain features, frequency-domain features, and acoustic impedance features to form a high-dimensional feature vector. S4.5, based on the high-dimensional feature vector constructed in S4.4, identifies defects and anomalies using a vector distribution difference algorithm, and outputs the anomaly level: Vector distribution difference algorithms include KL divergence, Wasserstein distance, or Mahalanobis distance: (1) The criterion for judging KL divergence is: when D KL When the value is less than the preset normal threshold, it is considered to be in a normal state, indicating that the current feature distribution is not significantly different from the defect-free baseline distribution; when the preset normal threshold is less than or equal to D... KL When the value is less than the preset minor anomaly threshold, it is considered a minor anomaly, indicating a slight deviation between the current feature distribution and the baseline distribution; when the preset minor anomaly threshold is less than or equal to D... KL When the value is less than the preset moderate anomaly threshold, it is judged as moderately abnormal, indicating that the current feature distribution deviates significantly from the baseline distribution; when D... KL When the value is greater than or equal to the preset moderate anomaly threshold, it is judged as a severe anomaly, indicating that the current feature distribution differs greatly from the baseline distribution; (2) The criteria for judging Wasserstein distance are as follows: when W < the preset normal distance threshold, it is judged as normal, indicating that the optimal matching cost between the current feature and the benchmark feature is extremely low; when the preset normal distance threshold ≤ W < the preset slight anomaly distance threshold, it is judged as slightly anomaly, indicating that there is a small matching deviation between the current feature and the benchmark feature; when the preset slight anomaly distance threshold ≤ W, it is judged as slightly anomaly. W When W < the preset moderate anomaly distance threshold, it is judged as moderate anomaly, indicating that the current feature deviates significantly from the baseline feature; when W ≥ the preset moderate anomaly distance threshold, it is judged as severe anomaly, indicating that the current feature deviates significantly from the baseline feature. (3) The criterion for determining Mahalanobis distance is: D M The condition is considered normal when the chi-square distribution quantile corresponding to feature dimension p is less than or equal to the preset confidence level. M >It is judged as an anomaly when the chi-square distribution quantile of the corresponding feature dimension p is under the preset confidence level; S5. Defect Type Judgment and Quantitative Assessment: Based on the statistical framework, the defect type is determined for the abnormal feature vector. The defect size is quantitatively assessed by combining acoustic empirical formulas. The assessment results are optimized by allocating weights according to the spatial stress distribution of the object being inspected.

2. The environmental noise-adaptive multi-spectral stress wave-acoustic print damage detection method according to claim 1, characterized in that: In S1, the enhanced excitation signal Represented as: in, For environmental noise signals, As an active incentive signal, For adaptive weighting functions, when the ambient noise level is ≥60dB When the ambient noise intensity is <60dB .

3. The environmental noise-adaptive multi-spectral stress wave-acoustic print damage detection method according to claim 2, characterized in that: In S2, the multi-channel synchronous acquisition device adopts an orthogonal arrangement strategy, with sensors arranged perpendicular to the surface of the object to capture longitudinal waves, and sensors arranged tangentially to the horizontal surface to capture transverse waves. A triaxial sensor group is used at the stress concentration points of the object. The sensor spacing... Based on the spatial sampling theorem c represents the sound velocity of the material being tested. The highest detection frequency is set to 150kHz, and the sensor spacing for steel objects is set to 15mm. Multi-band acoustic emission stress wave signals include low-frequency (20-40kHz), mid-frequency (40-80kHz), and high-frequency (80-150kHz) signals. Filtering out fixed interference frequency band signals specifically involves filtering out 50Hz and its integer multiples of harmonics, and collecting the host power switching frequency and its integer multiples of harmonics.

4. The environmental noise-adaptive multi-spectral stress wave-acoustic print damage detection method according to claim 1, characterized in that: S5 specifically includes: S5.

1. For feature vectors identified as abnormal in S4.5, the defect type is determined based on a multidimensional Bayesian framework: The prior probabilities of the multidimensional Bayesian framework are set based on historical detection data: P(crack) = 0.45, P(corrosion) = 0.30, P(delamination) = 0.15, P(void) = 0.10; The likelihood function is used as follows: Parameters are estimated using the maximum likelihood method. , This is a mean estimate; , For variance estimation; Posterior probability The formula for determining the defect type is: ; S5.

2. Quantitatively assess the size of the defect types determined in S5.1 using acoustic empirical formulas: The acoustic empirical formulas include: Crack length formula: , For the corresponding frequency band detection wavelength, The rate of change of impedance; Crack depth formula: , For defect echo time difference; The formula for crack width is: AR is an empirical aspect ratio, ranging from 0.1 to 0.3, and is adjusted according to the material hardness. S5.

3. Optimize the quantitative evaluation results of S5.2 by allocating weights according to the spatial stress distribution of the tested object: Stress weighting is assigned using a weighting function: in, The spatial stress distribution function of the object being tested. This is the maximum stress value. This is the weighting coefficient, with a value ranging from 0.5 to 2.

0.

5. The environmental noise-adaptive multi-spectral stress wave-acoustic print damage detection method according to claim 1, characterized in that: In S5.2, the defect size assessment error range is: length error ±10%, depth error ±15%, and width error ±20%.

6. The environmental noise-adaptive multi-spectral stress wave-acoustic print damage detection method according to claim 4, characterized in that: In S5.3, the typical region weights for stress weight allocation are: weld zone weight w = 2.5, bending zone weight w = 2.0, support point weight w = 1.8, and general region weight w = 1.0; the weight coefficients The value is adjusted according to the strength of the material being tested, for high-strength alloy materials. Values ​​range from 1.5 to 2.0 for common metallic materials. Values ​​range from 0.5 to 1.0 for composite materials. Values ​​range from 1.0 to 1.

5.

7. An environmental noise-adaptive multi-spectral stress wave-acoustic print damage detection system, used to perform the method as described in any one of claims 1-6, characterized in that: It includes a multi-spectral acoustic emission stress wave acquisition module, an acoustic texture map construction module, an intelligent analysis and diagnosis module, and a real-time monitoring and early warning module, which are sequentially connected by signals. The multi-spectral acoustic emission stress wave acquisition module is used to acquire passive and active environmental noise excitation signals, adjust the ratio of the two through an adaptive weighting function to form a mixed excitation signal, and simultaneously acquire multi-band acoustic emission stress wave signals under mixed excitation and filter out interference signals. The acoustic signature map construction module is used to receive the signal output by the multi-spectral acoustic emission stress wave acquisition module, acquire the factory baseline acoustic signature of the detection object and complete the acoustic signature calibration in combination with the actual working conditions, construct a three-dimensional acoustic signature map based on the calibrated signal and calculate the acoustic impedance tensor field. The intelligent analysis and diagnosis module is used to extract feature parameters of the three-dimensional acoustic map, identify defects and anomalies through the vector distribution difference algorithm, determine the defect type based on the statistical framework, and quantitatively evaluate the defect size by combining acoustic empirical formulas. The real-time monitoring and early warning module is used to receive defect data output by the intelligent analysis and diagnosis module, filter and confirm the validity of the data, analyze the defect development trend and predict the remaining life, calculate the overall confidence level based on the preset weight allocation and trigger graded early warning.

8. The environmental noise-adaptive multi-spectral stress wave-acoustic print damage detection system according to claim 7, characterized in that: The multi-spectral acoustic emission stress wave acquisition module includes a multi-channel synchronous acquisition unit, a frequency band separation unit, and a noise filtering unit, specifically: Multi-channel synchronous acquisition unit: It adopts a 12-channel hardware architecture, uses FPGA as the clock synchronization controller, with a clock accuracy of ±10ns, and is equipped with a 24-bit ADC chip with a dynamic range of >120dB to realize time synchronous acquisition of multi-channel signals at a sampling rate of 1MHz. Frequency band separation unit: Through real-time FFT frequency division technology, it synchronously separates the acoustic emission stress wave signals of low frequency 20-40kHz, mid frequency 40-80kHz, and high frequency 80-150kHz, and stores and transmits each frequency band independently; Noise filtering unit: Employs an IIR notch filter to filter out 50Hz and its harmonics, and collects the host power switching frequency and its harmonics, with a filter cutoff attenuation ≥40dB; The voiceprint map construction module includes a multi-level spectrum mapping unit, an acoustic impedance tensor calculation unit, and a spatial field reconstruction unit, specifically: Multi-level spectrum mapping unit: receives multi-band acoustic emission stress wave signals, collects factory reference spectrum characteristics under standard environment, and performs calibration in combination with actual working conditions after installation. Actual working conditions include temperature, stress, and boundary conditions, eliminating the influence of environmental differences on spectrum characteristics, and establishing a multi-level mapping relationship between frequency band, spatial location, and spectrum parameters. Acoustic impedance tensor calculation unit: Based on the calibrated acoustic impedance data output by the multi-level spectrum mapping unit, an acoustic impedance tensor model is constructed at each point in space. The model includes principal impedance components and coupled impedance components. The principal impedance components are diagonal elements, and the coupled impedance components are off-diagonal elements. The first, second, and third invariants of the tensor are calculated as core feature parameters. Spatial field reconstruction unit: Based on acoustic impedance tensor invariants as the basic data, spatial field reconstruction is performed using radial basis function interpolation. By setting boundary conditions, the interpolation accuracy is ensured, and finally a three-dimensional acoustic texture map covering the entire space of the detected object is constructed to realize the spatial continuous distribution representation of acoustic impedance characteristics. The intelligent analysis and diagnosis module includes a feature extraction unit, an anomaly identification unit, and a quantitative evaluation unit, specifically: Feature extraction unit: Extracts time-domain features, frequency-domain features, and acoustic impedance features from the voiceprint map. The time-domain features include eight items: root mean square value, peak value, peak factor, kurtosis, skewness, waveform factor, impulse factor, and margin factor. The frequency-domain features include spectral centroid, spectral width, spectral entropy, spectral peak frequency, spectral peak amplitude, spectral roll-off point, spectral flatness, spectral irregularity, dominant frequency ratio, and harmonic distortion. The acoustic impedance features include six items: impedance mean, impedance gradient, impedance anisotropy, impedance nonlinearity, impedance phase delay, and impedance quality factor, forming a high-dimensional feature vector. The feature extraction unit also includes a multi-resolution analysis module, which uses wavelet packet decomposition to perform time-frequency decomposition on the acoustic emission stress wave signal to obtain an energy feature vector containing the energy of each sub-band. The energy feature vector, together with the time domain features, frequency domain features, and acoustic impedance features, constitutes a high-dimensional feature vector. Anomaly identification unit: Built-in KL divergence, Wasserstein distance, and Mahalanobis distance calculation models, automatically selects the appropriate vector distribution difference algorithm to identify defects and anomalies, and outputs the anomaly level, which includes normal, slight, moderate, and severe. Quantitative evaluation unit: Integrates a multidimensional Bayesian statistical framework and an acoustic empirical formula library. After inputting feature vectors, it automatically calculates the probability of defect type and size parameters, and optimizes the evaluation results by combining stress weight allocation. The real-time monitoring and early warning module includes a data stream processing unit, a trend analysis unit, and an alarm decision unit, specifically: Data stream processing unit: integrates Kalman filtering algorithm and delay confirmation mechanism to filter and denoise diagnostic data and confirm its validity, and output stable defect data; Trend analysis unit: Built-in exponential degradation model, power-law degradation model, and linear-exponential hybrid model. The expression for the exponential degradation model is: : The expression for the power-law degeneracy model is: ; The expression for the linear-exponential mixed model is: ; The optimal model is selected based on the AIC criterion, and the expression for the AIC criterion is as follows: , where k is the number of model parameters and L is the likelihood value; Predicting Defect Development Trends and Remaining Life , The failure threshold Output the 95% confidence interval for the current time. , Obtained through Monte Carlo simulation; Alarm decision unit: Preset weight allocation, with weights of defect probability 0.3, defect size 0.3, growth trend 0.25, and location importance 0.15, based on overall confidence level. Triggering a tiered warning, Let be the confidence level of the i-th criterion. For the corresponding weights, When Conf≥0.8, a Level 1 alarm is triggered, which is for emergency handling. When 0.5≤Conf<0.8, a Level 2 alarm is triggered, which is for planned maintenance. When Conf<0.5, a Level 3 alarm is triggered, which is for continuous monitoring.

Images