Search

CN-121577757-B - Environmental noise adaptive multi-frequency spectrum stress wave-voiceprint damage detection method and system

CN121577757BCN 121577757 BCN121577757 BCN 121577757BCN-121577757-B

Abstract

The invention discloses an environment noise adaptive multi-frequency spectrum stress wave-voiceprint damage detection method and system, and relates to the technical field of nondestructive detection. The method comprises the steps of constructing passive-active mixed excitation, combining an active signal with environmental noise as a passive source, forming enhanced excitation through self-adaptive weight adjustment, synchronously acquiring 20-150kHz signals by multiple frequency bands, filtering interference, constructing a three-dimensional voiceprint map containing references and calibration, and realizing defect identification and quantification through acoustic impedance tensor inversion, a difference algorithm and a Bayesian framework. The system comprises an acquisition module, a voiceprint construction module, an intelligent analysis module, a monitoring and early warning module and a cooperative complete flow detection module. The invention solves the problems of single excitation limitation, environmental interference and complex structure detection, improves the detection sensitivity, accuracy and environmental adaptability, is suitable for detecting the damage of metal and composite material components, and has important engineering value.

Inventors

  • LEI LIANG
  • LONG ZHENYU
  • WANG HONGYU
  • WANG YIPING
  • LI JIAOBO

Assignees

  • 北京通泰恒盛科技有限责任公司

Dates

Publication Date
20260505
Application Date
20251203

Claims (8)

  1. 1. The environmental noise adaptive multi-frequency spectrum stress wave-voiceprint damage detection method is characterized by comprising the following steps of: S1, constructing passive-active mixed excitation, taking environmental noise as a passive broadband excitation source, combining an active excitation signal of 20-150kHz standard white noise sweep frequency, and adjusting the signal duty ratio of the environmental noise and the active excitation through a self-adaptive weight function to form an enhanced excitation signal; S2, multi-frequency band signal acquisition, namely acquiring multi-frequency band acoustic emission stress wave signals under the action of enhanced excitation by adopting a multi-channel synchronous acquisition device and taking an FPGA (field programmable gate array) as a clock synchronous controller, synchronously acquiring the multi-frequency band acoustic emission stress wave signals under the action of enhanced excitation, and filtering fixed frequency and harmonic interference thereof; S3, voiceprint map construction, namely acquiring factory reference voiceprints of a detection object in a standard environment, synchronously acquiring frequency response and acoustic impedance reference characteristics, completing voiceprint calibration by a compensation model in combination with the installed actual working condition, and constructing a three-dimensional voiceprint map by adopting radial basis function interpolation based on calibrated voiceprint data; S3.1, collecting factory reference voiceprints of the detection objects in a standard environment: Standard environment refers to a non-interference environment with a temperature of 20 + -2 ℃ and a relative humidity of 50 + -10% and a background noise of <40 dB; when active excitation is performed, the step is 1kHz or 500Hz, and the synchronous frequency measurement response is performed Expressed as: for the input of the excitation spectrum, For the output response spectrum, the synchronization measurement acoustic reactance reference value is expressed as: Wherein, the For detecting the density of the object material, c is the sound velocity, and R (f) is the reflection coefficient; Extracting an impedance mean value, an impedance standard deviation, skewness and kurtosis as reference characteristics, wherein: The impedance mean is expressed as: ; The standard deviation of the impedance is expressed as: ; The skewness is expressed as: ; Kurtosis is expressed as: ; S3.2, completing voiceprint calibration through a compensation model by combining the installed actual working conditions: The compensation model is expressed as: Wherein, the To measure acoustic impedance after installation, a stress compensation coefficient is installed , For mounting stress values, boundary condition compensation coefficients F is a function of boundary impedance, temperature compensation coefficient T is the temperature of the actual working condition, Is the reference temperature; The range of the value of (2) is 0.001 0.005/MPa; The range of the values is as follows 0.002 0.001/°C; S3.3, constructing a three-dimensional voiceprint map by adopting radial basis function interpolation based on the calibrated voiceprint data: The radial basis function interpolation formula is as follows: Wherein, the As a function of the thin-plate spline, For the coefficients to be determined, the data is solved by measuring points, Is a space point And measuring point Is the euclidean distance of (2); the interpolation process satisfies the boundary conditions: 、 、 ; S4, defect anomaly identification, namely, calculating tensor invariants through acoustic impedance tensor field inversion, extracting signal characteristics by adopting multi-resolution analysis and constructing high-dimensional characteristic vectors after data validity confirmation and filtering treatment, and identifying defect anomalies and outputting anomaly grades through a vector distribution difference algorithm, wherein the method specifically comprises the following steps of: s4.1, calculating tensor invariants through acoustic impedance tensor field inversion, wherein a tensor model is expressed as: Wherein, the Impedance of i-direction pressure to j-direction velocity, diagonal element 、 、 As the principal impedance component, off-diagonal elements 、 、 、 、 、 For coupling the impedance component; Of isotropic defect-free material And (2) and Time of day ; The amount of change in resistance of the defect-containing material is: ; the defect characteristic parameters satisfy: when it is determined that a defect is likely to exist; Judging that the defect is a cavity defect; When the anisotropic defect is judged; the tensor invariant includes: Wherein, the As a first invariant quantity, the first value is, Is a second invariant; Is a third invariant; The criteria are expressed as: Wherein, the For the first invariant of the current tensor, The threshold is a defect judgment threshold which is actually measured and determined, and the metal material threshold is set to be 0.08-0.12; S4.2, carrying out data validity confirmation on tensor invariants and related characteristic parameters: setting a confirmation window n=10 consecutive sampling points, and calculating a consistency criterion as follows: Wherein, MAD is the median absolute deviation, expressed as: ; The validation rules are: judging that the data is valid when C is more than or equal to 0.8, judging that the data is pending when C is more than or equal to 0.5 and less than or equal to 0.8, and carrying out supplementary sampling, judging that the data is invalid and rejecting when C is less than or equal to 0.5; s4.3, performing Kalman filtering processing on the data confirmed to be effective, wherein the Kalman filtering processing is used for optimizing the stability of the defect data, and the state equation is as follows: The observation equation is: Wherein the state transition matrix , Is the sampling interval; observation matrix ; Process noise covariance ; Measurement noise covariance r=1.0; Filtering gain P is a state covariance matrix; S4.4, carrying out time-frequency analysis on the filtered acoustic emission stress wave by adopting wavelet packet decomposition, wherein the formula is as follows: Wherein j is the number of decomposition layers and takes 1-5 layers, k is the index of a frequency sub-band, h is the coefficient of a wavelet filter, and db4, sym4 or coif4 wavelet types are selected; the formula for calculating the energy distribution of each sub-band is: an energy feature vector including each subband energy is formed, expressed as: , fusing the energy characteristic vector with the time domain characteristic, the frequency domain characteristic and the acoustic impedance characteristic to jointly form a high-dimensional characteristic vector; s4.5, identifying defect abnormality through a vector distribution difference algorithm based on the high-dimensional feature vector constructed in the S4.4, and outputting an abnormality grade: the vector distribution difference algorithm includes KL divergence, wasserstein distance, or Markov distance: (1) The KL divergence judgment criterion is that when D KL is smaller than a preset normal threshold, the current characteristic distribution is judged to be in a normal state, no significant difference is shown between the current characteristic distribution and the defect-free reference distribution, when the preset normal threshold is smaller than or equal to D KL and smaller than a preset slight abnormal threshold, the current characteristic distribution is judged to be slightly deviated from the reference distribution, when the preset slight abnormal threshold is smaller than or equal to D KL and smaller than a preset medium abnormal threshold, the current characteristic distribution is judged to be slightly deviated from the reference distribution, when D KL is larger than or equal to a preset medium abnormal threshold, the current characteristic distribution is judged to be severely abnormal, and the current characteristic distribution is extremely greatly deviated from the reference distribution; (2) The judgment criterion of the Wasserstein distance is that the W < preset normal distance threshold value is judged to be normal, which indicates that the optimal matching cost of the current feature and the reference feature is extremely low, the preset normal distance threshold value is less than or equal to W < preset slight abnormal distance threshold value is judged to be slightly abnormal, which indicates that small matching deviation exists between the current feature and the reference feature, the preset slight abnormal distance threshold value is less than or equal to W < preset moderate abnormal distance threshold value is judged to be moderately abnormal, which indicates that the matching deviation between the current feature and the reference feature is obvious, and the W is more than or equal to the preset moderate abnormal distance threshold value is judged to be severely abnormal, which indicates that the matching deviation between the current feature and the reference feature is extremely large; (3) The criterion of the Mahalanobis distance is that D M is less than or equal to the preset confidence level, the judgment is normal when the chi-square distribution score of the corresponding feature dimension p is normal, and D M is greater than the preset confidence level, the judgment is abnormal when the chi-square distribution score of the corresponding feature dimension p is abnormal; S5, judging and quantitatively evaluating the defect type, namely judging the defect type based on a statistical framework aiming at the abnormal feature vector, completing quantitative evaluation of the defect size by combining an acoustic empirical formula, and distributing weight to optimize an evaluation result according to the spatial stress distribution of the detection object.
  2. 2. The method for detecting environmental noise adaptive multi-spectral stress wave-voiceprint damage according to claim 1, wherein in S1, the enhanced excitation signal is generated Expressed as: Wherein, the In the event of an ambient noise signal, In order to be an active excitation signal, As a self-adaptive weight function, when the environmental noise intensity is more than or equal to 60dB When the environmental noise intensity is less than 60dB 。
  3. 3. The method for detecting environmental noise adaptive multi-frequency stress wave-voiceprint damage according to claim 2, wherein in S2, the multi-channel synchronous acquisition device adopts an orthogonal arrangement strategy, a sensor is arranged normally on the surface of a vertical detection object to capture longitudinal waves, a sensor is arranged tangentially on the horizontal surface to capture transverse waves, a stress concentration point of the detection object adopts a triaxial sensor group, and the sensor spacing is equal to the distance between the sensors Based on spatial sampling theorem C is the sound velocity of the detection target material, The highest detection frequency is 150kHz, and the sensor interval of a steel material detection object is set to be 15mm; The multi-band acoustic emission stress wave signals comprise low-frequency 20-40kHz, intermediate-frequency 40-80kHz and high-frequency 80-150kHz frequency band signals; the filtering of the fixed interference frequency band signal is specifically to filter 50Hz and integer multiple harmonic thereof, and collect the switching frequency of the host power supply and integer multiple harmonic thereof.
  4. 4. The environmental noise adaptive multi-frequency spectrum stress wave-voiceprint damage detection method according to claim 1, wherein S5 specifically comprises: s5.1, aiming at the feature vector judged to be abnormal in S4.5, judging the defect type based on a multidimensional Bayesian framework: The prior probability of the multidimensional Bayesian framework is set based on historical detection data, P (crack) =0.45, P (corrosion) =0.30, P (stratification) =0.15, P (void) =0.10; The likelihood function uses: estimating parameters by maximum likelihood method , Estimating the mean value; , Estimating for variance; posterior probability The formula for determining the defect type is: ; s5.2, finishing quantitative evaluation of the size of the defect type determined in the S5.1 by combining an acoustic empirical formula: the acoustic empirical formula includes: crack length formula: , for the detection of wavelengths in the corresponding frequency band, Is the rate of change of impedance; crack depth formula: , Is the defect echo time difference; the crack width formula is: AR is an empirical length-width ratio, the value is 0.1-0.3, and the AR is adjusted according to the hardness of the material; S5.3, distributing weights according to the spatial stress distribution of the detection object, and optimizing the quantitative evaluation result of S5.2: The stress weight distribution adopts a weight function: Wherein, the To detect the spatial stress distribution function of the object, At the value of the maximum stress to be applied, The weight coefficient is 0.5-2.0.
  5. 5. The environmental noise adaptive multi-spectrum stress wave-voiceprint damage detection method according to claim 1, wherein in S5.2, the defect size assessment error range is length error + -10%, depth error + -15% and width error + -20%.
  6. 6. The method for detecting environmental noise adaptive multi-spectral stress wave-voiceprint damage according to claim 4, wherein in S5.3, typical area weight values of stress weight distribution are weld zone weight w=2.5, bending zone weight w=2.0, supporting point weight w=1.8, general area weight w=1.0, and the weight coefficients are as follows The value of (2) is adjusted according to the strength of the material to be detected, and the high-strength alloy material The value is 1.5-2.0, and the common metal material The value is 0.5-1.0, and the composite material The value is 1.0-1.5.
  7. 7. An ambient noise-adapted multi-spectral stress wave-voiceprint damage detection system for performing the method of any one of claims 1-6, wherein: the system comprises a multi-frequency-spectrum acoustic emission stress wave acquisition module, a voiceprint map construction module, an intelligent analysis diagnosis module and a real-time monitoring and early warning module, wherein the multi-frequency-spectrum acoustic emission stress wave acquisition module, the voiceprint map construction module, the intelligent analysis diagnosis module and the real-time monitoring and early warning module are sequentially connected through signals; The multi-frequency spectrum acoustic emission stress wave acquisition module is used for acquiring an environmental noise passive excitation signal and an environmental noise active excitation signal, adjusting the duty ratio of the environmental noise passive excitation signal and the environmental noise active excitation signal through a self-adaptive weight function to form a mixed excitation signal, synchronously acquiring a multi-frequency band acoustic emission stress wave signal under the action of mixed excitation and filtering an interference signal; The voiceprint map construction module is used for receiving signals output by the multi-spectrum acoustic emission stress wave acquisition module, acquiring factory reference voiceprints of the detected objects, completing voiceprint calibration by combining with actual working conditions, constructing a three-dimensional voiceprint map based on the calibrated signals, and calculating acoustic impedance tensor fields; the intelligent analysis and diagnosis module is used for extracting characteristic parameters of the three-dimensional voiceprint map, identifying defect abnormality through a vector distribution difference algorithm, judging defect types based on a statistical framework, and quantitatively evaluating the defect sizes by combining an acoustic empirical formula; The real-time monitoring and early warning module is used for receiving the defect data output by the intelligent analysis and diagnosis module, filtering and confirming the effectiveness of the data, analyzing the defect development trend, predicting the residual life, calculating the total confidence coefficient based on preset weight distribution and triggering grading early warning.
  8. 8. The ambient noise adapted multi-spectral stress wave-voiceprint damage detection system of claim 7 wherein: the multi-frequency spectrum acoustic emission stress wave acquisition module comprises a multi-channel synchronous acquisition unit, a frequency band separation unit and a noise filtering unit, and is specific: The multichannel synchronous acquisition unit adopts a 12-channel hardware architecture, takes an FPGA as a clock synchronous controller, has clock precision of +/-10 ns, is configured with a 24-bit ADC chip, has a dynamic range of more than 120dB, and realizes time synchronous acquisition of multichannel signals, and the sampling rate is 1MHz; The frequency band separation unit synchronously separates acoustic emission stress wave signals of low frequency 20-40kHz, intermediate frequency 40-80kHz and high frequency 80-150kHz through a real-time FFT frequency division technology, and each frequency band is independently stored and transmitted; the noise filtering unit is used for filtering 50Hz and harmonic waves thereof, collecting host power supply switching frequency and harmonic waves thereof by adopting an IIR notch filter, and filtering cut-off attenuation is more than or equal to 40dB; the voiceprint map building module comprises a multi-level frequency spectrum mapping unit, an acoustic impedance tensor calculation unit and a space field reconstruction unit, and is particularly: The multi-level spectrum mapping unit is used for receiving multi-band acoustic emission stress wave signals, collecting factory reference spectrum characteristics in a standard environment, and carrying out calibration in combination with the installed actual working conditions, wherein the actual working conditions comprise temperature, stress and boundary conditions, eliminating the influence of environmental differences on the spectrum characteristics, and establishing a multi-level mapping relation of frequency band-space position-spectrum parameters; The acoustic impedance tensor calculation unit is used for constructing an acoustic impedance tensor model at each point in space based on the calibrated acoustic impedance data output by the multi-level frequency spectrum mapping unit, wherein the model comprises a main impedance component and a coupling impedance component, the main impedance component is a diagonal element, and the coupling impedance component is a non-diagonal element; The space field reconstruction unit takes the acoustic impedance tensor invariant as basic data, carries out space field reconstruction by adopting radial basis function interpolation, ensures interpolation accuracy by setting boundary conditions, and finally constructs a three-dimensional voiceprint map covering the whole space of the detection object to realize the space continuous distribution characterization of acoustic impedance characteristics; The intelligent analysis diagnosis module comprises a feature extraction unit, an abnormality identification unit and a quantitative evaluation unit, and is specific: The characteristic extraction unit is used for extracting time domain characteristics, frequency domain characteristics and acoustic impedance characteristics of the voiceprint map, wherein the time domain characteristics comprise 8 items of root mean square values, peak value factors, kurtosis, skewness, waveform factors, pulse factors and margin factors, the frequency domain characteristics comprise 6 items of spectrum centroid, spectrum width, spectrum entropy, spectrum peak frequency, spectrum peak amplitude, spectrum roll-off points, spectrum flatness, spectrum irregularity, main frequency ratio and harmonic distortion, and the acoustic impedance characteristics comprise 6 items of impedance mean value, impedance gradient, impedance anisotropy, impedance nonlinearity, impedance phase delay and impedance quality factor to form a high-dimensional characteristic vector; The characteristic extraction unit further comprises a multi-resolution analysis module, and performs time-frequency decomposition on the acoustic emission stress wave signals by adopting wavelet packet decomposition to obtain energy characteristic vectors containing energy of each sub-band, and the energy characteristic vectors, the time domain characteristics, the frequency domain characteristics and the acoustic impedance characteristics form high-dimensional characteristic vectors together; the anomaly identification unit is used for internally arranging a KL divergence, a Wasserstein distance and a Markov distance calculation model, automatically selecting an adaptive vector distribution difference algorithm to identify defect anomaly, and outputting anomaly grades, wherein the anomaly grades comprise normal, slight, moderate and serious; The quantitative evaluation unit integrates a multidimensional Bayes statistical framework and an acoustic empirical formula library, automatically calculates the probability of the defect type and the size parameter after inputting the feature vector, and optimizes the evaluation result by combining stress weight distribution; The real-time monitoring and early warning module comprises a data stream processing unit, a trend analysis unit and an alarm decision unit, and specifically comprises the following components: The data stream processing unit integrates a Kalman filtering algorithm and a delay confirmation mechanism, performs filtering denoising and validity confirmation on the diagnosis data, and outputs stable defect data; The trend analysis unit is internally provided with an index degradation model, a power law degradation model and a linear-index mixed model, The exponential degradation model expression is: : the power law degradation model expression is: ; the linear-exponential mixture model expression is: ; an optimal model is selected based on AIC criteria, the AIC criteria expression being: Wherein k is the number of model parameters, and L is a likelihood value; predicting defect development trend and residual life , In order to be a failure threshold value, For the current time, output 95% confidence interval , The method comprises the steps of obtaining through Monte Carlo simulation; The alarm decision unit is used for presetting weight distribution, wherein the weight distribution is that the defect probability is 0.3, the defect size is 0.3, the growing trend is 0.25, the position importance is 0.15, and the total confidence is calculated Triggering the grading early warning, For the confidence level of the i-th criterion, In order to correspond to the weight of the object, And when Conf is more than or equal to 0.8, triggering a primary alarm, wherein the primary alarm is emergency treatment, when Conf is more than or equal to 0.5 and less than or equal to 0.8, triggering a secondary alarm, wherein the secondary alarm is scheduled maintenance, and when Conf is less than or equal to 0.5, triggering a tertiary alarm, wherein the tertiary alarm is continuous monitoring.

Description

Environmental noise adaptive multi-frequency spectrum stress wave-voiceprint damage detection method and system Technical Field The invention relates to the technical field of nondestructive testing, in particular to an environment noise adaptive multi-frequency spectrum stress wave-voiceprint damage detection method and system. Background In the fields of industrial nondestructive testing, aerospace, petrochemical industry, rail transit, nuclear power equipment, intelligent manufacturing and the like, equipment structure health monitoring and defect detection are key links for guaranteeing production safety and prolonging equipment service life, the current mainstream detection technology takes traditional ultrasonic detection as a core, but the technology and similar derivative technology have obvious limitation in practical application, and are difficult to adapt to complex requirements of industrial scenes. From the technical application characteristics, the prior detection technology generally regards noise in an industrial environment as an interference signal, the traditional ultrasonic detection needs to construct a quiet detection environment (such as an isolation workshop and a sound insulation device) to work, so that a detection scene is strictly limited, meanwhile, most of the technologies adopt a single-band detection mode, are limited by the physical characteristics of sound waves, cannot consider penetration depth and detection resolution by single-band signals, namely, a low-frequency signal can penetrate thicker materials but is difficult to identify microscopic defects, and a high-frequency signal has high resolution but weak penetration capability, so that information of multi-scale defects (such as macroscopic structure cracks and surface micro-damages) is directly lost. In addition, the prior art relies on a shutdown detection mode, a production flow is required to be interrupted to create stable detection conditions, real-time monitoring of equipment in an operating state cannot be realized, in a defect judgment link, the traditional method mainly relies on manual experience to set a fixed threshold value, the adaptability to detection signals of different materials and under different working conditions is poor, false alarm or missing alarm is easily caused by threshold value deviation, and the detection reliability is further reduced. From the practical application defects, the limitations of the prior art are multiple restrictions, namely, the severe environmental requirements lead to significant increase of detection cost, additional investment is required for acquisition, installation and maintenance of sound insulation equipment, special industrial scenes such as the field, high temperature and high noise cannot be adapted, secondly, insufficient information utilization leads to insufficient detection coverage, defects of a certain scale can only be captured by a single-band signal, the whole health state of the equipment is difficult to comprehensively evaluate, for example, in the detection of metal parts, macro cracks of a welding line area and micro corrosion of the surface can be omitted, thirdly, the shutdown detection mode seriously affects the production efficiency, particularly in the detection of large-scale equipment such as an aeroengine and a nuclear power reactor, the economic loss of tens of thousands of yuan to millions of yuan can be caused by single shutdown, fourthly, the intelligent degree is low, the experience dependence on operators is strong, the judgment standard difference of different personnel is further amplified, fifthly, the quantification capability is insufficient, the prior art can only qualitatively judge whether the defects exist or not, and key parameters such as the size, the depth and the expansion rate of the defects are difficult to accurately evaluate, and the key parameters such as the equipment cannot provide accurate data support for the maintenance of the equipment. In summary, the existing nondestructive testing technology has the defects of environmental adaptability, testing efficiency, intelligentization level and quantification capability, and cannot meet the requirement of on-line, real-time, accurate and intelligent detection of equipment in the industrial 4.0 background, so that a novel testing scheme capable of breaking through the limitation of the traditional technology is needed. Disclosure of Invention The invention aims to provide an environmental noise adaptive multi-frequency spectrum stress wave-voiceprint damage detection method and system, which are used for solving the problems of environmental interference limitation, single-band information loss, insufficient quantitative capability and stop dependence of traditional detection, realizing online damage identification, quantitative evaluation and life prediction of metal and composite material components, and meeting the real-time, accurate and intelligent require