Fatigue crack damage positioning method based on array signal fusion guided wave-hidden Markov model

Abstract

The invention discloses a fatigue crack damage positioning method based on array signal fusion guided wave-hidden Markov model. The fatigue crack damage positioning method is used for improving the accuracy of online positioning of fatigue crack damage of an aviation structure under a service dynamic working condition. According to the method, uncertain distribution of guided wave observation vectors influenced by service dynamic working conditions is represented through a guided wave-hidden Markov model; probabilistic damage factors which are not influenced by service dynamic working conditions are extracted so as to reliably represent the degree of fatigue crack influence on array signals in the piezoelectric sensor network; and taking the probability damage factor as the damage characteristic of the array signal, and enhancing the damage influence through array signal fusion focusing, so as to realize accurate fatigue crack positioning. The method can greatly improve the accuracy of positioning the fatigue crack damage of the aviation structure under the influence of the service dynamic working condition.

Description

Technical Field

[0001] The present invention belongs to the technical field of aircraft structural health monitoring, and particularly relates to a fatigue crack damage location method based on array signal fusion guided wave - hidden Markov model. Background Art

[0002] Modern aircraft have increasingly high requirements for service performance and service life. The aircraft structure is the basis for safely achieving flight missions and meeting service life requirements. The failure of the structure will lead to the failure of the entire aircraft, and the service life of the structure determines the service life of the aircraft. Therefore, ensuring structural integrity is a key issue during the long - term service of aircraft. For aircraft structures, fatigue cracks are one of the most main and dangerous damage forms leading to structural failure. Using Structural Health Monitoring (SHM) technology to achieve accurate online fatigue crack damage diagnosis is of great significance for ensuring service safety and reducing maintenance costs.

[0003] Among many SHM technologies, the SHM method based on guided waves is considered to be one of the most promising methods due to its advantages such as wide monitoring range and sensitivity to small damages. In recent years, damage imaging methods based on piezoelectric sensor arrays have received extensive attention. They use array signal processing algorithms to fuse damage information from multiple monitoring channels, thereby realizing a focused and enhanced characterization of the structural damage location. Currently, a variety of guided wave array imaging methods have been developed, including delay and sum imaging, multiple signal classification, etc. However, the above - mentioned methods still face challenges when applied to aerospace engineering. During the service of aircraft, they usually face complex and harsh dynamic working conditions, including various time - varying uncertain factors, such as service environments like temperature, humidity, salt spray, and operating conditions like aerodynamic loads and variable boundary conditions, which seriously affect the accuracy of fatigue crack damage location, making the reliability of many imaging methods that work well in the laboratory decrease when applied to engineering applications.

[0004] The Hidden Markov Model (HMM) is a powerful probabilistic statistical model that contains a double stochastic process. It can be used for dynamic modeling of the uncertain distribution of guided wave characteristics under dynamic service conditions and to characterize the uncertain damage evolution process. In recent years, some scholars have applied this model to the vibration response signal modeling of rotating machinery structures and achieved mechanical structure fault diagnosis under dynamic service conditions. There have also been some studies that combine this model with guided wave SHM technology to achieve online quantitative diagnosis of fatigue cracks under dynamic service conditions. However, generally speaking, the research on such methods is still in the initial exploration stage. Existing research has only carried out HMM modeling for a single guided wave monitoring channel, with little exploration of the fusion of damage information from array signals, failing to fully utilize the large-area monitoring advantage of guided wave SHM technology and making it difficult to achieve accurate fatigue crack location. Therefore, it is necessary to study the guided wave-HMM method based on array signal fusion, which is of great significance for achieving accurate fatigue crack damage location under the influence of dynamic service conditions. Summary of the Invention

[0005] Aiming at the deficiencies of the above-mentioned existing technologies, the purpose of the present invention is to provide a fatigue crack damage location method based on guided wave-HMM with array signal fusion, so as to solve the problem that the traditional guided wave array imaging method is affected by uncertainties under dynamic service conditions, resulting in inaccurate fatigue crack damage location.

[0006] To achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0007] A fatigue crack damage location method based on guided wave-HMM with array signal fusion of the present invention is as follows:

[0008] 1) Collect m guided wave array signals of P guided wave monitoring channels included in the piezoelectric sensor network when the structure is in a healthy state and under the influence of dynamic service conditions; calculate m damage factors of each monitoring channel as healthy observation vectors, and establish a guided wave-Hidden Markov Model to characterize the uncertain distribution of healthy observation vectors affected by dynamic service conditions;

[0009] 2) Collect the guided wave array signals when the structure is in an unknown fatigue crack damage state and under the influence of dynamic service conditions, and calculate the damage factors of each channel as damage observation vectors;

[0010] 3) Repeat step 2) multiple times, and combine multiple damage observation vectors obtained under the influence of the structure damage state and dynamic service conditions to form an observation vector sequence;

[0011] 4) Input the observation vector sequence obtained in step 3) into the guided wave-Hidden Markov Model established in step 1), and calculate the probability damage factor as the channel feature;

[0012] 5) Divide the entire monitoring area composed of the piezoelectric sensor network into multiple pixel points, set the array signal fusion parameters, calculate the damage probability of each pixel point under each monitoring channel in the array, and accumulate and fuse the damage probabilities of all channels to obtain the imaging pixel value. Combine the imaging pixel values of all pixel points to obtain the pixel value matrix of fatigue crack damage;

[0013] 6) Set the pixel value threshold, search for the pixel points in the pixel value matrix of fatigue crack damage whose pixel values are higher than / equal to the threshold, and set the pixel values of the pixel points lower than the threshold to zero;

[0014] 7) Use the pixel values of the pixel points searched in step 6) as weights, perform weighted summation on their respective coordinates, and obtain the final fatigue crack damage location result.

[0015] Furthermore, the damage factors calculated in step 1) include the normalized cross-correlation moment damage factor DI1 and the root mean square deviation damage factor DI2, and the calculation formulas are as follows:

[0016]

[0017] In the formula, τ is the delay factor, r xy (τ) is the cross-correlation function of the reference and monitoring signals, r xx (τ) is the autocorrelation function of the reference signal, T is the signal length, H(t) is the reference signal, D(t) is the monitoring signal, and n is the order of the statistical moment.

[0018] Furthermore, the steps for establishing the guided wave-hidden Markov model in step 1) are as follows:

[0019] 11) Initialize the hidden Markov model parameters λ, including setting the number of hidden states to 2, forming the hidden state set S = {S1, S2}, where both S1 and S2 represent the structural health state; initialize the state transition probability a ij = 0.5, where i = 1, 2, j = 1, 2; the initial state probability π i = 0.5, and initialize the observation probability distribution b i (o t ) using the uniform initialization Gaussian mixture model clustering method, where o t represents the guided wave observation vector at time t;

[0020] 12) Optimize the hidden Markov model parameters λ through the Baum-Welch algorithm to obtain the trained model parameters a ij , π i and b i (o t ).

[0021] Furthermore, the observation vector sequence established in step 3) is composed of multiple consecutive observation vectors, and its expression is as follows:

[0022] O t ={o t-k+1 ,o t-k+2 ,…,o t}

[0023] In the formula, k is the number of observation vectors that make up the observation vector sequence O t , o t-k+1 is the guided wave observation vector at time t - k + 1, and o t-k+2 is the guided wave observation vector at time t - k + 2.

[0024] Furthermore, step 5) specifically includes:

[0025] 51) Define the damage feature of the p-th monitoring channel in the piezoelectric sensor network as the probability damage factor calculated by its guided wave-hidden Markov model, denoted as PDI p ;

[0026] 52) Divide the entire monitoring area composed of the piezoelectric sensor network into multiple pixel points, where the X and Y directions are each equally divided into c parts, and the position of each pixel point is represented by its coordinates (x, y) in the monitoring area;

[0027] 53) Set the array signal fusion parameters, and calculate the damage probability P p (x, y) of each pixel point under each monitoring channel in the array. The expression is as follows:

[0028]

[0029] In the formula, β is the size parameter of the array signal fusion, which controls the size of the influence area of the probability damage factor on the direct path of each channel of the array. R p (x, y) is the ratio of the sum of the distances from the pixel point (x, y) to the excitation element and the sensing element of the p-th monitoring channel of the array to the length of this monitoring channel. The expression is as follows:

[0030]

[0031] In the formula, (x pa , y pa ) and (x ps , y ps ) respectively represent the coordinates of the excitation element and the sensing element on the p-th monitoring channel of the array;

[0032] 54) Accumulate the damage probabilities of all monitoring channels in the array to obtain the final damage probability of each pixel point, which is used as the imaging pixel value P(x, y). The expression is as follows:

[0033]

[0034] 55) Combine the imaging pixel values of all pixel points to obtain the pixel value matrix of fatigue crack damage.

[0035] Further, step 6) specifically includes:

[0036] According to the pixel value matrix of fatigue crack damage obtained in step 5), set the pixel value threshold to 0.9 times the maximum pixel value, and search for the coordinates of all pixel points whose pixel values are higher than / equal to the pixel value threshold to form a set Z = {(x1, y1), (x2, y2), …, (x Q , y Q )}, where Q is the number of pixel points searched, x Q is the coordinate of the Qth pixel point searched in the X direction, and y Q is the coordinate of the Qth pixel point searched in the Y direction.

[0037] Further, the expression of the fatigue crack damage location result in step 7) is as follows:

[0038]

[0039] In the formula, x diag is the coordinate of the fatigue crack damage location result in the X direction, y diag is the coordinate of the fatigue crack damage location result in the Y direction, x q is the coordinate of the qth pixel point searched in the X direction, and y q is the coordinate of the qth pixel point searched in the Y direction.

[0040] Advantages of the present invention:

[0041] The present invention can effectively solve the problem of inaccurate fatigue crack damage location caused by the uncertainty of the dynamic working conditions of the aircraft during service affecting the guided wave array imaging, and realize the reliable location of structural fatigue crack damage under dynamic working conditions. It is of great significance for ensuring the service safety of the aircraft structure and reducing the operation and maintenance costs of the aircraft structure. Brief Description of the Drawings

[0042] Figure 1 is the flowchart of the method of the present invention;

[0043] Figure 2a is the physical diagram of the aviation lug connection structure used in the embodiment;

[0044] Figure 2b Schematic diagram of the dimensions of the aviation lug connection structure adopted in the embodiment;

[0045] Figure 3a Schematic diagram of the circular piezoelectric sensor array in the embodiment;

[0046] Figure 3b Schematic diagram of the setting of the guided wave array excitation-sensing monitoring channels in the embodiment;

[0047] Figure 4 Schematic diagram of the process of fatigue crack propagation at the hole edge of the lug structure in the embodiment;

[0048] Figure 5 Schematic diagram of the sample set of the health observation vectors of the monitoring channels PZT1 - PZT16 in the embodiment;

[0049] Figure 6 Schematic diagram of the observation probability distribution of the guided wave - hidden Markov model of the monitoring channels PZT1 - PZT16 in the embodiment;

[0050] Figure 7 Diagram of the fatigue crack damage location result based on the array signal fusion guided wave - hidden Markov model under the service dynamic load condition in the embodiment. Detailed implementation manner

[0051] For the convenience of understanding by those skilled in the art, the present invention will be further described below in conjunction with the embodiments and the accompanying drawings. The content mentioned in the implementation manner does not limit the present invention.

[0052] Refer to Figure 1 As shown, a fatigue crack damage location method based on the array signal fusion guided wave - hidden Markov model of the present invention is as follows:

[0053] 1) Collect m guided wave array signals of P guided wave monitoring channels included in the piezoelectric sensor network when the structure is in a healthy state and under the influence of the service dynamic condition; calculate m damage factors of each monitoring channel as the health observation vector, and establish a guided wave - hidden Markov model to characterize the uncertain distribution of the health observation vector affected by the service dynamic condition;

[0054] 2) Collect the guided wave array signals when the structure is in an unknown fatigue crack damage state and under the influence of the service dynamic condition, and calculate the damage factors of each channel as the damage observation vector;

[0055] 3) Repeat step 2) multiple times, merge the multiple damage observation vectors obtained under the influence of the structure damage state and the service dynamic condition, and form an observation vector sequence;

[0056] 4) inputting the observation vector sequence obtained in step 3) into the guided wave hidden Markov model established in step 1) to calculate the probabilistic damage factor as the channel feature;

[0057] 5) The entire monitoring area formed by the piezoelectric sensor network is divided into multiple pixel points, the array signal fusion parameters are set, the damage probability of each pixel point under each monitoring channel in the array is calculated, and the damage probabilities of all channels are accumulated and fused to obtain the imaging pixel value, and the imaging pixel values of all pixels are combined to obtain the pixel value matrix of fatigue crack damage;

[0058] 6) Setting a pixel value threshold, searching for pixel points whose pixel values are higher than / equal to the threshold in the pixel value matrix of fatigue crack damage, and setting the pixel values of pixel points whose pixel values are lower than the threshold to zero;

[0059] 7) The pixel values of the pixels searched in step 6) are used as weights, and the weighted summation of the respective coordinates is performed to obtain the final fatigue crack damage location result.

[0060] The object of this embodiment is an aviation ear piece connection structure, the material is 2024-T4 aluminum alloy, the structure width is 180mm, the length is 343.5mm, and the thickness is 5mm. Figure 2a and Figure 2b As shown, a bolt hole with a diameter of 48 mm is opened in the middle for installing bolts. A circular piezoelectric sensor array is arranged on the surface of the structure, including a total of 22 piezoelectric sensors, marked as PZT1 to PZT22, as shown in Figure 3a As shown in Figure 2, a total of 462 guided wave excitation-sensing monitoring channels are formed, such as Figure 3b The fatigue testing machine is used to perform fatigue tensile loading on the ear structure, so that fatigue crack damage is initiated and extended at the stress concentration area of the ear hole edge, as shown in Figure 2. Figure 4 shown.

[0061] The specific implementation method for locating fatigue crack damage on the ear structure is as follows:

[0062] 1. Collect array multi-channel health signals under service dynamic load conditions and construct a health observation vector sample set;

[0063] When the structure is in a healthy state, the guided wave signals of all monitoring channels of the array are collected during fatigue loading, and a total of m = 40 groups of healthy signals are collected; for each monitoring channel, the first group of signals is selected as the reference signal, and the normalized cross-correlation moment damage factor DI1 and the root mean square deviation damage factor DI2 are calculated to form a two-dimensional observation vector o = [DI1, DI2], and then a healthy observation vector sample set O = {o1, o2, o3, …, o 40};like Figure 5Shown is an example of the health observation vector sample set of one of the monitoring channels PZT1 - PZT16.

[0064] 2. Construct a guided wave - hidden Markov model for each monitoring channel of the array;

[0065] Based on the health observation vector sample set of each monitoring channel of the array, uniformly initialize the model parameters a ij and π i of the guided wave - hidden Markov model, where i = 1, 2, j = 1, 2; Initialize the observation probability distribution b i (o t ) using the uniform initialization Gaussian mixture model clustering method. According to the observation vector distribution, set the number of Gaussian components in the Gaussian mixture model to 2; Train the model parameters of the guided wave - hidden Markov model through the Baum - Welch algorithm. Taking the monitoring channels PZT1 - PZT16 as an example, after training, the state transition probability a 11 = 0.33, a 12 = 0.67, a 21 = 0.67, a 22 = 0.33, the initial state probability π1 = 1.00, π2 = 0.00, and the observation probability distribution characterized by the Gaussian mixture model is as Figure 6 shown.

[0066] 3. Calculate the probability damage factor for each monitoring channel of the array as the channel damage feature;

[0067] Set the length k of the observation vector sequence to 30. When the structure is in an unknown fatigue crack damage state, also collect the guided wave signals of all monitoring channels of the array 30 times during the fatigue loading process. Calculate the normalized cross - correlation moment damage factor and the root - mean - square deviation damage factor based on the healthy reference signal to form the observation vector. Thus, the 30 obtained observation vectors constitute the observation vector sequence O t = {o t-39 , o t-38 , …, o t-1 , o t}. Set the smoothing length l to 10, substitute O t into the guided wave - hidden Markov model to calculate the likelihood probability, and after smoothing, obtain the probability damage factor PDI p of each monitoring channel as the channel damage feature, so as to reliably characterize the influence degree of each channel affected by fatigue crack propagation under the working condition of dynamic fatigue load. The specific calculation expression of PDI p is as follows:

[0068]

[0069] 4. Array signal fusion to determine the fatigue crack damage location;

[0070] Set the size parameter β of the array signal fusion to 1.05. The lengths of the monitoring area formed by the circular sensor array in the X and Y directions are both 80 mm. Therefore, each is equally divided into 80 parts, that is, the length of each pixel is 1 mm. Calculate the damage probability of each pixel under each monitoring channel in the array, and accumulate the damage probabilities of all monitoring channels to achieve focus enhancement, obtaining a pixel value matrix, as Figure 7 shown. The removed part of the circle in the figure is the central bolt hole of the lug structure. Set the pixel value threshold according to 0.9 times the maximum pixel value in the pixel value matrix, set the pixel values of the pixels below the threshold to zero, and search for the coordinates of all pixels higher than / equal to the threshold. After weighted fusion of the searched coordinates, the fatigue crack damage location result is obtained. The positioning error from the actual fatigue crack position is about 2.1 mm, indicating that this method can accurately locate the fatigue crack damage of the structure under the service dynamic load condition.

[0071] There are many specific application ways of the present invention. The above is only the preferred implementation mode of the present invention. It should be pointed out that for those of ordinary skill in the art in this technical field, without departing from the principle of the present invention, several improvements can still be made, and these improvements should also be regarded as the protection scope of the present invention.

Claims

1. A fatigue crack damage location method based on array signal fusion guided wave - hidden Markov model, characterized in that, The steps are as follows: 1) Collect m guided wave array signals of P guided wave monitoring channels included in the piezoelectric sensor network under the influence of the healthy state and service dynamic conditions of the structure; calculate m damage factors of each monitoring channel as the healthy observation vector, and establish a guided wave-hidden Markov model to characterize the uncertain distribution of the healthy observation vector affected by the service dynamic conditions; 2) Collect the guided wave array signals of the structure under the influence of the unknown fatigue crack damage state and service dynamic conditions, and calculate the damage factors of each channel as the damage observation vector; 3) Repeat step 2) multiple times, and combine multiple damage observation vectors obtained under the influence of the structure damage state and service dynamic conditions to form an observation vector sequence; 4) Input the observation vector sequence obtained in step 3) into the guided wave-hidden Markov model established in step 1), and calculate the probability damage factor as the channel feature; 5) Divide the entire monitoring area composed of the piezoelectric sensor network into multiple pixel points, set the array signal fusion parameters, calculate the damage probability of each pixel point under each monitoring channel in the array, and accumulate and fuse the damage probabilities of all channels to obtain the imaging pixel value. Combine the imaging pixel values of all pixel points to obtain the pixel value matrix of the fatigue crack damage; 6) Set the pixel value threshold, search for the pixel points in the pixel value matrix of the fatigue crack damage whose pixel values are higher than / equal to the threshold, and set the pixel values of the pixel points lower than the threshold to zero; 7) Use the pixel values of the pixel points searched in step 6) as weights, and perform weighted summation on their respective coordinates to obtain the final fatigue crack damage location result.

2. The fatigue crack damage location method based on array signal fusion guided wave - hidden Markov model according to claim 1, wherein, The damage factors calculated in step 1) include the normalized cross-correlation moment damage factor DI1 and the root mean square deviation damage factor DI2, and the calculation formulas are as follows: where τ is the delay factor, r xy (τ) is the cross-correlation function of the reference and the monitoring signals, r xx (τ) is the autocorrelation function of the reference signal, T is the signal length, H(t) is the reference signal, D(t) is the monitoring signal, and n is the order of the statistical moment.

3. The fatigue crack damage location method based on array signal fusion guided wave - hidden Markov model according to claim 2, characterized in that, The steps for establishing the guided wave-hidden Markov model in step 1) are as follows: 11) Initialize the parameters λ of the hidden Markov model, including setting the number of hidden states to 2, forming the set of hidden states S = {S1, S2}, where both S1 and S2 represent the structural health state; initialize the state transition probability a ij = 0.5, where i = 1, 2, j = 1, 2; the initial state probability π i = 0.5, and initialize the observation probability distribution b using the uniform initialization Gaussian mixture model clustering method i (o t ), where o t represents the guided wave observation vector at time t; 12) Optimize the parameters λ of the hidden Markov model through the Baum-Welch algorithm to obtain the trained model parameters a ij , π i and b i (o t ).

4. The fatigue crack damage location method based on array signal fusion guided wave - hidden Markov model according to claim 1, characterized in that, The observation vector sequence established in step 3) is composed of multiple continuous observation vectors, and its expression is as follows: O t = {o t-k+1 , o t-k+2 , …, o t} where k is the number of observation vectors that make up the observation vector sequence O t ; o t-k+1 is the guided wave observation vector at time t - k + 1, and o t-k+2 is the guided wave observation vector at time t - k + 2.

5. The fatigue crack damage location method based on array signal fusion guided wave - hidden Markov model according to claim 1, characterized in that, Step 5) specifically includes: 51) Define the damage feature of the p-th monitoring channel in the piezoelectric sensor network as the probability damage factor calculated by its guided wave-hidden Markov model, denoted as PDI p ; 52) Divide the entire monitoring area composed of the piezoelectric sensor network into multiple pixel points, where the X and Y directions are equally divided into c parts respectively, and the position of each pixel point is represented by its coordinates (x, y) in the monitoring area; 53) Set the array signal fusion parameters and calculate the damage probability P p (x,y) of each pixel point under each monitoring channel in the array respectively. The expression is as follows:​ where β is the size parameter for array signal fusion, controlling the size of the influence region of the probability damage factor on the direct path of each channel of the array, and R p (x,y) is the ratio of the sum of the distances from the pixel point (x,y) to the excitation element and the sensing element of the p-th monitoring channel of the array to the length of the monitoring channel, and the expression is as follows: wherein, (x pa , y pa ) and (x ps , y ps ) respectively represent the coordinates of the excitation element and the coordinates of the sensing element on the p-th monitoring channel of the array; 54) Accumulate the damage probabilities of all monitoring channels in the array to obtain the final damage probability of each pixel point as the imaging pixel value P(x, y), and the expression is as follows: 55) Combine the imaging pixel values of all pixel points to obtain the pixel value matrix of the fatigue crack damage.

6. The fatigue crack damage location method based on array signal fusion guided wave - hidden Markov model according to claim 1, wherein Step 6) specifically includes: According to the pixel value matrix of fatigue crack damage obtained in step 5), set the pixel value threshold to 0.9 times the maximum pixel value, and search for the coordinates of all pixel points whose pixel values are higher than / equal to the pixel value threshold to form a set Z = {(x1, y1), (x2, y2), …, (x Q , y Q )}, where Q is the number of pixel points searched, x Q is the coordinate of the Qth pixel point searched in the X direction, and y Q is the coordinate of the Qth pixel point searched in the Y direction.

7. The fatigue crack damage location method based on array signal fusion guided wave - hidden Markov model according to claim 1, characterized in that, The expression of the fatigue crack damage location result in step 7) is as follows: where x diag is the X-direction coordinate of the fatigue crack damage location result, and y diag is the Y-direction coordinate of the fatigue crack damage location result, x q is the X-direction coordinate of the q-th pixel point searched, and y q is the Y-direction coordinate of the q-th pixel point searched.

Images