Structural damage early warning method based on guided wave hybrid probability characteristics and adaptive threshold

Abstract

The present invention discloses a structural damage early warning method based on waveguide mixing probability characteristics and adaptive thresholds, which belongs to the field of structural health monitoring, including estimating the probability distribution of waveguide feature samples of the structure in healthy and monitored states through an adaptive probability mixing modeling algorithm, obtaining the waveguide mixing probability characteristics by calculating the difference in probability distribution of the waveguide feature samples in the healthy and monitored states of the structure, adaptively determining the alarm threshold based on the waveguide mixing probability characteristics obtained in the healthy state of the structure, sliding averaging the waveguide mixing probability characteristics during the monitoring process, and performing damage early warning when the waveguide mixing probability characteristics after sliding averaging exceed the alarm threshold. The structural damage early warning method based on waveguide mixing probability characteristics and adaptive thresholds of the present invention is simple and efficient to implement, and can achieve reliable and adaptive structural damage early warning under the influence of service environment factors.

Description

Technical Field

[0001] The present invention relates to the technical field of structural health monitoring, and in particular to a structural damage early warning method based on guided wave mixed probability characteristics and adaptive thresholds. Background Art

[0002] Structural health monitoring technology can monitor the health status of the structure online, and then evaluate and predict the damage and remaining life of the structure. This condition-based maintenance method can effectively reduce the maintenance cost of the structure and the risk of accidents caused by damage to key structures while ensuring the safety of the structure. Among the structural health monitoring technologies, the structural health monitoring technology based on guided waves has the advantages of long monitoring distance, large monitoring area, and sensitivity to small damage. In recent years, it has attracted much attention in the fields of structural operation and maintenance such as aerospace, transportation, etc.

[0003] As the most basic and important component of structural health monitoring, structural damage warning is a systematic and automated process to identify whether damage has occurred. Structural health monitoring technology based on guided wave mainly identifies the occurrence and expansion of structural damage based on changes in guided wave characteristics such as waveguide signal amplitude and phase. In general, a fixed threshold of guided wave characteristics can be set based on empirical knowledge for damage alarm. However, in the environment where the structure is located, complex service environment factors such as randomly changing temperature, load, vibration, etc. will also cause changes in guided wave characteristics. This change is equivalent to the change in guided wave characteristics caused by structural damage, especially early damage, and it is difficult to distinguish. Therefore, for damage warning of serving structures, overcoming the uncertainty of environmental factors is a major difficulty. In addition, artificially set fixed thresholds will also introduce damage warning errors. Therefore, how to achieve reliable and adaptive structural damage warning is a major challenge faced by structural guided wave monitoring technology in practical applications.

[0004] Since the influence of service environment factors on guided wave characteristics shows strong uncertainty, and factors such as load, vibration, boundary conditions are difficult to measure directly, conventional environmental factor compensation methods are difficult to apply to the health monitoring of actual service structures. Considering that the probabilistic modeling method can estimate the random distribution of guided wave characteristics under the influence of service environment by fitting, domestic and foreign scholars have applied the probabilistic modeling method to damage monitoring research in recent years, providing a feasible way to solve the reliable monitoring of structural damage under the influence of uncertainty. These studies mainly use the probability distribution change of guided wave characteristics obtained on the service structure as an indicator for evaluating the damage state of the structure. Although this indicator suppresses the uncertainty of environmental factors to a certain extent and is sensitive to structural damage, the reliability of damage warning will still be affected by the accuracy of probability distribution estimation.

[0005] In order to meet the needs of long-term online and unattended monitoring, it is also critical to set the damage alarm threshold. Based on the training data obtained when the service structure is in a healthy state, the control chart can be used to determine the threshold. The control chart is an effective process control method, which mainly monitors the process through process quality characteristics and control limits. In the field of structural damage warning, structural health status monitoring is regarded as a process control, and the damage index is used as a quality characteristic reflecting the health status of the structure. The damage alarm threshold is the control limit, and then the damage warning is performed. Although the alarm threshold can be set through the control chart, the abnormal situation of the damage index under the healthy state of the structure may exceed the control limit, thereby causing a false alarm of damage.

[0006] In summary, although the waveguide monitoring method based on probability modeling can separate the features that are sensitive to damage from the influence of uncertain factors, its reliability is still affected by the accuracy of the probability distribution estimation of the waveguide characteristics, the setting of the alarm threshold, and the abnormal values ​​of the damage indicators. Summary of the invention

[0007] To solve the above problems, the present invention aims to provide a structural damage warning method based on waveguide hybrid probability characteristics and adaptive thresholds, which can effectively improve the adaptability and reliability of structural damage warning under the influence of service environment.

[0008] To achieve the above object, the technical solution of the present invention is:

[0009] The structural damage early warning method based on guided wave mixed probability characteristics and adaptive thresholds includes:

[0010] Step S11, when the structure is in a healthy state, obtain M guided wave feature samples s i , i=1,2,…,M;

[0011] Step S12, taking the M waveguide feature samples as training samples, randomly selecting N training samples from them to form a benchmark sample set S0, where 0 <N≤M / 2;

[0012] Step S13, estimating a reference probability distribution Φ0 of the reference sample set S0 by using an adaptive probabilistic hybrid modeling algorithm;

[0013] Step S14, sequentially substitute the M training samples into the updated training sample set to obtain M training sample sets S'1, S'2, ... S' M ;

[0014] Step S15, estimating the M training sample sets S'1, S'2, ... S' by the adaptive probabilistic hybrid modeling algorithm M The corresponding training probability distributions are Φ'1, Φ'2, ... Φ' M ;

[0015] Step S16, according to the calculation method of probability distribution difference, calculate the M training probability distributions Φ'1, Φ'2, ... Φ' respectively. M The probability distribution difference between the reference probability distribution Φ0 is the waveguide mixing probability feature, and the corresponding M training waveguide mixing probability features PDD'1, PDD'2, ...PDD' M ;

[0016] Step S17, according to the training waveguide mixing probability features PDD'1, PDD'2, ... PDD' M , through the formula:

[0017]

[0018] Adaptively calculate the alarm threshold λ, where u is the M training waveguide mixed probability features PDD'1, PDD'2, ...PDD' M The mean of .

[0019] The structural damage early warning method based on the waveguide mixed probability characteristics and adaptive thresholds also includes:

[0020] Step S21, in the monitoring state, obtain the monitoring waveguide characteristic sample s T , where T represents the Tth monitoring;

[0021] Step S22, each time a monitoring waveguide characteristic sample s is obtained T , the monitoring guided wave characteristic sample s T Substitute and update the monitoring sample set S of the last monitoring T-1 Get the current monitoring sample set S T ;

[0022] Step S23, obtaining the current monitoring sample set S through the adaptive probabilistic hybrid modeling algorithm T The corresponding monitoring probability distribution Φ T ;

[0023] Step S24, calculating the monitoring probability distribution Φ according to the calculation method of the probability distribution difference T The probability distribution difference between the reference probability distribution Φ0 is the waveguide mixing probability characteristic, and the monitoring waveguide mixing probability characteristic PDD is obtained. T ;

[0024] Step S25, calculate the control variable MA_PDD monitored for the Tth time T , calculated as follows:

[0025]

[0026] Where W is the sliding average window length;

[0027] Step S26: The T-th monitored control variable MA_PDD T Compared with the alarm threshold λ, if MA_PDD T >λ, a damage warning is issued; otherwise, no warning is given.

[0028] The above step S11 includes:

[0029] Step S111, under random load, collect the guided wave signal of the structure in a healthy state M times, each of the guided wave signals is recorded as f i (t), i = 1, 2, ..., M;

[0030] Step S112, extracting signal features and obtaining M waveguide feature samples s i , i=1,2,…,M.

[0031] In a specific embodiment, the calculation method of the first signal feature is as follows:

[0032]

[0033]

[0034]

[0035] Among them, SF 1,i is the first signal feature, t1 and t2 are the start time and end time of the intercepted signal segment, τ is the time lag parameter, r si is the waveguide signal f i (t) and the reference signal f s The cross-correlation between (t), r ss (τ) is the reference signal f s (t) Mutual correlation with itself.

[0036] The calculation method of the second signal feature is as follows:

[0037]

[0038] Among them, SF 2,i is the second signal feature, t1 and t2 are the start time and end time of the intercepted signal segment respectively, f s (t) is the reference signal, f i (t) is the waveguide signal.

[0039] The above waveguide feature samples i The expression is s i =[SF 1,i,SF 2,i ] T , where superscript T is the matrix transpose symbol, i=1,2,…,M.

[0040] The above-mentioned adaptive probabilistic mixture modeling algorithm includes,

[0041] Step S31, based on the sample set S, estimate each sample s by Gaussian kernel density i The density ρ i , the calculation formula is as follows:

[0042]

[0043] Where n is the sample s in the sample set S. i The number of samples i The dimension is D, d ij For sample s i With s j The Euclidean distance between c is the cutoff distance, the cutoff distance d c The expression is as follows:

[0044]

[0045] Among them, d Z (s i ) is sample s i With distance s i The distance between the closest Zth samples, Represents the largest integer not exceeding x;

[0046] Step S32, calculating each sample s i The minimum distance δ i , the calculation formula is as follows:

[0047]

[0048] Step S33, calculating each sample s i The density ρ i The minimum distance δ i The product of γ i , the calculation formula is as follows:

[0049] γ i =ρ i ×δ i ;

[0050] Step S34, calculating the product γ i The threshold value γ min , the calculation formula is as follows:

[0051] γmin =EX(ρ i )×d c ,

[0052] Where EX(ρ) is the density ρ i The mean of

[0053] Step S35, select the samples that satisfy the following inequality as the cluster center, and cluster the remaining samples in the sample set S into the cluster to which the cluster center closest to it belongs based on the nearest neighbor principle, to obtain the initial clustering result:

[0054] γ i >γ min &δ i >d c ;

[0055] Step S36, if there is a class in the initial clustering result whose number of samples is less than the dimension D, then remove the class and the corresponding samples to obtain the final clustering result, and the total number of samples is n';

[0056] Step S37, based on the final clustering result, calculating the mean and covariance matrix of each class, and calculating the ratio of the number of samples contained in each class to the total number of samples n' to obtain the weight corresponding to each class;

[0057] Step S38, using the mean, covariance matrix and weight of each class obtained in step S37 as initialization parameters, using the expectation maximization algorithm to estimate the probability distribution Φ of the sample set S, the probability density expression of the probability distribution Φ is as follows:

[0058]

[0059] Wherein, K is the number of sub-distributions in the probability distribution Φ, k=1,2,…,K; w k , μ k and Σ k are the weight, mean and covariance matrix of the kth sub-distribution, |·| is the determinant of the matrix, and T represents the matrix transpose symbol.

[0060] The calculation method of the above probability distribution difference is as follows:

[0061] Step S41, record the two compared probability distributions as probability distributions Φ B and the probability distribution Φ M , and the corresponding signal feature sample sets are signal feature sample sets S B And signal feature sample set S M , the signal feature sample set S B and the signal feature sample set S M The union of is the union SU , in the union S U Linear sampling is performed in the sample space covered to obtain the sampling feature sample set S MC , the sampling feature sample set S MC ={s1,s2,…,s R}, R is the sampling feature sample set S MC The sample size of

[0062] Step S42, respectively calculating the sampling feature sample set S MC In the probability distribution Φ B and the probability distribution Φ M The probability density Φ in B (S MC ) and probability density Φ M (S MC );

[0063] Step S43, calculating the probability density Φ B (S MC ) and the probability density Φ M (S MC ), the probability distribution difference PDD between them is calculated as follows:

[0064]

[0065] Here, ||·||2 is the 2-norm of the vector.

[0066] In a specific embodiment, the value of W is N / 2.

[0067] In a specific embodiment, the reference signal f s (t) Take the average signal of the waveguide signal f1(t) and the waveguide signal f2(t).

[0068] Beneficial effect: The structural damage warning method based on waveguide mixed probability characteristics and adaptive thresholds of the present invention is simple and efficient to implement, can adaptively and efficiently fit the random distribution of waveguide characteristics under service environment factors; can adaptively set damage warning thresholds; can achieve reliable and adaptive structural damage warning under the influence of service environment factors.

[0069] In order to make the above features and advantages of the invention more obvious and easy to understand, embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 Schematic diagram of the monitored structure and the position of the piezoelectric sensor in a specific embodiment of the present invention.

[0071] Figure 2The present invention is a flow chart of a structural damage early warning method based on waveguide mixing probability characteristics and adaptive thresholds.

[0072] Figure 3 Schematic diagram of the benchmark sample set S0 and the benchmark probability distribution Φ0.

[0073] Figure 4 For 40 training waveguide mixed probability features PDD'1, PDD'2, ...PDD' 40 Schematic diagram of .

[0074] Figure 5 Schematic diagram of the distribution of guided wave feature samples monitored at different crack lengths.

[0075] Figure 6 Schematic diagram of the changes in the monitoring guided wave mixing probability characteristics obtained from 320 monitorings.

[0076] Figure 7 Schematic diagram of the control variables and damage warning results obtained from 320 monitorings.

[0077] In the drawings, like reference numerals refer to the same drawing elements. DETAILED DESCRIPTION

[0078] In order to make the purpose and technical solution of the embodiment of the present invention clearer, the technical solution of the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings of the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all of the embodiments. Based on the described embodiment of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0079] In this specific embodiment, random load is used as a service environment factor, the ear piece is used as the monitored structure, and the cracks on the hole edge of the ear piece are regarded as damage. Combined with the guided wave damage monitoring technology, the implementation process of the method of the present invention is specifically explained by taking the early warning of the initiation of cracks on the hole edge of the ear piece structure under the service environment factor of random load.

[0080] Figure 1 FIG. 1 is a schematic diagram of the monitored structure and the position of the piezoelectric sensor in a specific embodiment of the present invention. Figure 1As shown, the monitored structure 1 is a rectangular ear piece structure with a width of 100 mm and a length of 300 mm. There is a simulated damage on the monitored structure 1, and the simulated damage is an ear hole 11 with a diameter of 30 mm. The ear hole 11 has a crack 12. A piezoelectric sensor PZT1 and a piezoelectric sensor PZT2 are arranged next to the ear hole 11 for monitoring the crack 12. The piezoelectric sensor PZT1 is used as an excitation element of the waveguide signal, and the piezoelectric sensor PZT2 is used as a guide signal. Response element, the expansion direction of the crack 12 is vertically downward, the distance between the center of the piezoelectric sensor PZT1 and the bottom of the monitored structure 1 is 25 mm, the distance between the center of the piezoelectric sensor PZT1 and the vertical center line of the monitored structure 1 is 75 mm, the piezoelectric sensor PZT2 is kept horizontal with the piezoelectric sensor PZT1 and is located on the right side of the piezoelectric sensor PZT1, and the center of the piezoelectric sensor PZT2 is 150 mm away from the center of the piezoelectric sensor PZT1.

[0081] Figure 2 The flowchart of the structural damage early warning method based on the waveguide mixed probability characteristics and adaptive threshold of the present invention is shown, which specifically includes:

[0082] Step S11, obtaining a sample of guided wave characteristics in a healthy state of the structure.

[0083] The step S11 specifically includes:

[0084] Step S111, under random load, collect the guided wave signal of the structure in a healthy state 40 times, each guided wave signal is recorded as f i (t), i = 1, 2, ..., 40;

[0085] Step S112, extracting signal features and forming a waveguide feature sample. In this specific embodiment, two signal features are extracted to form a two-dimensional waveguide feature sample, that is, D=2. i (t) The first signal feature SF extracted 1,i The calculation method is as follows,

[0086]

[0087]

[0088]

[0089] Among them, t1 and t2 are the start time and end time of the intercepted signal segment, τ is the time lag parameter, r si is the waveguide signal f i (t) and the reference signal f sThe cross-correlation between (t), r ss (τ) is the reference signal f s (t) is cross-correlated with itself. In this specific embodiment, the reference signal f s (t) Take the average signal of f1(t) and f2(t).

[0090] Based on the waveguide signal f i (t) The second signal feature SF extracted 2,i The calculation method is as follows,

[0091]

[0092] Wherein, t1 and t2 are the start time and end time of the intercepted signal segment respectively.

[0093] The two signal features extracted by the above calculation method constitute the waveguide signal f i (t) of the guided wave characteristic sample s i , that is, s i =[SF 1,i ,SF 2,i ] T , where superscript T is the matrix transpose symbol, i = 1, 2, …, 40.

[0094] Step S12, constructing a reference sample set S0, more specifically, based on the waveguide signals f1(t) to f 40 (t), 40 waveguide feature samples s'1, s'2, ..., s' can be extracted 40 As training samples, in this specific embodiment, the number of training samples M is 40, and 20 samples are randomly selected from the 40 training samples to form a benchmark sample set S0, and the sample capacity N of the benchmark sample set S0 is 20.

[0095] Step S13, estimating the baseline probability distribution Φ0 of the baseline sample set S0 by using the adaptive probabilistic hybrid modeling algorithm in step S3.

[0096] The step S3 more specifically includes:

[0097] Step S31, based on the sample set S, estimate each sample s by Gaussian kernel density i The density ρ i , the calculation formula is as follows:

[0098]

[0099] Where n is the sample s in the sample set S i The number of samples, a sample s i The dimension is D, d ij For sample si With s j The Euclidean distance between c is the cutoff distance, the cutoff distance d c The expression is as follows:

[0100]

[0101] Among them, d Z (s i ) is sample s i With distance s i The distance between the closest Zth samples, Represents the largest integer not exceeding x;

[0102] Step S32, calculate each sample s i The minimum distance δ i , the calculation formula is as follows:

[0103]

[0104] Step S33, calculate each sample s i The density ρ i With the minimum distance δ i The product of γ i , the calculation formula is as follows:

[0105] γ i =ρ i ×δ i ;

[0106] Step S34, calculate the product γ i The threshold value γ min , the calculation formula is as follows:

[0107] γ min =EX(ρ i )×d c ,

[0108] Where EX(ρ) is the density ρ i The mean of

[0109] Step S35, select the samples that satisfy the following inequality as the cluster center, and cluster the remaining samples in the sample set S into the cluster to which the cluster center closest to it belongs based on the nearest neighbor principle, to obtain the initial clustering result:

[0110] γ i >γ min &δ i >d c ;

[0111] Step S36, if there is a class in the initial clustering result whose number of samples is less than the sample dimension D, then remove the class and the corresponding samples to obtain the final clustering result, and the total number of samples is n';

[0112] Step S37, based on the final clustering result, the mean and covariance matrix of each class are calculated, and at the same time, the ratio of the number of samples contained in each class to the total number of samples n' is calculated to obtain the weight corresponding to each class;

[0113] Step S38, using the mean, covariance matrix and weight of each class obtained in step S37 as initialization parameters, using the expectation maximization algorithm to estimate the probability distribution Φ of the sample set S, the probability density expression of the probability distribution Φ is as follows:

[0114]

[0115] Wherein, K is the number of sub-distributions in the probability distribution Φ, k=1,2,…,K; w k , μ k and Σ k are the weight, mean and covariance matrix of the kth sub-distribution, |·| is the determinant of the matrix, and T represents the matrix transpose symbol.

[0116] In the step S13, the sample set S in the step S3 is a reference sample set S0. Figure 3 is a schematic diagram of the benchmark sample set S0 and the benchmark probability distribution Φ0, as shown Figure 3 As shown, the horizontal axis is the first signal feature SF1, the vertical axis is the second signal feature SF2, the dots represent the waveguide feature samples, and the circles represent the sub-distribution of the reference probability distribution Φ0.

[0117] Step S14, forming a training sample set. More specifically, the 40 training samples are sequentially substituted into the updated training sample set to obtain 40 training sample sets S'1, S'2, ... S' 40 .

[0118] The following takes the training sample set S'1 as an example to illustrate the substitution and update method of the training sample set: the benchmark sample set S0 is the initial training sample set, when the benchmark sample set S0 = {s' 1’ ,s' 2’ ,…,s' 20’}, substitute the training sample s'1 into the benchmark sample set S0, and remove s' 1’ Get the training sample set S'1, S'1 = {s' 2’ ,s' 3’ ,…,s' 20’ ,s'1}. Similarly, the training sample set S'2={s' 3’ ,…,s'20’ ,s'1,s'2}, until the training sample set S' 40 ={s' 21 ,s' 22 ,…,s' 39 ,s' 40}.

[0119] Step S15, estimating 40 training sample sets S'1, S'2, ... S' by the adaptive probabilistic hybrid modeling algorithm in step S3 40 The corresponding training probability distributions are Φ'1, Φ'2, ... Φ' 40 .

[0120] Step S16, respectively calculate 40 training probability distributions Φ'1, Φ'2, ... Φ' 40 The probability distribution difference between the reference probability distribution Φ0 is the waveguide mixing probability feature, and 40 training waveguide mixing probability features PDD'1, PDD'2, ... PDD' are obtained. 40 .

[0121] Figure 4 For 40 training waveguide mixed probability features PDD'1, PDD'2, ...PDD' 40 Schematic diagram of the training probability distribution order 1 to 40, the horizontal axis is the training probability distribution order 1 to 40, and the vertical axis is the corresponding waveguide mixing probability feature.

[0122] Step S17, based on the training waveguide mixing probability features PDD'1, PDD'2, ... PDD' 40 , through the formula:

[0123]

[0124] Adaptively calculate the alarm threshold to obtain the alarm threshold λ, where M is 40 in this specific embodiment, and u is 40 waveguide mixing probability features PDD'1, PDD'2, ...PDD' 40 In this specific embodiment, the value of the alarm threshold λ is 0.48.

[0125] Step S21, constructing a waveguide feature sample under monitoring state. More specifically, under monitoring state, collecting the waveguide signal of the structure under random load. The waveguide signal collected by the Tth monitoring is recorded as f T (t), based on the waveguide signal f T (t) extracting the waveguide signal f according to the calculation method of the first signal characteristic and the second signal characteristic T (t) Corresponding waveguide feature SF 1,T and SF 2,T , construct monitoring guided wave feature samples s TIn this specific embodiment, a total of 320 monitorings are performed, so T=1, 2, ..., 320.

[0126] In a specific embodiment, the length of the crack 12 of the structure is observed by an electron microscope during the monitoring process to verify the damage warning effect of the method of the present invention. During the 80th, 139th, 200th, 260th, and 320th monitoring, the crack 12 on the surface of the structure expands to 1 mm, 2 mm, 3 mm, 4 mm, and 5 mm, respectively.

[0127] Figure 5 The distribution diagram of the guided wave feature samples monitored under different crack lengths is shown. The horizontal axis is the first signal feature, and the vertical axis is the second signal feature. The crack lengths are 0mm, 0-1mm, 1-2mm, 2-3mm, 3-4mm, and 4-5mm, respectively. Figure 5 It can be seen that at different lengths, the distribution of the guided wave feature samples overlaps, especially when there is no crack and the crack extension does not exceed 1mm, the distribution of the guided wave feature samples almost overlaps, so it is difficult to directly rely on the distribution of the guided wave feature samples for reliable damage warning.

[0128] Step S22, each time a monitoring waveguide characteristic sample s is obtained T , the guided wave characteristic sample s will be monitored T Substitute and update the monitoring sample set S of the last monitoring T-1 Get the current monitoring sample set S T , the initial monitoring sample set is the benchmark sample set S0.

[0129] The following takes the monitoring sample set S1 as an example to illustrate the substitution and update method of the monitoring sample set: the benchmark sample set S0 is the initial training sample set, when the benchmark sample set S0 = {s'1', s' 2’ ,…,s' 20’}, substitute the monitoring waveguide characteristic sample s1 into the reference sample set S0, remove s'1 in the reference sample set S0, and obtain the monitoring sample set S1, S1 = {s' 2’ ,s' 3’ ,…,s' 20’ ,s1}. By analogy, we can get the monitoring sample set S2={s' 3’ ,…,s' 20’ ,s1,s2}, etc.

[0130] Step S23, obtaining the current monitoring sample set S through the adaptive probabilistic hybrid modeling algorithm in step S3 T The corresponding monitoring probability distribution Φ T .

[0131] Step S24, calculate the monitoring probability distribution Φ TThe probability distribution difference between the reference probability distribution Φ0 is the waveguide mixing probability characteristic, and the monitoring waveguide mixing probability characteristic PDD is obtained. T .

[0132] Figure 6 Schematic diagram of the change of the monitoring guided wave mixing probability characteristics obtained from 320 monitorings, where the upper axis is the length of the structural crack observed by the electron microscope during the monitoring process, the lower axis is the number of monitorings, and the ordinate is the guided wave mixing probability characteristics. Figure 6 It can be seen that the waveguide mixing probability feature extracted by the present invention is more sensitive to the expansion of structural cracks, and when the crack expands to 2mm, the value of the waveguide mixing probability feature quickly expands to the maximum; when there is no crack, the waveguide mixing probability feature will also fluctuate randomly due to the influence of random loads, and this fluctuation increases the difficulty of reliable early warning of structural damage. In addition, Figure 6 The alarm threshold is also given. It can be seen that in the absence of crack damage, the guided wave mixing probability characteristics will also exceed the alarm threshold. Therefore, the reliability of directly using the guided wave mixing probability characteristics for damage warning needs to be improved.

[0133] Step S25, calculate the control variable MA_PDD monitored for the Tth time T , the calculation formula is as follows:

[0134]

[0135] Wherein, W is the sliding average window length. In this specific embodiment, W is 10.

[0136] Step S26, obtain the structural damage warning result, more specifically, control variable MA_PDD T Compared with the alarm threshold λ, when the control variable MA_PDD T When the alarm threshold is exceeded, a damage warning is issued.

[0137] The calculation method of the probability distribution difference between step S16 and step S24 is as follows:

[0138] Step S41, the probability distributions of the two comparisons are respectively recorded as Φ B and Φ M , and the corresponding signal feature sample sets are S B and S M , S B and S M The union of U , in the union S U Linear sampling is performed in the sample space covered to obtain the sampling feature sample set S MC , sampling feature sample set S MC ={s1,s2,…,sR}, R is S MC The sample size, in this embodiment, R is 10000;

[0139] Step S42, using the probability density expression in step S38, respectively calculate the sampling feature sample set S MC In the probability distribution Φ B and the probability distribution Φ M The probability density Φ in B (S MC ) and probability density Φ M (S MC );

[0140] Step S43, calculate two probability densities Φ B (S MC ) and Φ M (S MC ), the probability distribution difference PDD between them is calculated as follows:

[0141]

[0142] Here, ||·||2 is the 2-norm of the vector.

[0143] Figure 7 The schematic diagram of the control variables and damage warning results obtained from 320 monitorings, where the lower horizontal axis is the number of monitorings, the upper horizontal axis is the crack length, and the vertical axis is the control variable. Figure 7 It can be seen that compared with the waveguide mixing probability characteristics, the control variables in the present invention are less affected by random load environmental factors, and the results of structural damage warning are more accurate.

[0144] In summary, the structural damage early warning method based on the waveguide mixed probability characteristics and adaptive thresholds of the present invention realizes a reliable early warning of damage of a 1 mm crack in the structure under the influence of service environment factors.

[0145] Although the present invention has been disclosed as above by way of embodiments, it is not intended to limit the present invention. Any person having ordinary knowledge in the technical field may make some changes and modifications without departing from the spirit and scope of the present invention. Therefore, the protection scope of the present invention shall be determined by the scope of the attached patent application.

Claims

1. A structural damage early warning method based on waveguide mixing probability characteristics and adaptive thresholds, characterized in that: include, Step S11, when the structure is in a healthy state, obtain M guided wave feature samples s i , i=1,2,…,M; Step S12, taking the M waveguide feature samples as training samples, randomly selecting N training samples from them to form a benchmark sample set S0, where 0 <N≤M / 2; Step S13, estimating a reference probability distribution Φ0 of the reference sample set S0 by using an adaptive probabilistic hybrid modeling algorithm; Step S14: Substitute the M training samples into the updated training sample set in sequence to obtain M training sample sets S'1, S'2, ... S' M ; Step S15, estimating the M training sample sets S'1, S'2, ... S' by the adaptive probabilistic hybrid modeling algorithm M The corresponding training probability distributions are Φ'1, Φ'2, ... Φ' M ; Step S16, according to the calculation method of probability distribution difference, calculate the M training probability distributions Φ'1, Φ'2, ... Φ' respectively. M The probability distribution difference between the reference probability distribution Φ0 is the waveguide mixing probability feature, and the corresponding M training waveguide mixing probability features PDD'1, PDD'2, ...PDD' M ; Step S17, according to the training waveguide mixing probability features PDD'1, PDD'2, ... PDD' M , through the formula: Adaptively calculate the alarm threshold λ, where u is the M training waveguide mixed probability features PDD'1, PDD'2, ...PDD' M The mean of The adaptive probabilistic hybrid modeling algorithm includes: Step S31, based on the sample set S, estimate each sample s by Gaussian kernel density i The density ρ i , the calculation formula is as follows: Where n is the sample s in the sample set S. i The number of samples i The dimension is D, d ij For sample s i With s j The Euclidean distance between c is the cutoff distance, the cutoff distance d c The expression is as follows: Among them, d Z (s i ) is sample s i With distance s i The distance between the closest Zth samples, Represents the largest integer not exceeding x; Step S32, calculating each sample s i The minimum distance δ i , the calculation formula is as follows: Step S33, calculating each sample s i The density ρ i The minimum distance δ i The product of γ i , the calculation formula is as follows: c i =ρ i ×d i ; Step S34, calculating the product γ i The threshold value γ min , the calculation formula is as follows: c min =EX(ρ i )×d c , Where EX(ρ) is the density ρ i The mean of Step S35, select the samples that satisfy the following inequality as the cluster center, and cluster the remaining samples in the sample set S into the cluster to which the cluster center closest to it belongs based on the nearest neighbor principle, to obtain the initial clustering result: c i >c min &d i >d c ; Step S36, if there is a class in the initial clustering result whose number of samples is less than the dimension D, then remove the class and the corresponding samples to obtain the final clustering result, and the total number of samples is n'; Step S37, based on the final clustering result, calculating the mean and covariance matrix of each class, and calculating the ratio of the number of samples contained in each class to the total number of samples n' to obtain the weight corresponding to each class; Step S38, using the mean, covariance matrix and weight of each class obtained in step S37 as initialization parameters, using the expectation maximization algorithm to estimate the probability distribution Φ of the sample set S, the probability density expression of the probability distribution Φ is as follows: Wherein, K is the number of sub-distributions in the probability distribution Φ, k=1,2,…,K; w k , μ k and Σ k are the weight, mean and covariance matrix of the kth sub-distribution, respectively, |·| is the determinant of the matrix, and T represents the matrix transpose symbol; The calculation method of the probability distribution difference is as follows: Step S41, record the two compared probability distributions as probability distributions Φ B and the probability distribution Φ M , and the corresponding signal feature sample sets are signal feature sample sets S B And signal feature sample set S M , the signal feature sample set S B and the signal feature sample set S M The union of is the union S U , in the union S U Linear sampling is performed in the sample space covered to obtain the sampling feature sample set S MC , the sampling feature sample set S MC ={s1,s2,…,s R }, R is the sampling feature sample set S MC The sample size of Step S42, respectively calculating the sampling feature sample set S MC In the probability distribution Φ B and the probability distribution Φ M The probability density Φ in B (S MC ) and probability density Φ M (S MC ); Step S43, calculating the probability density Φ B (S MC ) and the probability density Φ M (S MC ), the probability distribution difference PDD between them is calculated as: Here, ||·||2 is the 2-norm of the vector.

2. The structural damage early warning method based on waveguide mixed probability characteristics and adaptive threshold as claimed in claim 1 is characterized in that: Also includes, Step S21, in the monitoring state, obtain the monitoring waveguide characteristic sample s T , where T represents the Tth monitoring; Step S22, each time a monitoring waveguide characteristic sample s is obtained T , the monitoring guided wave characteristic sample s T Substitute and update the monitoring sample set S of the last monitoring T-1 Get the current monitoring sample set S T ; Step S23, obtaining the current monitoring sample set S through the adaptive probabilistic hybrid modeling algorithm T The corresponding monitoring probability distribution Φ T ; Step S24, calculating the monitoring probability distribution Φ according to the calculation method of the probability distribution difference T The probability distribution difference between the reference probability distribution Φ0 is the waveguide mixing probability characteristic, and the monitoring waveguide mixing probability characteristic PDD is obtained. T ; Step S25, calculate the control variable MA_PDD monitored for the Tth time T , calculated as follows: Where W is the sliding average window length; Step S26: The T-th monitored control variable MA_PDD T Compared with the alarm threshold λ, if MA_PDD T >λ, a damage warning is issued; otherwise, no warning is given.

3. The structural damage early warning method based on waveguide mixed probability characteristics and adaptive threshold as claimed in claim 1 is characterized in that: The step S11 comprises: Step S111, under random load, collect the guided wave signal of the structure in a healthy state M times, each of the guided wave signals is recorded as f i (t), i = 1, 2, ..., M; Step S112, extracting signal features and obtaining M waveguide feature samples s i , i=1,2,…,M.

4. The structural damage early warning method based on waveguide mixed probability characteristics and adaptive threshold as claimed in claim 3 is characterized in that: The signal characteristics are calculated as follows: Among them, SF 1,i is the first signal feature, t1 and t2 are the start time and end time of the intercepted signal segment, τ is the time lag parameter, r si is the waveguide signal f i (t) and the reference signal f s The cross-correlation between (t), r ss (τ) is the reference signal f s (t) Mutual correlation with itself.

5. The structural damage early warning method based on waveguide mixed probability characteristics and adaptive threshold as claimed in claim 4 is characterized in that: The signal characteristics are calculated as follows: Among them, SF 2,i is the second signal feature, t1 and t2 are the start time and end time of the intercepted signal segment respectively, f s (t) is the reference signal, f i (t) is the waveguide signal.

6. The structural damage early warning method based on waveguide mixing probability characteristics and adaptive thresholds as claimed in claim 5, characterized in that: The waveguide characteristic sample s i The expression is s i =[SF 1,i ,SF 2,i ] T , where superscript T is the matrix transpose symbol, i=1,2,…,M.

7. The structural damage early warning method based on waveguide mixed probability characteristics and adaptive threshold as claimed in claim 2, characterized in that: The value of W is N / 2.

8. The structural damage early warning method based on waveguide mixing probability characteristics and adaptive thresholds according to any one of claims 4 to 5, characterized in that: The reference signal f s (t) Take the average signal of the waveguide signal f1(t) and the waveguide signal f2(t).

Images