Method for identifying internal crack of cylindrical shell component

Abstract

The application discloses a kind of cylindrical shell component internal crack identification method, it is related to cylindrical shell component technical field.The method steps include: collecting acoustic vibration signal data and carrying out pulse detection, obtain pulse family characteristic vector;Pulse family characteristic vector is subjected to consistency test and band matching degree analysis to obtain pulse variation coefficient and band matching index, the crack confidence level is obtained in combination of the two, greater than preset threshold value then determine that there is crack;After confirming crack, wave propagation characteristic vector is obtained based on pulse family characteristic vector, wave propagation correlation coefficient is obtained by dynamic event regularization method, the first order difference variance of wave propagation characteristic vector is calculated to obtain wave propagation difference variance, combined to obtain comprehensive difference degree, crack candidate abnormal points are determined accordingly, high-confidence crack source area is obtained by neighborhood propagation analysis;Cylindrical shell micro area is divided, and zoning weight is obtained by analytic hierarchy process, and the number of high-confidence crack source area in zoning is combined to determine zoning crack risk level.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cylindrical shell components, in particular to a cylindrical shell component internal crack identification method. BACKGROUND

[0002] As a key bearing structure widely used in industrial equipment, the cylindrical shell component is prone to fatigue cracks under long-term service in complex alternating loads and harsh environments. The initiation and propagation of cracks can significantly weaken the structural integrity, and if not identified and located in time, it may cause catastrophic accidents. Therefore, developing an in-service monitoring technology that can accurately identify internal cracks in components and assess their risk status is of great significance to ensure the safe operation of major equipment and achieve predictive maintenance.

[0003] One of the traditional crack identification methods is the time difference positioning technology based on acoustic emission or vibration signals. This method arranges a sensor array on the surface of the structure, captures the stress wave signals generated by crack propagation or external excitation, and calculates the relative time difference of stress wave arrival at different sensors. A hyperbolic equation system is constructed using geometric relationships, and the spatial coordinates of the crack source are finally solved to achieve static positioning of the crack.

[0004] However, the traditional method focuses on solving a single spatial coordinate of the crack event, lacks analysis of the propagation law of crack dynamic response signals, making it difficult to distinguish between transient signal interference and real crack response signals, thus unstable in structural safety assessment of cylindrical shell components, ultimately increasing the risk of sudden rupture accidents of cylindrical shell components. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides a cylindrical shell component internal crack identification method to solve the problems in the background art.

[0006] To achieve the above purpose, the present application is implemented by the following technical scheme: a cylindrical shell component internal crack identification method, comprising the following steps:

[0007] Step S1: arranging a sensor array on the surface of the cylindrical shell component, actively exciting the cylindrical shell component and collecting acoustic vibration signals through the sensor array to obtain acoustic vibration signal data;

[0008] Step S2: performing pulse detection on the acoustic vibration signal data to calculate acoustic vibration pulse family feature vectors;

[0009] Step S3: performing consistency test on the acoustic vibration pulse family feature vectors to obtain pulse variation coefficients, performing frequency band matching degree analysis on the acoustic vibration pulse feature vectors to obtain frequency band matching indexes, and combining the pulse variation coefficients and the frequency band matching indexes to calculate crack confidence, when the crack confidence is greater than a preset threshold, then the cylindrical shell component has a crack.

[0010] Step S4: After determining the existence of the crack, a wave propagation characteristic vector is calculated based on the acoustic vibration pulse family characteristic vector; a correlation analysis is performed on the wave propagation characteristic vector and a preset theoretical propagation characteristic vector by a dynamic event normalization method to obtain a wave propagation correlation coefficient; a variance of a first-order difference of the wave propagation characteristic vector is calculated to obtain a wave propagation difference variance;

[0011] Step S5: A comprehensive difference degree is calculated based on the wave propagation correlation coefficient and the wave propagation difference variance; a crack candidate abnormal point is obtained based on the comprehensive difference degree, and a neighborhood propagation analysis is performed on the crack candidate abnormal point to obtain a high-confidence crack source area;

[0012] Step S6: The cylindrical shell component is divided into micro-zones to obtain cylindrical shell sub-zones, and a risk quantification is performed on the cylindrical shell sub-zones by an analytic hierarchy process to obtain a cylindrical shell sub-zone weight; a sub-zone crack risk level is obtained by combining the cylindrical sub-zone weight and the number of high-confidence crack source areas in the sub-zone, thereby realizing crack identification of the cylindrical shell component.

[0013] Preferably, the sensor array arranged on the surface of the cylindrical shell component comprises the following specific steps:

[0014] According to the circumferential sensing belt of the cylindrical shell component as the weft, two types of sensors are arranged in adjacent pairs along the circumferential direction on each circumferential sensing belt: at a preset installation coordinate point, an acoustic emission sensor and a vibration sensor are arranged in an adjacent pair as a sensor unit pair; the number of sensor unit pairs is determined by the diameter of the cylindrical shell component and the monitoring accuracy, and the number of sensor unit pairs required to be arranged on each circumferential sensing belt is ≥ , wherein the preset maximum arc length interval is adjusted by the diameter of the cylindrical shell component, and the larger the diameter, the smaller the preset maximum arc length interval, so that the sensor array is uniformly distributed, thereby ensuring that the arc length distance between adjacent sensors is controlled.

[0015] Preferably, the method of exciting the cylindrical shell component by an active source and collecting acoustic vibration signals by the sensor array to obtain acoustic vibration signal data comprises the following steps:

[0016] Under the condition that the cylindrical shell component has no crack, a low-amplitude, wide-frequency steady excitation source is applied to the cylindrical shell component, and acoustic emission response signals (t) and vibration response signals (t) of all acoustic emission channels and all vibration channels in a continuous time length (t) are synchronously collected, wherein cg=1, 2,..., M;

[0017] For the vibration sensor in the sensor unit pair, the arithmetic mean value of the response signal of each channel in the time domain is calculated as the zero drift baseline value of the channel:

[0018]

[0019] in, is the zero-drift baseline value of the cg-th sensing channel, representing the local output of that channel in a crack-free state. The duration of steady-state excitation, The vibration response signal acquired by the cg-th sensing channel;

[0020] A controllable transient excitation is applied at a fixed position on the cylindrical shell component, and the pulse response signals of all channels are acquired synchronously. The channel with the largest amplitude value in this type of sensor is selected as the reference channel cg2 of this type of sensor. The time delay deviation, amplitude compression coefficient, and phase drift deviation of other channels cg in this type of sensor relative to the reference channel cg2 are calculated:

[0021]

[0022] in, This indicates the time delay deviation between the sensing channel cg and the reference channel cg2. This is the vibration pulse response signal of the sensing channel cg. This indicates the time-axis translation of the reference channel cg2. The vibration pulse response signal, This is the time offset. Indicates the duration of the transient vibration pulse response signal;

[0023] Calculate the amplitude compression factor:

[0024]

[0025] in, This represents the amplitude compression factor of the sensing channel cg relative to the reference channel cg2. This represents the vibration pulse response signal of the reference channel cg2. This represents the vibration pulse response signal of the sensor channel cg;

[0026]

[0027] in, This represents the phase shift deviation of the sensing channel cg relative to the reference channel cg2 at frequency f. It is a Fourier transform operator, where f is the frequency variable, representing the frequency components of the signal. The pulse response signal of the sensing channel cg Fourier transform, The pulse response signal of the reference channel cg2 Fourier transform, () indicates the phase angle when taking a complex number;

[0028] For each sensor category, the calibration parameters for each sensor channel will be... , , , Each parameter is stored in a separate calibration parameter library. During subsequent real-time monitoring, any channel will respond to the acquired vibration response signal. (t) Compensation is performed using the corresponding calibration parameters to obtain standard vibration signal data. :

[0029]

[0030] in, This represents the phase shift deviation of the sensing channel cg relative to the reference channel cg2 at frequency f. This represents the zero-drift baseline value of the cg-th sensing channel. This represents the amplitude compression factor of the sensing channel cg relative to the reference channel cg2. For inverse Fourier transform operators, For Fourier transform operators, It is the imaginary unit. It is a time shift The vibration response signal afterward This indicates the time delay deviation of the sensing channel cg relative to the reference channel cg2;

[0031] Similarly, acoustic emission response signal (t) correction method and vibration response signal The procedure is the same as for (t), requiring the same calibration steps for the acoustic emission channel sensor to ultimately obtain standard acoustic emission signal data. ;

[0032] The final acoustic vibration signal data includes and .

[0033] Preferably, the step of pulse detection of acoustic vibration signal data and calculation of acoustic vibration pulse family feature vector includes the following steps:

[0034] Standard vibration signal data for acoustic and vibration signal data For example, calculating standard vibration signal data The short-time moving average is used to subtract the standard vibration signal data from the short-time moving average to obtain the dynamic signal with zero baseline. :

[0035]

[0036] wherein, is the zero baseline vibration dynamic signal of the cgth channel, is the standard vibration signal data of the cgth channel, is the length of the short-time time window;

[0037] Calculating dynamic amplitude threshold for pulse detection :

[0038]

[0039] wherein, is the dynamic threshold, is the noise adjustment coefficient, the value range is [4, 6], the higher the value, the more conservative the detection, is the real-time noise standard deviation;

[0040] When ∣ ∣> , mark as the vibration event start point . When the signal amplitude is continuously lower than the threshold for a duration 2 signal sampling periods, mark as the vibration event end point , extract the signal segment within the , time window as a vibration candidate pulse family to be analyzed;

[0041] For each candidate vibration pulse family, in the vibration pulse family signal segment, the sliding window maximum value method is used for peak value detection to obtain the peak time sequence , wherein N is the total number of pulses in the vibration pulse family;

[0042] Calculate the time difference between adjacent pulse peaks to obtain the pulse interval sequence Δ [n]:

[0043]

[0044] wherein, expresses the nth pulse interval, expresses the peak time of the nth pulse, expresses the peak time of the n+1th pulse;

[0045] Calculate the average pulse interval and interval standard deviation of the vibration pulse family to describe its time distribution characteristics:

[0046]

[0047] wherein, is the average pulse interval, N is the total number of pulses, and n is the pulse index;

[0048]

[0049] wherein, is the standard deviation of the vibration pulse interval;

[0050] The envelope e(t) of the vibration pulse family signal is calculated using the Hilbert transform:

[0051]

[0052] wherein, is the analytical envelope of the vibration pulse family signal, is the vibration dynamic signal of the zero baseline of the cgth channel, is the Hilbert transform;

[0053] The envelope fluctuation intensity, i.e. the ratio of the standard deviation and the mean value of the envelope, is calculated. The greater the value, the more intense and unstable the energy release:

[0054]

[0055] wherein, is the vibration envelope fluctuation intensity, T is the duration of the vibration pulse family, T= - , is the vibration pulse start time point, is the vibration pulse end time point, is the mean value of the analytical envelope; the power spectrum P(f) of the vibration pulse signal band of the vibration pulse family is obtained by performing a fast Fourier transform, first identify all potential peaks in the power spectrum that satisfy the peak amplitude of the global highest peak value, for each potential peak , determine its 3dB low frequency cutoff frequency and high frequency cutoff frequency , form the frequency band ; , calculate the energy of each frequency band ; finally, select the frequency band with the largest energy as the dominant frequency band of the vibration pulse family ;

[0056] For each vibration candidate pulse family detected by each channel, the vibration pulse family feature vector [ , , , , , This feature vector serves as the digital representation of the pulse family. Similarly, pulse family detection is performed on the acoustic emission channel to ultimately obtain the acoustic emission pulse family feature vector. =[ , , , , , ],in, The starting point of the acoustic emission event. The end point of the acoustic emission event. The standard deviation of the acoustic emission pulse interval. The intensity of acoustic emission envelope fluctuations. This is the dominant frequency band for acoustic emission. The amplitude of the acoustic emission signal is used to obtain the acoustic vibration pulse family feature vector, which includes the vibration pulse family feature vector and the generation pulse family feature vector.

[0057] Preferably, the step of performing a consistency check on the eigenvectors of the acoustic-vibration pulse family to obtain the pulse variation coefficient includes the following steps:

[0058] For the vibration pulse family eigenvectors in the acoustic-vibration pulse family eigenvectors, the pulse family eigenvectors In Set a time alignment window as the time base. All pulse family eigenvectors are sorted by their start time. Sorting, based on a pulse family trigger time For the seed, search all that meet the criteria. By analyzing the family of vibration pulses under certain conditions, a cluster of candidate events can be obtained.

[0059] A consistency test is performed on the eigenvectors of the K vibration pulse families in the candidate event cluster, and the coefficient of variation of the pulse interval standard deviation in the candidate event cluster is calculated:

[0060]

[0061] in, The coefficient of variation is the standard deviation of the vibration pulse interval. This represents the pulse interval standard deviation of the eigenvector of the Kth vibration pulse family, where std() is the standard deviation function and mean() is the mean function.

[0062] Calculate the intensity of vibration envelope fluctuations Coefficient of variation:

[0063]

[0064] in, The coefficient of variation of the intensity of the vibration envelope fluctuations. the fluctuation strength of the vibration envelope of the Kth vibration pulse cluster feature vector;

[0065] the coefficient of variation of the vibration pulses is:

[0066]

[0067] wherein, the coefficient of variation of the vibration pulses, the weight of the standard deviation of the pulse interval, the weight of the fluctuation strength of the envelope, = 1, the threshold value of the standard deviation of the pulse interval, the threshold value of the fluctuation strength of the envelope;

[0068] the coefficient of variation of the vibration pulses is calculated in the same way as the coefficient of variation of the acoustic emission pulses .

[0069] Preferably, the frequency band matching degree analysis of the acoustic vibration pulse feature vector is performed to obtain a frequency band matching index, including the following steps:

[0070] based on the dominant frequency band of the acoustic emission channel signal in the pulse cluster feature vector and the dominant frequency band of the vibration channel signal the weighted average value of the dominant frequency band of the acoustic emission and vibration signals in the candidate cluster is calculated respectively:

[0071]

[0072] wherein, the weighted average value of the acoustic emission dominant frequency band, the acoustic emission dominant frequency band of the kth acoustic emission pulse cluster feature vector, the acoustic emission amplitude of the kth acoustic emission pulse cluster feature vector;

[0073]

[0074] wherein, the weighted average value of the vibration dominant frequency band, the vibration dominant frequency band of the kth vibration pulse cluster feature vector, the vibration signal amplitude of the kth vibration pulse cluster feature vector;

[0075] the frequency band matching index is obtained by performing frequency band matching degree analysis on the pulse cluster feature vector:

[0076]

[0077] wherein,​ is a frequency band matching index, is a frequency alignment tolerance, D is a Euclidean distance of energy center of gravity space of acoustic emission and vibration signal, is an energy center of gravity distance threshold.

[0078] Preferably, the crack confidence is:

[0079] The crack confidence is calculated by combining the vibration pulse variation coefficient, the acoustic emission pulse coefficient and the frequency band matching index:

[0080]

[0081] wherein, is a crack confidence, is a vibration pulse variation coefficient, is a frequency band matching index, is an acoustic emission pulse variation coefficient.

[0082] Preferably, the correlation analysis of the wave propagation characteristic vector and the preset theoretical propagation characteristic vector by the dynamic event regularization method to obtain a wave propagation correlation coefficient comprises the following specific steps:

[0083] For each candidate point G(i,j) of the search grid and each preset direction of the candidate point , the candidate point G(i,j) is regarded as a time starting point of wave propagation, and a virtual path is generated along the direction from the point, and the virtual path and the sensor number set along each direction, the signal arrival time recorded by each sensor corresponding to the vibration pulse family characteristic vector , the signal peak amplitude recorded by each sensor , for any direction to be analyzed , a narrowband analysis area with an axial angle of is defined on the cylindrical shell development diagram, all sensor points located in the narrowband area are screened out to form an analysis point set of the direction = {( , )}, a time sequence and an amplitude sequence are generated;

[0084] For the path with G(i,j) as the starting point and the direction angle , the theoretical sequence is generated according to the sound velocity model and the attenuation model of the intact shell: and , wherein, is the distance from the nth point on the path to the starting point;

[0085] The wave propagation correlation coefficient is obtained by calculating the wave propagation eigenvector and the preset theoretical propagation eigenvector of each direction through a dynamic event normalization method and performing correlation analysis:

[0086] ;

[0087] ;

[0088] The wave propagation correlation coefficient is obtained by calculating the wave propagation eigenvector and the preset theoretical propagation eigenvector of each direction through a dynamic event normalization method and performing correlation analysis: The maximum and minimum values are normalized to obtain The maximum and minimum values are normalized to obtain ; ;

[0089] The wave propagation correlation coefficient is obtained by calculating the wave propagation eigenvector and the preset theoretical propagation eigenvector of each direction through a dynamic event normalization method and performing correlation analysis: :

[0090]

[0091] wherein, is the wave propagation correlation coefficient, is the time sequence weight, is the normalized time sequence correlation index, is the amplitude weight, is the normalized amplitude correlation index, FX is the total number of directions, fx is the direction index, and DTW() is a time sequence dynamic normalization method function.

[0092] Preferably, the first-order difference of the wave propagation eigenvector is calculated to obtain the wave propagation difference variance, including the following specific steps:

[0093] The first-order difference of the wave propagation eigenvector is calculated to obtain the wave propagation difference variance:

[0094] ;

[0095] The wave propagation difference variance is obtained by calculating the first-order difference of the wave propagation eigenvector:

[0096]

[0097] wherein, is the wave propagation difference variance, is the variance of the time sequence first-order difference, is the variance of the amplitude first-order difference, and K is the total number of pulse family eigenvectors in the candidate event cluster.

[0098] Preferably, the comprehensive difference degree is:

[0099] The comprehensive difference degree is calculated by combining the wave propagation correlation coefficient and the wave propagation difference variance:

[0100]

[0101] wherein, is the comprehensive difference degree, is the wave propagation correlation coefficient, is the wave propagation difference variance.

[0102] The application provides a cylindrical shell component internal crack identification method, relates to machine learning and deep learning technology, and has the following beneficial effects:

[0103] (1) The crack confidence is obtained by combining the pulse variation coefficient and the frequency band matching index, and a dual determination mechanism of feature consistency and energy matching degree is constructed. The pulse variation coefficient tests the consistency of the characteristics of different channel pulse families, and excludes single channel misjudgment caused by individual differences of sensors or local interference. The frequency band matching index ensures that the identified signal is related to the stress wave and structural vibration associated events generated by the crack, rather than irrelevant environmental noise or equipment interference, by correlating the dominant frequency band and energy center distance of the acoustic emission and vibration signal. The combination of the two effectively improves the accuracy of crack existence determination and avoids missed judgment, providing reliable pre-determination basis for subsequent crack positioning and analysis.

[0104] (2) The comprehensive difference degree is calculated by combining the wave propagation correlation coefficient and the wave propagation difference variance. The wave propagation correlation coefficient compares the measured crack evolution characteristics with the theoretical evolution characteristics by dynamic event regularization method to determine whether the signal conforms to the physical law of crack propagation. The wave propagation difference variance quantifies the smoothness of the evolution characteristic vector to identify signal abnormal jumps caused by the existence of cracks. The comprehensive difference degree combines the two after normalization, which can not only exclude interference points that do not conform to the theoretical evolution law, but also filter out real crack associated points with abnormal signal fluctuations, and then form a high credible crack source area through neighborhood propagation analysis, greatly improving the accuracy and credibility of crack positioning.

[0105] (3) The partition crack risk level is calculated by combining the cylindrical shell partition weight and the number of high credible crack source areas in the partition. The partition weight quantifies the inherent risk basis of different regions by combining stress level, material performance, geometric characteristics and other factors through the analytic hierarchy process. The number of high credible crack source areas reflects the distribution density of actual cracks in the region. The risk level calculated by combining the two is no longer a general assessment of the whole component, but can accurately identify the key regions with high inherent risk and multiple crack distribution, providing clear direction for subsequent maintenance decision-making and helping to focus on high-risk partitions for repair, improving the efficiency and pertinence of cylindrical shell component safety management. BRIEF DESCRIPTION OF DRAWINGS

[0106] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0107] Fig. 1 This is a flowchart of the steps for identifying internal cracks in a cylindrical shell component proposed in this invention.

[0108] Fig. 2 This is a step hierarchy diagram of obtaining crack confidence in a method for identifying internal cracks in a cylindrical shell component proposed in this invention;

[0109] Fig. 3 This is a step hierarchy diagram for obtaining the risk level of zoned cracks in a method for identifying internal cracks in a cylindrical shell component proposed in this invention. Detailed Implementation

[0110] 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 skilled in the art without creative effort are within the scope of protection of the present invention.

[0111] Please see Figs. 1-3 The present invention provides a technical solution: a method for identifying internal cracks in cylindrical shell components.

[0112] Step S1: Arrange a sensor array on the surface of the cylindrical shell component, use an active excitation source to excite the cylindrical shell component, and collect acoustic vibration signals through the sensor array to obtain acoustic vibration signal data.

[0113] Based on the geometric and mechanical properties of the cylindrical shell component, the sensing zones and sensor positioning are planned, following the latitude and longitude grid principle to ensure full coverage monitoring. The circumferential sensing zones, acting as latitude lines, are initially set at approximately 100-200 mm from the boundaries at both ends of the cylindrical shell and at the midpoint of the shell's total length. For extra-long components, additional rings are added at equal intervals according to accuracy requirements to control the axial monitoring spacing. The axial sensing lines, acting as longitude lines, are evenly distributed along the generatrix of the cylindrical shell, intersecting with the circumferential sensing zones to form grid points. On each circumferential sensing zone, two types of sensors are paired and installed adjacently along the circumference: at preset installation coordinate points, acoustic emission sensors and vibration sensors are installed adjacently with extremely small spacing (usually a few centimeters), forming a sensor unit pair.

[0114] It should be noted that the acoustic emission sensor can be a harmonic resonant piezoelectric acoustic emission sensor, and the vibration sensor can be an industrial piezoelectric accelerometer. The harmonic resonant piezoelectric acoustic emission sensor with a resonant frequency of 150 kHz-400 kHz is used to capture high-frequency stress waves of cracks, and the industrial piezoelectric accelerometer with a low-frequency (1 Hz-5 kHz) response is used to monitor structural vibration.

[0115] The number of sensor (acoustic emission sensor and vibration sensor) unit pairs is determined by the diameter of the cylindrical shell member and the monitoring accuracy. The number of sensor unit pairs required to be arranged on each circumferential sensing band is ≥ The preset maximum arc length interval (such as 500 mm) is adjusted by the diameter of the cylindrical shell member. The larger the diameter, the smaller the preset maximum arc length interval should be, so that the sensor array is uniformly distributed, thereby ensuring that the arc length distance between adjacent sensors is controlled. For stress concentration areas such as welds, openings, and supports, it is necessary to implement encryption distribution on the adjacent axial and circumferential grid lines. Generally, the interval is reduced to 1 / 2 to 1 / 3 of the basic value, forming a local high-density monitoring area to capture potential defects.

[0116] During the sensor installation process, standard processes must be strictly followed to ensure signal fidelity. Before installation, the selected points must be surface treated, and an angle grinder or sandpaper must be used to completely remove paint, rust, and oxidation layers until the metal gloss is exposed and the surface flatness is better than Ra 3.2 μm. The treated installation surface must be evenly coated with a special acoustic coupling agent (such as silicone or epoxy coupling agent) to fill the microscopic gaps in the contact surface and exclude air, ensuring efficient conduction of acoustic and vibration energy. Sensor fixation can be selected according to surface characteristics, such as magnetic base adsorption, special adhesive bonding, or mechanical clamp locking. During installation, the sensor must be perpendicular to the shell surface, and uniform and appropriate clamping force must be applied to avoid damage from excessive tightening or failure from excessive loosening. Clear identification must be made and matched with the point distribution map number one by one, and finally a space positioning accurate, installation quality controlled acoustic-vibration synchronous sensing network is formed.

[0117] To eliminate the effects of installation stress, temperature difference and surface roughness on the consistency of the sensor network sensitivity, calibration is needed at the initial installation of the sensor (ensuring that the cylindrical shell member is in a crack-free state) and re-calibration is needed periodically (such as every quarter or when the ambient temperature changes more than 10°C) to compensate for long-term drift. The object of the calibration parameter measurement is the sensor itself, aiming to eliminate the individual differences of the sensor and the installation effects, so even in the case of a cracked member, regular calibration can still effectively compensate for the sensor drift, but the initial calibration needs to be carried out in a crack-free state to establish a reference. During the service of the object, in order to compensate for the drift of the sensor performance over time and the environment, regular or online calibration triggered under certain conditions (such as after a sharp temperature change) is needed. Online calibration acknowledges that the internal state of the member is unknown, and its main purpose is to correct the slow drift of parameters such as sensitivity and baseline caused by sensor aging, coupling agent drying, temperature changes, etc. During online calibration, only the baseline calibration step under steady-state excitation can be repeated, or a special calibration source (such as a broken lead analog source) that does not damage the structure is used to correct the transient characteristics. The result of the online calibration is used to update the compensation parameters, but its reference is still the relative relationship of the network established by the initial calibration.

[0118] The initial calibration is carried out under a low-amplitude, wide-band steady-state excitation source on the cylindrical shell member in a crack-free state, and the acoustic emission response signals (t) and the vibration response signals (t) of all acoustic emission channels and all vibration channels are synchronously collected within a continuous time (t) and the vibration response signals (t) of all acoustic emission channels and all vibration channels are synchronously collected within a continuous time

[0119] (t) and the vibration response signals (t) of all acoustic emission channels and all vibration channels are synchronously collected within a continuous time

[0120] dt

[0121] wherein, is the zero-drift baseline value of the cgth sensing channel, representing the local output of the channel in a crack-free state, is the duration of the steady-state excitation, is the vibration response signal collected by the cgth sensing channel.

[0122] At the same time as calculating the zero-drift baseline value of each channel, the noise standard deviation of the channel is calculated using the same vibration response signal :

[0123]

[0124] in, Let cg be the noise standard deviation of the cg-th sensing channel, and t be the time index.

[0125] Initial calibration was performed with the cylindrical shell component free of cracks. A controllable transient excitation (such as a standard tap) was applied to a fixed position on the cylindrical shell component, and the pulse response signals of all channels were acquired simultaneously. The channel with the largest amplitude in this type of sensor was selected as the reference channel cg2. The time delay deviation, amplitude compression factor, and phase drift deviation of the other channels cg in this type of sensor relative to the reference channel cg2 were calculated.

[0126]

[0127] in, This indicates the time delay deviation between the sensing channel cg and the reference channel cg2. This is the vibration pulse response signal of the sensing channel cg. This indicates the time-axis translation of the reference channel cg2. The vibration pulse response signal, This is the time offset. This indicates the duration of the transient vibration pulse response signal.

[0128] It should be noted that, It is an application of the cross-correlation function, used to find... and Optimal time alignment of (t) The time offset that causes the integral result to reach its maximum value is calculated by solving for this. To determine the delay deviation, the integral operation is performed on the vibration pulse response signal of the sensing channel cg. Translate along the time axis with reference channel cg2 Vibration pulse response signal after The product of the two signals is calculated, and the integration process occurs within the duration of the transient vibration pulse response signal. Through cross-correlation analysis, the time offset that enables the vibration pulse response signal of the sensing channel cg to achieve the best match (i.e., the maximum integral value) with the vibration pulse response signal of the reference channel cg2 after translation is found. ,this This refers to the time delay deviation between the sensing channel cg and the reference channel cg2.

[0129] Calculate the amplitude compression factor:

[0130]

[0131] wherein, represents the amplitude compression factor of the sensor channel cg with respect to the reference channel cg2, represents the vibration impulse response signal of the reference channel cg2, represents the vibration impulse response signal of the sensor channel cg.

[0132]

[0133] wherein, represents the phase drift deviation of the sensor channel cg with respect to the reference channel cg2 at frequency f, is the Fourier transform operator, f is the frequency variable, and represents the frequency components of a signal, is the Fourier transform of the vibration impulse response signal of the sensor channel cg, is the Fourier transform of the vibration impulse response signal of the reference channel cg2, () represents taking the phase angle of a complex number.

[0134] For each sensor category, the calibration parameters , , , of each sensor channel are stored in an independent calibration parameter library, and in subsequent real-time monitoring, a low-energy, periodic active excitation source (such as a piezoelectric ceramic actuator installed on the surface of the component) is used to excite the cylindrical shell component. The excitation signal is a short pulse or a beam of a specific frequency, and its energy is sufficient to excite stress waves and structural vibrations that can be captured by the sensor array, but it will not affect the integrity of the component. Through this active excitation, high-quality, synchronized acoustic emission and vibration signals for subsequent crack identification can be periodically obtained, and the collected vibration response signals (t) are compensated using their corresponding calibration parameters to obtain standard vibration signal data :

[0135]

[0136] wherein, represents the phase drift deviation of the sensor channel cg with respect to the reference channel cg2 at frequency f, is the zero-drift baseline value of the cgth sensor channel, represents the amplitude compression factor of the sensor channel cg with respect to the reference channel cg2, is the inverse Fourier transform operator, is the Fourier transform operator, is the imaginary unit, is the time shift the subsequent vibration response signal, represents the time delay deviation of the sensing channel cg relative to the reference channel cg2.

[0137] Similarly, the acoustic emission response signal the correction method of the vibration response signal (t) is the same way, the same calibration steps need to be taken for the acoustic emission channel sensor, and finally the standard acoustic emission signal data is obtained. The final acoustic vibration signal data includes and .

[0138] It should be noted that in the real-time monitoring stage, a low-energy, periodic active excitation source is used to excite the cylindrical shell member. The excitation signal is a short pulse or a beam of a specific frequency, and its energy is sufficient to excite stress waves and structural vibrations that can be captured by the sensor array, but it will not affect the integrity of the member. Through this active excitation, high-quality, synchronized acoustic emission and vibration response signals for subsequent crack identification algorithms can be periodically obtained.

[0139] It should be noted that the calibration excitation is applied in the initial installation without cracks. The requirement is highly standardized and repeatable, and the purpose of calibration is to accurately measure the inherent parameters (baseline, noise, inter-channel difference) of the sensor network itself, providing a reference for subsequent signal compensation. Once this parameter library is established, it can be used for a long time when the sensor network does not change. The real-time monitoring excitation is periodically applied to the member, and the calibration requirement is engineering reliability and convenience, which can effectively excite stress waves / vibrations containing the required frequency components. The monitoring algorithm identifies cracks by analyzing the difference between the response of this excitation and the theoretical intact model, rather than directly comparing with a calibration response. Therefore, the monitoring excitation does not require complete consistency with the calibration excitation, but it should ensure that key parameters such as the main frequency range are within the system design range.

[0140] After completing the baseline calibration of all sensors, accurately map the physical installation points of all M sensor unit pairs to the two-dimensional unfolded coordinate diagram of the cylindrical shell (side view); this diagram is generated using the equal interval method or the center line method, and forms an orthogonal coordinate system with the hoop coordinate and the axial coordinate. Each unit pair is represented as a unique coordinate point on the diagram and is labeled with a number. Through the sensor unit pair number, each installation point in the point diagram is accurately associated with the corresponding two calibration records (one acoustic emission and one vibration) in the baseline library, forming a one-to-one mapping relationship between space and data.

[0141] Step S2: Pulse detection is performed on the acoustic vibration signal data, and the acoustic vibration pulse family feature vector is calculated.

[0142] The standard vibration signal data of the acoustic vibration signal data The short-time moving average of the standard vibration signal data is calculated The dynamic signal of zero baseline is obtained by subtracting the short-time moving average from the standard vibration signal data :

[0143]

[0144] wherein, is the vibration dynamic signal of zero baseline of the cgthchannel, is the standard vibration signal data of the cgthchannel, is the length of the short-time time window.

[0145] It should be noted that the length of the short-time time window should be selected according to the signal characteristics, and the value should be much larger than the main period of the signal, but less than the typical duration of the pulse family to be detected. For high-frequency acoustic emission signals, 1-5 ms can be usually taken; for low-frequency vibration signals, 10 to 50 ms can be usually taken.

[0146] It should be noted that the signal is processed into the dynamic signal of zero baseline, the short-time average value of the signal itself is subtracted in real time, the average value is forced to be zero, and the slow baseline fluctuation caused by the sensor electronic noise, temperature drift and the like is eliminated, so that the signal always fluctuates up and down around the zero value line. A stable and reliable reference criterion is provided for subsequent amplitude threshold detection; if this processing is not performed, the drifting baseline will cause the threshold to fail, thereby causing serious false alarms (noise is mistaken for an event) or missed alarms (real crack signals are submerged). Step S2 receives the standard signal data calibrated in S1, it should be noted that step S1 calibrates the static differences between the sensor channels, and the zero baseline processing in this step aims to eliminate the dynamic baseline drift caused by electronic noise, temperature transient and the like during signal acquisition in real time. The levels and purposes are different, and they work together to ensure the accuracy of the subsequent pulse detection.

[0147] The dynamic amplitude threshold value for pulse detection is calculated :

[0148]

[0149] wherein, is the dynamic threshold value, is a noise adjustment coefficient, the value range is [4, 6], the higher the value, the more conservative the detection, is the real-time noise standard deviation.

[0150] It should be noted that the noise adjustment coefficient The monitoring should include environmental noise intensity. For low-noise environments with weak electromagnetic interference and stable equipment vibration (e.g., laboratories, precision machinery workshops), a value of 4 to 5 is recommended. For high-noise environments with strong electromagnetic interference and multiple devices operating collaboratively (e.g., chemical processing workshops, metallurgical plants), a value of 5 to 6 is recommended. Real-time noise standard deviation is also important. To extract (5-10) pulses within the current monitoring period before pulse detection. The standard deviation of the standard vibration signal data within the range is denoted as the real-time noise standard deviation. .

[0151] When | |> At that time, it is marked as the starting point of the vibration event. When the signal amplitude remains below the threshold Duration The point where the vibration event ends is marked after two signal sampling cycles. .Will[ , The signal segment within the time window is extracted as a candidate pulse family for vibration to be analyzed. For example, if the sampling frequency is 1MHz, then the sampling period is 1... If it remains below the threshold 2 The decision is now complete.

[0152] For each candidate vibration pulse family, peak detection is performed within the vibration pulse family signal segment using the sliding window maximum value method. The window length is set to 1 / 5 of the dominant period of the signal (the dominant period is calculated from the dominant frequency of the steady-state excitation signal in step S1; for example, if the dominant frequency is 1kHz, the dominant period is 1ms, and the window length is 0.2ms). When the amplitude of the center sampling point within the window is greater than the amplitudes of all other points within the window, and this amplitude > When the peak value is determined to be a valid pulse peak, the peak time sequence is finally obtained. Where N is the total number of pulses in the vibration pulse family, and the maximum value of the peak value of all pulses is obtained. As the amplitude of the vibration signal.

[0153] Calculate the time difference between adjacent pulse peaks to obtain the pulse interval sequence Δ [n]:

[0154]

[0155] in, Expressing the nth pulse interval, This indicates the peak time of the nth pulse. This indicates the peak time of the (n+1)th pulse.

[0156] The average pulse interval and interval standard deviation of the vibration pulse family are calculated to describe its time distribution characteristics:

[0157]

[0158] wherein, is the average pulse interval, N is the total number of pulses, and n is the pulse index.

[0159]

[0160] wherein, is the standard deviation of the vibration pulse interval.

[0161] The envelope e(t) of the vibration pulse family signal is calculated using Hilbert transform:

[0162]

[0163] wherein, is the analytical envelope of the vibration pulse family signal, is the vibration dynamic signal of the zero baseline of the cgthchannel, and j is the imaginary unit, is the Hilbert transform.

[0164] The envelope fluctuation strength is calculated, and the greater the value, the more intense and unstable the energy release:

[0165]

[0166] wherein, is the vibration envelope fluctuation strength, T is the duration of the vibration pulse family, and T= is the vibration pulse time starting point, is the vibration pulse ending time point, is the mean value of the analytical envelope.

[0167] The main frequency band of the extracted pulse family is extracted. The power spectrum P(f) of the vibration pulse signal band of the vibration pulse family is obtained by performing fast Fourier transform, and first, all potential peaks in the power spectrum that satisfy the peak amplitude of the global highest peak value are identified. For each potential peak , its 3dB low frequency cutoff and high frequency cutoff are determined according to P(f)= , forming the frequency band ; The energy of each frequency band is calculated ; finally, the frequency band with the largest energy is selected as the dominant frequency band of the vibration pulse family . ​​​​

[0168] For each vibration candidate pulse family detected in each channel, a vibration pulse family feature vector is obtained =[ , , , , , ] as the digital representation of the pulse family, and the pulse family detection is performed on the acoustic emission channel to obtain an acoustic emission pulse family feature vector =[ , , , , , ] wherein, is the acoustic emission event start point, is the acoustic emission event end point, is the standard deviation of the acoustic emission pulse interval, is the acoustic emission envelope fluctuation strength, is the acoustic emission dominant frequency band, is the acoustic emission signal amplitude. Finally, a vibration-acoustic pulse family feature vector is obtained, including the vibration pulse family feature vector and the acoustic emission pulse family feature vector.

[0169] Step S3: consistency test is performed on the vibration-acoustic pulse family feature vector to obtain a pulse variation coefficient, a frequency band matching degree analysis is performed on the vibration-acoustic pulse family feature vector to obtain a frequency band matching index, and a crack confidence is calculated by combining the pulse variation coefficient and the frequency band matching index. When the crack confidence is greater than a preset threshold, it is determined that the cylindrical shell member has a crack.

[0170] For the vibration pulse family feature vector in the vibration-acoustic pulse family feature vector, the in the vibration pulse family feature vector is taken as a time reference, and a time alignment window (e.g., 1-5 milliseconds) is set, which should be greater than the maximum expected time for the wave to propagate between adjacent sensors, but much smaller than the minimum interval between two independent events. All vibration pulse family feature vectors are sorted according to their start times , and a vibration pulse family trigger time is taken as a seed to search for all vibration pulse families that meet the condition to form an initial time cluster. Then, the standard deviation of the pulse peak time in the cluster is calculated, and if the standard deviation is too large, a clustering algorithm based on the peak time difference (such as DBSCAN) is used to purify the initial cluster to remove outliers and form a final candidate event cluster. The consistency test is performed on the K vibration pulse family feature vectors in the candidate event cluster to calculate the variation coefficient of the vibration pulse interval standard deviation in the candidate event cluster :

[0171]

[0172] where, is the coefficient of variation of the pulse interval standard deviation, is the pulse interval standard deviation of the Kth pulse family eigenvector, std() is the standard deviation function, and mean() is the mean function.

[0173] The coefficient of variation of the fluctuation intensity of the vibration envelope is calculated as:

[0174] where, is the coefficient of variation of the fluctuation intensity of the vibration envelope, is the fluctuation intensity of the vibration envelope of the Kth pulse family eigenvector.

[0175] The coefficient of variation of the vibration pulse is then calculated as:

[0176]

[0177] where, is the coefficient of variation of the vibration pulse, is the weight of the pulse interval standard deviation, is the weight of the fluctuation intensity of the envelope, + = 1, is the threshold value of the pulse interval standard deviation, is the threshold value of the fluctuation intensity of the envelope.

[0178] It should be noted that the weights of the fluctuation intensity of the envelope and the pulse interval standard deviation are determined based on the principal component analysis (PCA) contribution of the historical data set to meet the best suppression effect on a specific type of interference. and By collecting a large number of interference signals in the crack-free state, the distribution of the corresponding characteristic values (such as the 99th percentile) is counted to calibrate.

[0179] In the same way as the calculation of the coefficient of variation of the vibration pulse , the coefficient of variation of the acoustic emission pulse is calculated .

[0180] For the candidate event cluster, based on the dominant frequency band of the acoustic emission channel signal in the pulse family eigenvector and the dominant frequency band of the vibration channel signal , the weighted average values of the dominant frequency bands of the acoustic emission and vibration signals in the candidate cluster are calculated respectively:

[0181] ​

[0182] wherein, is the acoustic emission dominant frequency band weighted average value, is the acoustic emission dominant frequency band of the kth acoustic emission pulse family eigenvector, is the acoustic emission amplitude of the kth acoustic emission pulse family eigenvector.

[0183]

[0184] wherein, is the vibration dominant frequency band weighted average value, is the vibration dominant frequency band of the kth vibration pulse family eigenvector, is the vibration signal amplitude of the kth vibration pulse family eigenvector.

[0185] By analyzing the band matching degree of the pulse family eigenvector, a band matching index is obtained:

[0186]

[0187] wherein, is the band matching index, is the frequency alignment tolerance, D is the energy barycenter space Euclidean distance of the acoustic emission and vibration signals, is the energy barycenter distance threshold.

[0188] It should be noted that the frequency alignment tolerance A reasonable range is usually preset according to the sensor characteristics, for example, based on the fluctuation characteristics of the dominant frequency bands of the acoustic emission and vibration signals, the energy barycenter distance threshold By collecting a large number of interference signals in the crack-free state, the signals in the crack-free state only contain pure interference such as environmental noise and equipment vibration, by collecting a large number of such signals and calculating the distribution of the energy barycenter distance (such as taking the 99% quantile as the threshold), a reliable benchmark can be established to distinguish real crack events from random interference. The reason why it is not dynamically calibrated in real-time acquisition is that real-time signals may be mixed with crack components, which will pollute the accuracy of threshold calibration.

[0189] By combining the vibration pulse coefficient of variation, the acoustic emission pulse coefficient and the band matching index, a crack confidence is calculated:

[0190]

[0191] wherein, is the crack confidence, is the vibration pulse coefficient of variation, Band matching index, Acoustic emission pulse coefficient of variation.

[0192] It should be noted that the crack confidence is calculated in combination with the vibration pulse coefficient of variation, the acoustic emission pulse coefficient and the band matching index. The vibration pulse coefficient of variation can exclude single-channel misjudgment of the vibration channel caused by sensor individual difference or local interference by testing the consistency of the multi-channel pulse family characteristics (such as pulse interval standard deviation and envelope fluctuation intensity) of the vibration channel, and ensure that the vibration signal meets the synchronization rule of the true crack excitation at the feature level. The acoustic emission pulse coefficient can exclude single interference of the acoustic emission channel in the same way by testing the consistency of the multi-channel pulse family characteristics (such as pulse interval standard deviation and envelope fluctuation intensity) of the acoustic emission channel, and ensure that the acoustic emission signal matches the synchronization of the crack event at the feature level. The band matching index can verify whether the two types of signals are derived from the stress wave-structure vibration correlation event caused by the crack or irrelevant environmental noise or equipment interference by correlating the dominant band weighted average value and the energy barycenter spatial Euclidean distance of the acoustic emission and vibration signals, and ensure that the signal has the physical properties of the crack event at the energy correlation level. The crack confidence formed by the cubic root geometric average combination of the three can realize the three-dimensional cross verification of “vibration channel feature consistency, acoustic emission channel feature consistency, and cross-channel energy correlation”, which can more comprehensively exclude single-channel interference and non-correlated signals, further improve the accuracy of crack existence determination, and effectively avoid missed judgment and misjudgment.

[0193] The final decision logic is based on the binary decision of the comprehensive confidence index: if the crack confidence is greater than or equal to the preset threshold , the candidate event is confirmed as a real crack event and output to the subsequent analysis link; otherwise, if the crack confidence is lower than the threshold, the event is regarded as an interference signal and is removed.

[0194] Step S4: After determining the crack existence, the wave propagation feature vector is calculated based on the acoustic vibration pulse family feature vector; the correlation analysis is performed on the wave propagation feature vector and the preset theoretical propagation feature vector by the dynamic event regularization method to obtain the wave propagation correlation coefficient; the variance of the first-order difference of the wave propagation feature vector is calculated to obtain the wave propagation difference variance.

[0195] This step takes the output of S3, and the sensors included in the event confirmed as a real crack are mapped to their coordinates , ), wherein ∈[0,2π) is the ring coordinate (radian or arc length), ∈[0,L] represents the axial coordinates (L is the shell length). Analyze the spatial distribution of the confirmed energy clusters to determine the direction of energy propagation. Map each pulse family within the energy cluster to its corresponding sensor's spatial coordinates. For each signal channel in each type of sensor, taking the vibration channel as an example, the vibration signal amplitude is... Using the eigenvectors of the pulse family as weights, calculate the weighted average center coordinates of the event. , As the initial positioning point:

[0196] ;

[0197] ;

[0198] Where K is the total number of vibration pulse family feature vectors in the candidate event cluster, and k is the index of the vibration pulse family feature vector.

[0199] It should be noted that the calculation formulas given in step S4 are all illustrated using data such as the peak amplitude of the vibration channel as examples. While the acoustic emission channel and the vibration channel are complementary in crack detection, once the localization stage begins, the weighted average center coordinates calculated by the acoustic emission channel and the vibration channel are relatively close, and the theoretical propagation paths and amplitude attenuation laws of the two types of signals are similar. Furthermore, vibration accelerometers typically have superior low-frequency response and higher signal-to-noise ratio, enabling them to more stably capture the overall structural vibration modes induced by cracks. Their signal attenuation laws and propagation time delays are more consistent with classical elastic wave propagation theory on an engineering scale, thus making the wave propagation characteristic vector calculation model based on the vibration channel more robust and practical for engineering applications. Therefore, only data from the vibration channel is used for crack location detection.

[0200] With coordinates ( , The analysis region is defined around the center. Within this region, it is divided into I grids along the circumferential direction and J grids along the axial direction, forming I... J is a regular grid, and each grid point G(i,j) corresponds to a unique continuous coordinate. , ), , , , Let i = 1, 2, ..., I, j = 1, 2, ..., J; I and J take... An odd number of 5 such that ( , The corresponding intermediate grid point is defined. For each grid point G(i,j), a set of angles with the axis is defined. Angled passage (such as) =±30°, ±45°, ±60°).

[0201] It should be noted that the fundamental reason for setting up multi-directional propagation analysis is that a single-directional model cannot accurately describe the physical process of wave propagation on a curved surface. By setting up multiple sets of typical directions (such as circumferential 0°, axial 90°, and ±45° oblique directions) for parallel analysis, cross-validation can be achieved by utilizing the response differences of wavefronts in different directions to the spatial location of the crack. The redundancy design between directions resists interference from a single path, thereby significantly improving the accuracy and reliability of the positioning results.

[0202] For each candidate point G(i,j) in the search grid and each preset direction of the candidate point Candidate point G(i,j) is considered as the starting point of wave propagation time, and the wave propagation begins from this point along the direction... Generate virtual sensing paths. Along the virtual paths in each direction, find the set of sensor numbers involved in the real crack event determined in step S3, and record the signal arrival time of the vibration pulse family feature vector corresponding to each sensor. The peak amplitude of the signal recorded by each sensor For any direction to be analyzed On the development diagram of the cylindrical shell, define a line that makes an angle of θ with the axis. The narrowband analysis region is defined. All sensor points located within or near this narrowband region are selected to form the analysis point set for that direction. ={( , )}.right Sort the data according to their projected distances relative to the theoretical wave source point to obtain an ordered sequence. Find the earliest arrival time. =min( The corresponding sensor Calculate the relative positions of each point. Time difference: Generate time series sequences Find the maximum amplitude. =max( The corresponding sensor Calculate the relative points Amplitude attenuation ratio: Generate amplitude sequence .

[0203] It should be noted that in the narrow-band analysis region, starting from candidate point G(i,j), a line is drawn on the cylindrical shell unfolded diagram with an angle of θ to the axis. The theoretical propagation line is used as the center line. Then, based on the pre-set tolerance width parameter (which is usually 1 to 2 times the average spacing between sensors), a parallel strip-shaped region is generated with the center line as the reference and the vertical distance between the two sides of the boundary being the tolerance width parameter / 2. This region is the narrow band analysis region. Finally, the actual set of sensors participating in the analysis in this direction is determined by calculating and screening the vertical distance from all sensor coordinate points to the center line (less than or equal to the tolerance width parameter / 2).

[0204] For a starting point G(i,j) and a direction angle of... The path, based on the sound velocity model of an intact shell. and decay model Generate theoretical sequences: and ,in, The path (with candidate point G(i,j) as the starting point of wave propagation, along a preset direction) The distance from the nth point on the extended virtual sensing path to the starting point.

[0205] It should be noted that in the sound speed model With decay model The following parameters were determined through prior experimental calibration: Under a crack-free cylindrical shell component, a standard excitation was applied at a known location, and for each direction angle... The wave velocity c in that direction is calculated by measuring the ratio of the time difference of arrival of the signal from the sensor pair to its geodesic distance along the shell. )= Geodetic distance refers to the shortest path distance between two physical sensor locations along the casing; simultaneously, by fitting the exponential decay law of the received signal amplitude A(d) with distance d at different distances, A(d) = * Inversely calculate the attenuation coefficient in this direction Finally, a continuous model is formed by fitting functions to the data in all directions.

[0206] For each grid point G(i,j), along FX preset directions A theoretical wavefront propagation sequence is generated. To prevent nonlinear distortion of the sequence due to missing sensors or positioning errors, the wave propagation characteristic vector is calculated using the dynamic event warping method, and correlation analysis is performed between the wave propagation characteristic vector and the preset theoretical propagation characteristic vector for each direction to obtain the wave propagation correlation coefficient.

[0207] Time-series correlation index ;

[0208] Amplitude Correlation Index ;

[0209] Calculate all directions of the current grid pointMaximum value of and minimum value of is normalized as Similarly, is normalized as .

[0210] Wave propagation correlation coefficient :

[0211]

[0212] wherein, is the wave propagation correlation coefficient, is the time series weight, is the normalized time series correlation index, is the amplitude weight, is the normalized amplitude correlation index, FX is the total number of directions, fx is the direction index, and DTW() is the dynamic time warping function.

[0213] It should be noted that, in the stress wave propagation process of the cylindrical shell member, the regularity of the time series of wave arrival time is generally more stable than that of the amplitude decay sequence, because the time information is mainly affected by the wave velocity, which is a material inherent property and less affected by the environment; in contrast, the signal amplitude is significantly affected by the propagation path attenuation, the change of sensor coupling state, and random noise, and its reliability is relatively low. Therefore, in practical applications, the time series weight is usually given a higher value (for example, = 0.8), while the amplitude weight is given a lower value (for example, = 0.2), and + = 1.

[0214] Quantify the smoothness of the measured sequence itself and identify abnormal jumps. Calculate the variance of the first-order difference of the wave propagation feature vector to obtain the wave propagation difference variance:

[0215]

[0216] After maximum and minimum value normalization of and , the wave propagation difference variance is calculated:

[0217]

[0218] wherein, is the wave propagation difference variance, is the variance of the first-order difference of the time series, and ​​The variance of the first-order difference of the amplitude.

[0219] Step S5: Calculate the comprehensive difference degree by combining the wave propagation correlation coefficient and the wave propagation difference variance; based on the comprehensive difference degree, obtain the crack candidate abnormal point, and perform neighborhood propagation analysis on the crack candidate abnormal point to obtain a high-confidence crack source area.

[0220] The comprehensive difference degree is calculated by combining the wave propagation correlation coefficient and the wave propagation difference variance:

[0221]

[0222] Wherein, The comprehensive difference degree is, The wave propagation correlation coefficient is, The wave propagation difference variance is.

[0223] The value of is mapped to the cylindrical shell development diagram grid to generate a two-dimensional difference mapping diagram, and coordinate points with a comprehensive difference degree greater than a threshold value are recorded as candidate abnormal points.

[0224] It should be noted that the comprehensive difference degree is calculated by combining the wave propagation correlation coefficient and the wave propagation difference variance. The closer the wave propagation correlation coefficient is to 1, the more similar the measured sequence is to the theoretical sequence, indicating that the measured wave propagation conforms to the normal law of the structure without cracks, and the more normal the measured wave is, the lower the possibility of the existence of cracks. The closer the wave propagation correlation coefficient is to 0, the more the measured sequence deviates from the theoretical sequence, indicating that the measured wave propagation is destroyed (such as wave reflection / refraction caused by cracks), and the more abnormal the measured wave is, the higher the possibility of the existence of cracks. The wave propagation difference variance is an exponential mapping value of the first-order difference variance of the measured sequence. The greater the value, the more normal the measured wave is; the smaller the value, the more abnormal the measured wave is. The "normal / abnormal" correlation logic of the wave propagation correlation coefficient is completely consistent, both of which are "greater value is normal and smaller value is abnormal", and they provide dual abnormality judgment basis of "theoretical consistency degree" and "signal smoothness" for the comprehensive difference degree.

[0225] Perform spatial clustering analysis (such as the DBSCAN algorithm) on all candidate abnormal points, and merge the points connected in space to form a plurality of independent and connected suspected crack source areas.

[0226] For the suspected crack source area G, each suspected crack source area is composed of a series of grid points. All grid points in the suspected crack source area G are regarded as a plane object with uniform mass, and the geometric center O (i.e. the arithmetic mean of all coordinates) is calculated. With the center O as the center, an annular peripheral analysis region is defined, wherein, ( , To the suspected crack initiation area Euclidean distance from the geometric center Inner diameter The outer diameter is used to ensure an appropriate distance between the analysis area and the source region in order to capture the true energy propagation effect. After obtaining the suspected crack source region, based on the pulse family feature vectors FV accumulated within the current detection time window (e.g., 15 minutes), the event-weighted center coordinates located at... The pulse family feature vectors with a crack confidence level greater than 0.8 constitute a set of high-confidence events. .

[0227] With center O as the origin, along multiple preset directions (Same as step S4) Draw a ray that extends to the outer area. The boundary. These rays are the radial analysis paths. ,in It is the direction angle.

[0228] For each high-confidence event in the outer region point ( , ), calculate its path to each radiation path The vertical distance to the event point. ( , It is assigned to the radiation path that is closest to it. Up. Simultaneously, calculate the event point. ( , ) in path Find the projection point on the center O and calculate the radial distance from that projection point to the center O. For each radiation path Collect all event points belonging to this path. Arrange these points according to their radial distance from the center O. Sort the distances from smallest to largest. This establishes a distance sequence from the "nearest source" to the "farest source". Calculate the gradient between adjacent points on this ordered distance sequence: Where c is the index of the c-th point in the distance sequence. This represents the vibration amplitude at point c. Let c represent the distance from the c-th point to the center. Calculate the time delay gradient between adjacent points on this ordered distance sequence. ,in, Indicates the time to reach the c-th point; if the path Simultaneously satisfying spatial and trend continuity ( <0 and If the ratio of the points with a value greater than 0 (denoted as high-confidence event points) exceeds 0.7, it is considered that the path is an effective propagation trace, indicating that there is a crack propagation behavior in this direction that conforms to the physical law. If the points on a path do not satisfy the law, it may mean that there is no crack propagation in this direction, or the signal is disturbed.

[0229] Based on the effective propagation traces of each suspected crack source area, the comprehensive confidence score of the crack source area is calculated:

[0230]

[0231] wherein, is the crack source area comprehensive confidence score, is the crack source area path length coherence confidence score, is the path multi-direction confidence score.

[0232] It should be noted that the crack source area path length coherence confidence score is calculated based on the average length of all effective propagation traces of the source area. The more event points a trace contains, the higher the value, indicating stronger statistical significance. For each effective propagation trace of the suspected crack source area G , the number of high-confidence event points it contains is counted , the average number of high-confidence event points of all finite propagation traces in the suspected crack source area G is calculated, =min( ,1), the minimum high-confidence event point threshold is 3 by default. The path multi-direction confidence score is calculated based on the number of directions of the effective propagation traces possessed by the source area. The more different directions are simultaneously pointed to by independent traces, the higher the value, indicating stronger spatial consistency of the evidence, =min( ,1), the minimum effective propagation trace number threshold is 2 by default.

[0233] According to the comprehensive confidence score , the suspected crack source area is finally classified: high-confidence source area: ≥0.8, confirmed as a real crack and entered the risk assessment. Medium-confidence source area: 0.5≤ <0.8, marked as to be observed and recommended for enhanced monitoring. Low-confidence source area: <0.5, considered as interference and removed.

[0234] Step S6: The cylindrical shell component is divided into micro-zones to obtain cylindrical shell partitions. The risk of the cylindrical shell partitions is quantified by the analytic hierarchy process to obtain the weight of the cylindrical shell partitions. The partition crack risk level is obtained by combining the weight of the cylindrical partitions and the number of high-confidence crack source areas in the partitions, thereby realizing crack identification of the cylindrical shell component.

[0235] In step S6, the transition from crack detection positioning to risk assessment phase is made, which directly depends on the output results of step S5, ensuring the coherence and practicality of the process. Step S5 obtains high-confidence crack source areas through the comprehensive difference and neighborhood propagation analysis. Step S6 first divides the cylindrical shell component into micro-zones, then quantifies the inherent risk weight of each partition through the analytic hierarchy process, and counts the number of high-confidence crack source areas in each partition (spatial mapping of the geometric center coordinates of each high-confidence crack source area). Finally, the number of high-confidence crack source areas is taken as input, combined with the partition weight to calculate the comprehensive risk value, so as to determine the comprehensive risk level of each partition (low risk, medium risk, high risk).

[0236] The cylindrical shell component is divided into micro-zones, and the partition principle is determined according to the mechanical simulation and structural characteristics: based on finite element analysis, high stress concentration areas and geometric discontinuity areas are extracted as independent partitions, and geometric discontinuity and material heterogeneity parts are preferentially identified, including welds (including heat-affected zones), hole edges, support connection transition zones, and ring-shaped rib plate and shell connection parts. These parts are prone to stress concentration due to structural mutations and need to be separately divided into independent partitions. Then, the remaining areas are divided into regular grids according to the sensor array density, ensuring that the size of a single partition is not greater than the positioning accuracy, and finally a partition set covering the entire cylindrical shell is generated.

[0237] The analytic hierarchy process is used to quantify the inherent risk weight of each partition, and the specific process is as follows: (1) Construct a hierarchical structure model: take the partition risk assessment as the target layer, take the key factors affecting the partition risk (such as average stress level, stress concentration coefficient, fatigue sensitivity, detectability, and material performance) as the criterion layer, and take each partition as the scheme layer. (2) Construct a judgment matrix: for each factor in the criterion layer, use the 1-9 scale method to compare each partition with each other, and construct a judgment matrix A, where the element represents the importance of partition f1 relative to partition f2 for this risk factor. (3) Calculate the weight vector and check the consistency: calculate the maximum eigenvalue of the judgment matrix A and its corresponding eigenvector, normalize the eigenvector to obtain the weight vector of each partition for this risk factor. Then calculate the consistency ratio CR= , if CR<0.1, the consistency of the judgment matrix is acceptable, otherwise the judgment matrix needs to be adjusted. (4) Calculate the comprehensive weight: repeat steps 2-3 to calculate the weight of each risk factor relative to the total target, and finally obtain the comprehensive inherent risk weight of each partition by weighted synthesis .

[0238] The risk of the cylindrical shell partition is quantified by the analytic hierarchy process, and the weight of the cylindrical shell partition is obtained:

[0239]

[0240] wherein, is the zone weight of the cylindrical shell, is the ith evaluation dimension, is the weight of the ith evaluation dimension.

[0241] It should be noted that the ith evaluation dimension is the specific value of the five key factors of the zone risk, such as the average stress level, the stress concentration coefficient, the fatigue sensitivity, the detectability, and the maximum and minimum normalized values of the material performance, i.e. .

[0242] For each micro-zone, the comprehensive risk value R is calculated:

[0243]

[0244] wherein, R is the comprehensive risk value, is the weight of the zone, is the adjustment coefficient, which is 0.5 by default, represents the number of high-confidence crack source zones in the zone.

[0245] Based on the size of the comprehensive risk value R, the crack risk level of each zone is divided into three levels: low risk, medium risk, and high risk. When R is lower than the threshold value R_low, it is determined to be a low risk level; when R is between the threshold values R_low and R_high, it is determined to be a medium risk level; when R is higher than the threshold value R_high, it is determined to be a high risk level, thereby realizing the division of the crack risk level of the cylindrical shell component.

[0246] A complete internal crack identification method for cylindrical shell components is proposed in this paper. The method realizes the whole process identification from signal acquisition to risk assessment through six core steps: first, the acoustic emission and vibration sensor arrays are arranged on the surface of the cylindrical shell according to the latitude and longitude grid principle and the system self-calibration is completed, and the vibration signals are collected; then the pulse detection is performed on the signals, and the pulse family feature vector is calculated; then the crack confidence is obtained through consistency test and frequency band matching degree analysis, and it is judged whether there is a crack; after confirming the crack, the wave propagation feature vector is calculated based on the pulse family feature vector, and the wave propagation correlation coefficient and the wave propagation difference variance are obtained by combining the dynamic event regularization method and the first-order difference variance; then the comprehensive difference degree is calculated by combining the two, the crack candidate abnormal point is located, and the high-confidence crack source zone is obtained through neighborhood propagation analysis; finally, the micro-zone division is performed on the cylindrical shell, the zone weight is quantified by the analytic hierarchy process, the zone crack risk level is calculated combined with the number of high-confidence crack source zones in the zone, and the crack identification and risk assessment are realized.

[0247] The crack confidence is obtained by combining the pulse variation coefficient and the band matching index, and a dual decision mechanism of feature consistency and energy matching degree is constructed. The pulse variation coefficient tests the consistency of the characteristics of the pulse family in different channels, and excludes the single channel misjudgment caused by individual differences or local interference of the sensor; the band matching index ensures that the recognized signal is related to the stress wave and structural vibration caused by crack, rather than irrelevant environmental noise or equipment interference, by correlating the dominant frequency band and energy center distance of acoustic emission and vibration signals. The combination of the two effectively improves the accuracy of crack existence judgment and avoids missed judgment, providing reliable pre-judgment basis for subsequent crack positioning and analysis.

[0248] The comprehensive difference degree is calculated by combining the wave propagation correlation coefficient and the wave propagation differential variance, and its key significance lies in realizing the accurate screening and positioning of crack location. The wave propagation correlation coefficient compares the measured crack evolution characteristics with the theoretical evolution characteristics by dynamic event regularization method to judge whether the signal conforms to the physical law of crack propagation; the wave propagation differential variance quantifies the smoothness of the evolution feature vector to identify the abnormal jump of the signal caused by the existence of cracks. The comprehensive difference degree combines the two after normalization, which can not only exclude interference points that do not conform to the theoretical evolution law, but also filter out real crack-related points with abnormal signal fluctuations, and then form a high-confidence crack source area through neighborhood propagation analysis, greatly improving the accuracy and reliability of crack positioning.

[0249] The partition crack risk level is calculated by combining the cylindrical shell partition weight and the number of high-confidence crack source areas in the partition, and its core significance lies in realizing the differentiated and targeted evaluation of crack risk. The partition weight quantifies the inherent risk basis of different regions (such as higher weight for high stress concentration areas) by combining stress level, material performance, geometric characteristics, etc. through the analytic hierarchy process; the number of high-confidence crack source areas reflects the distribution density of actual cracks in the region. The risk level calculated by combining the two is no longer a general assessment of the whole component, but can accurately identify the key areas with high inherent risk and multiple crack distribution, providing clear direction for subsequent maintenance decisions and helping to focus on high-risk partitions for repair, improving the efficiency and pertinence of cylindrical shell component safety management.

[0250] It has to be noted that, in the present document, the terms "first", "second", etc. merely serve to identify different entities or actions and do not necessarily require or imply any actual relationship or order between these entities or actions. Moreover, the terms "comprises", "comprising", or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. In other words, without further restriction, reference to elements will not, without more limitations, exclude additional, unrecited elements of a process, method, article, or apparatus.

[0251] While embodiments of the application have been shown and described, it is to be understood that the application is not limited to these embodiments. Rather, it is the intention to cover any variations and modifications, provided they fall within the scope of the application as defined by the appended claims and their equivalents.

Claims

1. A method for identifying internal cracks in a cylindrical shell component, characterized in that: Includes the following steps: Step S1: Arrange a sensor array on the surface of the cylindrical shell component, apply active source excitation to the cylindrical shell component, and collect acoustic vibration signals through the sensor array to obtain acoustic vibration signal data; Step S2: Perform pulse detection on the acoustic vibration signal data and calculate the acoustic vibration pulse family feature vector; Step S3: Perform a consistency check on the acoustic vibration pulse family feature vectors to obtain the pulse variation coefficient. Perform a frequency band matching degree analysis on the acoustic vibration pulse feature vectors to obtain the frequency band matching index. Combine the pulse variation coefficient and the frequency band matching index to calculate the crack confidence level. When the crack confidence level is greater than the preset threshold, the cylindrical shell component has a crack. Step S4: After confirming the existence of the crack, calculate the wave propagation characteristic vector based on the acoustic pulse family characteristic vector; perform correlation analysis on the wave propagation characteristic vector and the preset theoretical propagation characteristic vector using the dynamic event warping method to obtain the wave propagation correlation coefficient; calculate the variance of the first difference of the wave propagation characteristic vector to obtain the wave propagation difference variance. Step S5: Calculate the comprehensive difference degree by combining the wave propagation correlation coefficient and the wave propagation differential variance; based on the comprehensive difference degree, obtain the crack candidate anomaly points, perform neighborhood propagation analysis on the crack candidate anomaly points, and obtain the high confidence crack source region. Step S6: Divide the cylindrical shell component into micro-regions to obtain cylindrical shell partitions. Quantify the risk of each cylindrical shell partition using the analytic hierarchy process (AHP) to obtain the partition weights. Combine the partition weights and the number of high-confidence crack initiation zones within each partition to obtain the crack risk level of the partition, thereby achieving crack identification of the cylindrical shell component.

2. The method for identifying internal cracks in a cylindrical shell component according to claim 1, characterized in that: The process of deploying a sensor array on the surface of the cylindrical shell component includes the following specific steps: Using the circumferential sensing zones of the cylindrical shell component as latitude lines, two types of sensors are arranged adjacently in pairs along the circumferential direction on each circumferential sensing zone: at preset installation coordinate points, acoustic emission sensors and vibration sensors are installed adjacently in pairs, forming a sensor unit pair; the number of sensor unit pairs is determined by the diameter of the cylindrical shell component and the monitoring accuracy, and the required number of sensor unit pairs on each circumferential sensing zone is ≥ The preset maximum arc length spacing is adjusted by the diameter of the cylindrical shell component. The larger the diameter, the smaller the preset maximum arc length spacing should be, so that the sensor array is evenly distributed, thereby ensuring that the arc length distance between adjacent sensors is controlled.

3. The method for identifying internal cracks in a cylindrical shell component according to claim 2, characterized in that: The process of actively exciting the cylindrical shell component and acquiring acoustic vibration signals through a sensor array to obtain acoustic vibration signal data includes the following steps: Under the condition of a crack-free cylindrical shell component, a low-amplitude, wide-frequency steady-state excitation source is applied to the cylindrical shell component, and the acoustic emission channels and all vibration channels are simultaneously acquired for a duration of [duration missing]. Acoustic emission response signal inside (t) and vibration response signal (t), where cg = 1, 2, ..., M; For the vibration sensor in the sensing unit pair, the arithmetic mean of the response signal of each channel in the time domain is calculated as the zero-drift baseline value of that channel: ； in, is the zero-drift baseline value of the cg-th sensing channel, representing the local output of that channel in a crack-free state. The duration of steady-state excitation, The vibration response signal acquired by the cg-th sensing channel; A controllable transient excitation is applied at a fixed position on the cylindrical shell component, and the pulse response signals of all channels are acquired synchronously. The channel with the largest amplitude value in this type of sensor is selected as the reference channel cg2 of this type of sensor. The time delay deviation, amplitude compression coefficient, and phase drift deviation of other channels cg in this type of sensor relative to the reference channel cg2 are calculated: ； in, This indicates the time delay deviation between the sensing channel cg and the reference channel cg2. This is the vibration pulse response signal of the sensing channel cg. This indicates the time-axis translation of the reference channel cg2. The vibration pulse response signal, This is the time offset. Indicates the duration of the transient vibration pulse response signal; Calculate the amplitude compression factor: ； in, This represents the amplitude compression factor of the sensing channel cg relative to the reference channel cg2. This represents the vibration pulse response signal of the reference channel cg2. This represents the vibration pulse response signal of the sensor channel cg; ； in, This represents the phase shift deviation of the sensing channel cg relative to the reference channel cg2 at frequency f. It is a Fourier transform operator, where f is the frequency variable, representing the frequency components of the signal. The pulse response signal of the sensing channel cg Fourier transform, The pulse response signal of the reference channel cg2 Fourier transform, () represents the phase angle when taken as a complex number; For each sensor category, the calibration parameters for each sensor channel will be... , , , Each parameter is stored in a separate calibration parameter library. During subsequent real-time monitoring, any channel will respond to the acquired vibration response signal. (t) Compensation is performed using the corresponding calibration parameters to obtain standard vibration signal data. : ； in, This represents the phase shift deviation of the sensing channel cg relative to the reference channel cg2 at frequency f. This represents the zero-drift baseline value of the cg-th sensing channel. This represents the amplitude compression factor of the sensing channel cg relative to the reference channel cg2. For inverse Fourier transform operators, For Fourier transform operators, It is the imaginary unit. It is a time shift The vibration response signal afterward This indicates the time delay deviation of the sensing channel cg relative to the reference channel cg2; Similarly, acoustic emission response signal (t) correction method and vibration response signal The procedure is the same as for (t), requiring the same calibration steps for the acoustic emission channel sensor to ultimately obtain standard acoustic emission signal data. ; The final acoustic vibration signal data includes and .

4. The method for identifying internal cracks in a cylindrical shell component according to claim 3, characterized in that: The process of pulse detection of acoustic vibration signal data and calculation of acoustic vibration pulse family feature vectors includes the following steps: Standard vibration signal data for acoustic and vibration signal data For example, calculating standard vibration signal data The short-time moving average is used to subtract the standard vibration signal data from the short-time moving average to obtain the dynamic signal with zero baseline. : ； in, This represents the zero-baseline vibration dynamic signal of the cg-th channel. This is the standard vibration signal data for the cg-th channel. The length of the short-time window; Calculate the dynamic amplitude threshold for pulse detection : ； in, For dynamic thresholds, This is the noise adjustment coefficient, with a value range of [4, 6]. A higher value indicates a more conservative detection. The real-time noise standard deviation; When | |> At that time, it is marked as the starting point of the vibration event. When the signal amplitude remains below the threshold Duration The point where the vibration event ends is marked after two signal sampling cycles. ,Will[ , The signal segments within the time window are extracted and used as a candidate family of vibration pulses to be analyzed. For each candidate vibration pulse family, peak value detection is performed within the vibration pulse family signal segment using the sliding window maximum value method to obtain the peak time sequence. , where N is the total number of pulses in the vibration pulse family; Calculate the time difference between adjacent pulse peaks to obtain the pulse interval sequence Δ [n]: ； in, Expressing the nth pulse interval, This indicates the peak time of the nth pulse. This indicates the peak time of the (n+1)th pulse; Calculate the average pulse interval and standard deviation of this vibration pulse family to describe its temporal distribution characteristics: ； in, Where N is the average pulse interval, N is the total number of pulses, and n is the pulse index; ； in, The standard deviation of the vibration pulse interval; The envelope e(t) of the family of vibration pulse signals is calculated using the Hilbert transform: ； in, The analytical envelope of the vibration pulse family of signals. This represents the zero-baseline vibration dynamic signal of the cg-th channel. For Hilbert variations; Calculate the envelope fluctuation intensity, which is the ratio of the standard deviation to the mean of the envelope. A larger value indicates more violent and unstable energy release. ； in, Let T be the intensity of the vibration envelope fluctuation, and T be the duration of the vibration pulse family, T= - , The starting point of the vibration pulse time. The end time of the vibration pulse. To analyze the mean of the envelope, a fast Fourier transform is performed on the vibration pulse signal band of the vibration pulse family to obtain its power spectrum P(f). First, all signals in the power spectrum that satisfy the peak amplitude are identified. The potential peak value of the global maximum peak amplitude, for each potential peak value According to P(f)= Determine its 3dB low-frequency cutoff. and high frequency cutoff , forming a frequency band , ]; Calculate the energy of each frequency band Ultimately, the frequency band with the highest energy was selected as the dominant frequency band for this family of vibration pulses. ; For each candidate vibration pulse family detected by each channel, a vibration pulse family feature vector is obtained. =[ , , , , , This feature vector serves as the digital representation of the pulse family. Similarly, pulse family detection is performed on the acoustic emission channel to ultimately obtain the acoustic emission pulse family feature vector. =[ , , , , , ],in, The starting point of the acoustic emission event. The end point of the acoustic emission event. The standard deviation of the acoustic emission pulse interval. The intensity of acoustic emission envelope fluctuations. This is the dominant frequency band for acoustic emission. The amplitude of the acoustic emission signal is used to obtain the acoustic vibration pulse family feature vector, which includes the vibration pulse family feature vector and the generation pulse family feature vector.

5. The method for identifying internal cracks in a cylindrical shell component according to claim 4, characterized in that: The process of performing a consistency check on the eigenvectors of the acoustic-vibration pulse family to obtain the pulse variation coefficient includes the following steps: For the vibration pulse family eigenvectors in the acoustic-vibration pulse family eigenvectors, the pulse family eigenvectors In Set the time alignment window as the time base. All pulse family eigenvectors are sorted by their start time. Sorting, based on a pulse family trigger time For the seed, search all that meet the criteria. By analyzing the family of vibration pulses under certain conditions, a cluster of candidate events can be obtained. A consistency test is performed on the eigenvectors of the K vibration pulse families in the candidate event cluster, and the coefficient of variation of the pulse interval standard deviation in the candidate event cluster is calculated: ； in, The coefficient of variation is the standard deviation of the vibration pulse interval. This represents the pulse interval standard deviation of the eigenvector of the Kth vibration pulse family, where std() is the standard deviation function and mean() is the mean function. Calculate the intensity of vibration envelope fluctuations Coefficient of variation: ； in, The coefficient of variation of the intensity of the vibration envelope fluctuations. The vibration envelope undulation intensity of the Kth vibration pulse family eigenvector; The coefficient of variation of the vibration pulse is: ； in, The vibration pulse variation coefficient, The weights are the standard deviations of the pulse intervals. The weights for the intensity of envelope fluctuations. + =1, The threshold value is the standard deviation of the pulse interval. The threshold value for the intensity of envelope undulation; Calculation method of the coefficient of variation of the same vibration pulse Calculate the acoustic emission pulse variation coefficient .

6. The method for identifying internal cracks in a cylindrical shell component according to claim 5, characterized in that: The step of performing frequency band matching degree analysis on the acoustic vibration pulse feature vector to obtain the frequency band matching index includes the following steps: Based on the dominant frequency band of the acoustic emission channel signal in the pulse family eigenvectors The dominant frequency band of the vibration channel signal Calculate the weighted average values ​​of the dominant frequency bands of acoustic emission and vibration signals within the candidate clusters, respectively: ； in, The weighted average value of the dominant acoustic emission frequency band. The dominant frequency band of acoustic emission is the eigenvector of the k-th acoustic emission pulse family. The acoustic emission amplitude is the eigenvector of the k-th acoustic emission pulse family. ； in, It is the weighted average value of the dominant frequency band of vibration. The dominant frequency band of the characteristic vector of the k-th vibration pulse family is... The amplitude of the vibration signal is represented by the eigenvector of the k-th vibration pulse family. By performing frequency band matching degree analysis on the eigenvectors of the acoustic-vibration pulse family, the frequency band matching index is obtained: ； in, For frequency band matching index, Where D is the frequency alignment tolerance, and D is the spatial Euclidean distance between the energy centroids of the acoustic emission and vibration signals. This represents the energy centroid distance threshold.

7. The method for identifying internal cracks in a cylindrical shell component according to claim 6, characterized in that: The confidence level for the crack is: The crack confidence level is calculated by combining the vibration pulse variation coefficient, acoustic emission pulse coefficient, and frequency band matching index. ； in, For crack confidence level, The vibration pulse variation coefficient, For frequency band matching index, is the acoustic emission pulse variation coefficient.

8. The method for identifying internal cracks in a cylindrical shell component according to claim 7, characterized in that: The process of performing correlation analysis on the wave propagation characteristic vector and the preset theoretical propagation characteristic vector using the dynamic event normalization method to obtain the wave propagation correlation coefficient includes the following specific steps: For each candidate point G(i,j) in the search grid and each preset direction of the candidate point Candidate point G(i,j) is considered as the starting point of wave propagation time, and the wave propagation begins from this point along the direction... Generate virtual sensing paths; a set of virtual paths and sensor numbers along each direction; and the signal arrival time recorded by the vibration pulse family feature vector corresponding to each sensor. The peak amplitude of the signal recorded by each sensor For any direction to be analyzed On the development diagram of the cylindrical shell, define a line that makes an angle of θ with the axis. The narrowband analysis region is defined, and all sensor points located within this narrowband region are selected to form the analysis point set for that direction. ={( , Generate time series sequences. and amplitude sequence ; For a starting point G(i,j) and a direction angle of... Based on the sound velocity model and attenuation model of the intact shell, the theoretical sequence is generated for the path: and ,in, It is the distance from the nth point on the path to the starting point; Correlation analysis was performed using the dynamic event warping method to calculate the wave propagation characteristic vector and the preset theoretical propagation characteristic vector for each direction, resulting in the wave propagation correlation coefficient: ； ； Calculate all directions of the current grid point Normalization of maximum and minimum values ​​yields ,right Perform maximum and minimum value normalization to obtain ; Wave propagation correlation coefficient : ； in, The correlation coefficient for wave propagation. For time series weights, It is the normalized time-series correlation index. For magnitude weighting, is the normalized amplitude correlation index, FX is the total number of directions, fx is the direction index, and DTW() is the time series dynamic warping function.

9. The method for identifying internal cracks in a cylindrical shell component according to claim 8, characterized in that: The calculation of the variance of the first-order difference of the wave propagation eigenvector, to obtain the wave propagation difference variance, includes the following specific steps: Calculate the variance of the first-order difference of the wave propagation eigenvectors to obtain the wave propagation difference variance: ； Wave propagation differential variance: ； in, Let the wave propagation difference variance be... The variance of the first-order time difference. Let K be the variance of the first-order difference of the amplitude, and K be the total number of pulse family eigenvectors in the candidate event cluster.

10. The method for identifying internal cracks in a cylindrical shell component according to claim 9, characterized in that: The overall degree of difference is: By combining the wave propagation correlation coefficient and the wave propagation difference variance, the comprehensive difference degree is calculated: ； in, To assess the overall difference, The correlation coefficient for wave propagation. Let Variance be the difference between wave propagation variances.

Images