Bayesian and Kriging Model-Based Inversion Method for Metal Corrosion Damage Evaluation

Abstract

The present invention provides a metal corrosion damage evaluation inversion method based on Bayesian and Kriging models, which includes: collecting ultrasonic guided wave signals in experiments to construct an ultrasonic guided wave signal data set; establishing a two-dimensional simulation model for non-damaged guided wave propagation; calibrating the mechanical properties of the simulation model, and improving the fit between the two-dimensional simulation model for non-damaged guided wave propagation and the experiment through the Bayesian optimization algorithm; introducing the equivalent thickness as an identification index to identify the corrosion damage path; solving the width and depth information of corrosion damage on different paths, and introducing the Kriging model during the solving process to improve the fineness of corrosion damage information; using the minimum distance method to locate the damage center and obtaining the boundary points of the damage by inverse-solving the Kriging model; fitting all boundary points by the least squares method.

Description

Technical Field

[0001] The present invention relates to non-destructive testing technology based on ultrasonic guided waves, and particularly to a model calibration method based on Bayesian optimization and an inversion method for quantitative evaluation of metal corrosion damage based on Kriging model. Background Art

[0002] The rapid development of modern industry has brought about widely used large metal structures, along with complex external environments. When metal structures are in a humid environment or in contact with water, steam, and other corrosive media, corrosion will occur. As one of the main reasons for the decline in structural strength, the degree of material corrosion will gradually deepen with the continuation of working time, posing a great hidden danger to structural safety. Corrosion damage not only damages the materials on the surface of the structure, resulting in the destruction of the original design shape of the structure, but also changes the mechanical and other physical properties of the structure, seriously affecting the original shape, strength, and other design indicators of the structure, significantly reducing the safety performance of the structure. And when the corrosion damage reaches a certain level, it may greatly endanger public social safety. Therefore, for common metal structures, especially components that have reached a certain service time, the aging of the structure and the accompanying corrosion damage during the service process are common, and the corrosion damage is a completely irreversible process after it occurs. Therefore, the detection of corrosion damage during the maintenance of key structures is a very necessary and important process. Being able to take effective measures to detect the corrosion damage on the structure and evaluate the expansion trend of corrosion can guide maintenance and support personnel to formulate repair and maintenance measures for the damage as early as possible, and minimize the destructive effect of corrosion on the structural strength and remaining service life.

[0003] Guided wave non-destructive testing (GWNDT) analyzes the signal information in the propagation waveform, such as the propagation characteristics of waves, through excitation / sensing elements arranged on the structure, and analyzes the obtained structural state information by combining signal processing methods. Piezoelectric transducers are mostly selected as the excitation / sensing elements, and the transducers are pasted on the surface of the device to be tested. By applying an alternating voltage to the positive and negative electrodes of the transducer, according to the piezoelectric effect, the piezoelectric material deforms and generates ultrasonic guided waves in the structure. The waveform passing through the damage area carries components containing damage information. By extracting and analyzing the components, and finally using a data fusion algorithm to reconstruct the damage.

[0004] Common damage algorithms usually propose damage indices and map them to damage information. However, common damage algorithms are difficult to analyze complex signals. At the same time, due to the variability of the experimental process, variables such as temperature and material parameters affect the accuracy of the detection results. Therefore, this paper proposes a high-precision numerical model calibration method, which uses Bayesian optimization to optimize the elastic constants to ensure the accuracy of numerical modeling. Then, the concept of equivalent thickness is proposed, and signals corresponding to different thicknesses are generated using the numerical model. The damage path is determined by the similarity between the signals. After determining the damage path, a damage numerical model is established, and parametric scanning is performed on the width and depth information of the corrosion damage. A dataset is constructed using the similarity between the model output signal and the experimental signal as an index. A surrogate model is constructed through the Kriging model to predict unsampled points and achieve high-precision inversion of the damage width and depth information. The minimum distance method is used to determine the damage center, and geometric relationships are used to determine the damage boundary points. All boundary points are fitted to obtain the damage quantification prediction result. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for high-resolution damage quantitative detection in the case of unknown experimental environment and material parameters of the detection object, which can cooperate with the numerical model to quickly invert unknown parameters, calibrate the numerical model, and improve the accuracy of damage quantification. At the same time, a surrogate model is constructed to accurately estimate unsampled points, realizing high-resolution inversion of damage information and improving the quantitative accuracy of corrosion damage.

[0006] To solve the above problems, the present invention proposes a metal corrosion damage evaluation inversion method based on the Bayesian and Kriging models, which is characterized in that it includes the following steps:

[0007] S1. Ultrasonic guided wave signal acquisition: Paste piezoelectric ceramics on the surface of the metal specimen to be tested, excite through a signal generator, connect the output signal to the sensor through a signal amplifier, and then perform signal acquisition to form an ultrasonic guided wave signal dataset;

[0008] S2. Construct a complete two-dimensional simulation model of ultrasonic guided wave propagation without damage based on the COMSOL MULTIPHYSICS commercial software;

[0009] S3. Calibrate the mechanical properties of the simulation model; Introduce the Bayesian optimization algorithm to optimize the two-dimensional simulation model of ultrasonic guided wave propagation without damage, providing a basis for constructing a corrosion damage signal library;

[0010] S4. Identification of the corrosion damage path of the metal specimen. Introduce the equivalent thickness EI of the metal specimen as the identification index, and obtain the equivalent thickness index EI corresponding to the metal specimen by solving the propagation velocity υ g , and the identification index is expressed as:

[0011]

[0012] where d damage is the damage propagation distance on this path, and d intact is the non-damage propagation distance on this path, υ g1 is the propagation speed corresponding to the thickness of the damaged metal specimen, and υ g2 is the propagation speed corresponding to the thickness of the non-damaged metal specimen, υ g is the propagation speed corresponding to EI;

[0013] S5. Solve the width and depth information of the corrosion damage of the metal specimen on different paths;

[0014] S51. Construct a mapping model between the corrosion damage information of the metal specimen and the ultrasonic guided wave signal, introduce the Kriging model, and increase the fineness of the damage information;

[0015] S52. Through the established Kriging model, predict the unsampled points on the metal specimen, realize the inverse solution of the corrosion width-depth combination with the minimum RMSE value, and solve the optimal corrosion width-depth combination of the metal specimen under this damage path;

[0016] S53. Repeat step S52 for each damage path on the metal specimen to solve the damage information of each path;

[0017] S6. Use the minimum distance method to locate the damage center of the metal specimen, and solve the point with the minimum sum of distances to each damage path in the full geometric space;

[0018]

[0019] where x i , y i are the coordinates of each point in the scanning space, d() represents solving the distance from any point to all damage paths, and argmin() represents the minimum value of all combinations;

[0020] S7. Draw a perpendicular line from the damage center of the metal specimen to the propagation path, draw a circle with the foot of the perpendicular as the center and the damage length as the diameter, solve the intersection points of the circle and the damage path, and obtain the boundary points of the damage;

[0021] S8. Fit all the boundary points of the damage of the metal specimen by the least squares method.

[0022] Furthermore, step S3 specifically includes the following steps:

[0023] S31. Evaluate the objective function, and use the difference between the output signal of the non-damage guided wave propagation two-dimensional simulation model and the experimental signal as the objective function, which is expressed as:

[0024]

[0025] Among them, \(x\) represents a \(d\)-dimensional decision vector, represents the decision space, and signal observed is the signal output by the simulation model, and signal measured is the experimental signal;

[0026] S32. Construct a prior model of the objective function through the initially selected elastic constants, and determine the next evaluation point of the elastic modulus combination by maximizing the acquisition function;

[0027] S33. Add the newly evaluated points to the historical observation data set, update the elastic modulus parameter combination of the non-destructive guided wave propagation two-dimensional simulation model, solve the elastic modulus corresponding to the minimum value of the objective function, and narrow the distance between the non-destructive guided wave propagation simulation model and the actual physical entity model.

[0028] Furthermore, step S4 specifically includes the following steps:

[0029] S41. Based on the optimized non-destructive guided wave propagation two-dimensional simulation model, perform parametric scanning on the thickness of the metal specimen plate, and solve the ultrasonic guided wave signals corresponding to different thicknesses;

[0030] S42. Compare the signals output by the corrosion damage detection experiment with the signal set output by the non-destructive guided wave propagation two-dimensional simulation model, measure the approximation degree of the two groups of signals with the RMSE value as an index, and obtain the equivalent thickness index EI;

[0031] EI = RMSE(intact signal library - signal experimental ) (2)

[0032] Among them, intact signal library represents the signal library corresponding to different thicknesses constructed by the non-destructive guided wave propagation simulation model, and signal experimental represents the experimental signal, and RMSE() is the root mean square error, which represents the square root of the ratio of the sum of the squares of the deviations between the predicted value and the true value to the number of observations.

[0033] Preferably, step S51 specifically includes the following steps:

[0034] S511. Construct a damage model according to the calibrated mechanical properties of the metal specimen material;

[0035] S512. Perform parametric scanning within the specified damage parameter space, solve the ultrasonic guided wave signals corresponding to different corrosion damage widths and damage depths, and construct an ultrasonic guided wave signal library;

[0036] S513. Calculate the RMSE values of each signal in the ultrasonic guided wave signal library and the target experimental signal, and construct a dataset DI of the corrosion damage information of the metal specimen and the RMSE index;

[0037] DI i = RMSE(damage signal library(i) - Signal experimental ) (3)

[0038] where damage signal library represents the signal library corresponding to different corrosion depths and widths constructed by the damage guided wave propagation simulation model, and signal experimental represents the experimental signal;

[0039] S514. Introduce the Kriging model to construct a mapping model between the corrosion damage information and the ultrasonic guided wave signal, and increase the fineness of the damage information.

[0040] Preferably, the implementation process of increasing the fineness of the damage information by introducing the Kriging model in step S514 is as follows:

[0041] Construct a mapping relationship between the corrosion damage information of the metal specimen and the signal difference index; the model consists of two parts: residual and Gaussian process prior:

[0042] f(X) = X T ω, Y = f(X) + ε (4)

[0043] where X = {x1, x2,..., x n} is the given corrosion damage width and depth of the metal specimen as the input dataset, Y = {DI1, DI2,..., DI n} represents the set of RMSE indexes, ω is the weight coefficient, ∈ is the residual, and f(X) represents the prior function following the Gaussian process; given the learning samples, assume that the regression residual follows an independent and identically distributed Gaussian distribution, that is

[0044] For f(X), by means of the Gaussian distribution, it is expressed as:

[0045]

[0046] where, represents the Gaussian process, and this Gaussian distribution is determined by the mean m(x) and the covariance . Among them, m(x) is set to 0, and (X, X′) represents any two different sample points in the parameter space, then the covariance represents the covariance matrix between each dimension of the two vectors, that is, the kernel function, and is further expressed as:

[0047]

[0048] Predict the new sample x that is not sampled in the damage information dataset of the metal specimen using the Kriging model * and there is

[0049] y * = f(x * ) + ∈ (7)

[0050] It follows a Gaussian conditional distribution:

[0051]

[0052] where x * is the input value, that is, the two variables of the length and width of the corrosion damage is the output mean, that is, the RMSE value solved through the simulation signal and the experimental signal, coυ(y * ) is the output of the variance and there is:

[0053]

[0054]

[0055] That is, the combination of the width and depth of the corrosion damage obtained by prediction

[0056] Compared with the prior art, the present invention has the following advantages:

[0057] 1. The present invention uses ultrasonic guided waves as the detection method. This detection method has high reliability through the experiment of pasting piezoelectric wafers and has the potential to be applied in complex environments.

[0058] 2. In the present invention, the model calibration is based on the Bayesian optimization algorithm. The sampling points in the parameter space are selected through the posterior probability, and the model is updated using the collected information to complete the optimization.

[0059] 3. The present invention constructs the mapping relationship between damage information and signal indicators based on the Kriging algorithm, predicts the unsampled points based on the Bayesian theory, and uses the genetic algorithm to inversely solve the damage information. It overcomes the quantitative error and improves the accuracy of damage location and quantification.

[0060] 4. The present invention proposes the minimum distance method for quickly locating the center of the corrosion damage. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 is a schematic diagram of the laboratory test device of the present invention;

[0062] Figure 2Flowchart of the model calibration method based on Bayesian optimization and the inversion method for quantitative evaluation of metal corrosion damage based on Kriging model in an embodiment of the present invention;

[0063] Figure 3 Sensor layout diagram in the present invention;

[0064] Figure 4 Schematic diagram of the Bayesian optimization result in the present invention;

[0065] Figure 5 、 Figure 6 Schematic diagram of the equivalent thickness in the present invention;

[0066] Figure 7 Schematic diagram of the damage center location result in the present invention;

[0067] Figure 8 Schematic diagram of the damage boundary point location method in the present invention;

[0068] Figure 9 Quantitative comparison diagram of corrosion damage in the present invention.

[0069] Reference numerals: 1 - signal acquisition end; 2 - signal amplifier; 3 - signal generator; 4 - structure to be inspected Detailed implementation manners

[0070] For a better understanding of the technical solution of the present invention, the following further describes in detail the specific implementation manners of the present invention in conjunction with the drawings and embodiments. The same reference numerals in the drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the drawings, unless otherwise specified, the drawings do not have to be drawn to scale.

[0071] Taking laboratory tests as an example, the entire test system includes a signal acquisition end 1, a signal amplifier 2, a signal generator 3, and a structure to be inspected 4, as Figure 1 shown. Piezoelectric wafers are bonded to the surface of the structure to be inspected, and the output line of the signal generator is connected to the signal amplifier and then to the piezoelectric wafers to generate ultrasonic guided waves for transmission in the structure to be inspected. Single sensors are excited one by one, and the remaining sensors receive the signals, which are displayed and stored through an oscilloscope.

[0072] As Figure 2 shown, the inversion method for quantitative evaluation of corrosion damage based on Bayesian optimization and Kriging model disclosed in the present invention is implemented as follows:

[0073] S1. Ultrasonic guided wave signal acquisition; To verify the reliability of the method proposed in this patent, guided wave damage verification is carried out for the corrosion damage of a metal plate, as Figure 3As shown. The thickness of the specimen is 4 mm. A circular corrosion damage with a depth of 2 mm and a diameter of 20 mm is machined at a position 150 mm away from the plate edge. The piezoelectric ceramics are pasted on the surface of the specimen to be tested, which serves as both a sensor to receive signals and an excitation end to emit signals. The sensor layout and the damage position are as Figure 1 shown. The thickness change on the sensing path will affect the arrival time of the guided wave signal. Therefore, we select the direct component for analysis and analyze the change of characteristics to analyze and evaluate the damage situation.

[0074] It is excited by a signal generator. The output signal is connected to the sensor through a signal amplifier, and then signal acquisition is carried out. The specific sensor layout is as Figure 3 . A chirp signal of 5 - 500 kHz is selected for excitation, and the 5 - cycle toneburst narrowband signal component with a center frequency of 80 kHz is extracted. The sampling rate is 5 MHz. Therefore, the A0 mode can be easily extracted to reflect the nature of the damage. Among them, there are 11 excitation ends and 11 receiving ends, and a total of 11×11 = 121 groups of signals are collected.

[0075] S2. Build a complete two - dimensional simulation model of non - damaged guided wave propagation based on the commercial software COMSOL MULTIPHYSICS.

[0076] S3. The mechanical properties of the object to be inspected have a great influence on the received ultrasonic signal. In order to enhance the accuracy of damage detection and extract signal characteristics more accurately, first, the mechanical properties of the simulation model need to be calibrated to build a more accurate numerical model to achieve high - precision damage detection of the metal plate.

[0077] S31. Evaluate the objective function. Take the difference between the output signal of the two - dimensional simulation model of non - damaged guided wave propagation and the experimental signal as the objective function, which is expressed as:

[0078]

[0079] Among them, x represents a d - dimensional decision vector, represents the decision space, signal observed is the signal output by the simulation model, and signal measured is the experimental signal;

[0080] S32. Build a prior model of the objective function through the initially selected elastic constants, and determine the next evaluation point of the elastic modulus combination by maximizing the acquisition function;

[0081] S33. Add the newly evaluated points to the historical observation dataset, update the elastic modulus parameter combination of the non-damaged guided wave propagation two-dimensional simulation model, solve for the elastic modulus that minimizes the corresponding objective function, and narrow the gap between the non-damaged guided wave propagation simulation model and the actual physical entity model.

[0082] To verify the stability of Bayesian optimization, 8 groups of signals (T1-R1, T2-R3, T3-R3, T4-R4, T8-R8, T9-R9, T10-R10, T11-R11) with a distance of 150 mm were selected for elastic constant inversion, and the optimization results are as Figure 4 shown. The median of each group of data was selected as the final optimization result and applied to the subsequent numerical model.

[0083] S4. Identification of the corrosion damage path of the metal plate; Due to the thickness reduction caused by corrosion, it will greatly affect the propagation speed on this sensing path. To identify whether there is damage on this path, the equivalent thickness EI is introduced as an identification index. By solving the propagation speed υ g corresponding to the equivalent thickness, the corresponding equivalent thickness index EI is obtained. The identification index is expressed as:

[0084]

[0085] where d damage is the damage propagation distance on this path, d intact is the non-damaged propagation distance on this path, υ g1 is the propagation speed corresponding to the damaged thickness, and υ g2 is the propagation speed corresponding to the non-damaged thickness;

[0086] S41. As can be easily seen from the above formula, the equivalent thickness is the thickness corresponding to the group velocity υ g . Based on the optimized elastic constant information, update the established numerical model, perform parametric scanning on the plate thickness, and solve the ultrasonic guided wave signals corresponding to different thicknesses.

[0087] S42. Compare the damage detection experimental signals with the signal set output by the numerical model respectively, and use the RMSE value as an index to measure the approximation degree of the two groups of signals, then the equivalent thickness index can be solved. Here only show the equivalent thickness solution results obtained with T5 and T6 as excitations, as Figure 5 , Figure 6 shown.

[0088] S5. After completing the identification of the damage path, it is necessary to solve the width and depth information of the corrosion damage on different paths.

[0089] S51. To accurately correspond damage information with ultrasonic signals, it is necessary to construct a mapping model between the two, introduce the Kriging model, and increase the fineness of damage information.

[0090] S511. Construct a damage model according to the calibrated mechanical properties of the metal plate material.

[0091] S512. Conduct parametric scanning within the specified damage parameter space, solve the ultrasonic guided wave signals corresponding to different damage widths and depths, and construct an ultrasonic guided wave signal library.

[0092] S513. Calculate the RMSE values of each signal in the ultrasonic guided wave signal library and the target experimental signal, and construct a data set DI of corrosion damage information and RMSE index.

[0093] DI i = RMSE(damage signal library(i) - signal experimental )

[0094] S514. Introduce the Kriging model, which has applications in fields such as time series analysis, image processing, and automatic control, construct a mapping model between damage information and ultrasonic guided wave signals, and increase the fineness of damage information.

[0095] Construct a mapping relationship between the corrosion damage information of the metal plate and the signal difference index; assume that the model consists of two parts: residual and Gaussian process prior, as follows:

[0096] f(X) = X T ω, Y = f(X) + ∈

[0097] where X = {x1, x2,..., x n} is the input data set of the corrosion damage width and depth of the given metal specimen, Y = {DI1, DI2,..., DI n}, represents the set of RMSE indices, ω is the weight coefficient, ∈ is the residual, and f(X) represents the prior function following the Gaussian process; in the case of given learning samples, assume that the regression residual follows an independent and identically distributed Gaussian distribution, that is

[0098] For f(X), with the help of the Gaussian distribution, represent it as:

[0099]

[0100] where, represents the Gaussian process, and this Gaussian distribution is composed of the mean m(x) and the covariance Determine that m(x) is set to 0, and (X, X′) represents any two different sample points in the parameter space. Then the covariance represents the covariance matrix between the dimensions of two vectors, that is, the kernel function; it can be further expressed as:

[0101]

[0102] Use the Kriging model to predict the new unsampled sample x in the damage information dataset * Then there is

[0103] y * = f(x * ) + ∈

[0104] It follows a Gaussian conditional distribution:

[0105]

[0106] where x * is the input value, that is, the two variables of the length and width of the corrosion damage, is the mean of the output, that is, the RMSE value obtained by solving the simulation signal and the experimental signal. Coυ(y * ) is the output of the variance and there is:

[0107]

[0108]

[0109] That is, the combination of the corrosion damage width and depth obtained by prediction.

[0110] S52. Through the established Kriging model, predict the unsampled points on the metal plate to realize the inverse solution of the corrosion width-depth combination with the minimum RMSE value, and solve the optimal corrosion width-depth combination under this damage path.

[0111] S53. Repeat step S52 for each damage path on the metal plate to solve the damage information of each path.

[0112] S6. Use the minimum distance method to locate the damage position on the metal plate and solve the point with the minimum sum of distances to all damage paths in the entire geometric space;

[0113]

[0114] where x i , y i are the coordinates of each point in the scanning space, d() represents solving the distance from any point to all damage paths, and argmin() represents the minimum value of all combinations; the solution result is asFigure 7 As shown, the minimum point is (150, 250), which is consistent with the actual damage center.

[0115] S7. Draw a perpendicular line from the damage center on the metal plate to the propagation path. Take the foot of the perpendicular as the center of the circle and draw a circle with the damage length as the diameter. Solve the intersection points of the circle and the damage path to obtain the boundary points of the damage, as Figure 8 shown.

[0116] S8. Fit all the boundary points of the damage on the metal plate by the least square method. The result is as Figure 9 shown.

[0117] The present invention uses ultrasonic guided wave signals as the analysis basis, which is simple and convenient to operate and can adapt to complex and harsh external field conditions. The present invention creatively proposes a model calibration method based on Bayesian optimization, with a relatively fast operation rate, and realizes a calibration scheme for a high-fidelity numerical model. And an innovative minimum distance method is proposed to realize the rapid positioning of the corrosion damage center position. At the same time, with the help of the Kriging model algorithm, the damage information is mapped to the signal index. Finally, through the damage imaging algorithm, a positioning and quantitative evaluation scheme for corrosion damage can be realized.

[0118] Finally, it should be noted that the above-described embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A metal corrosion damage evaluation inversion method based on Bayesian and Kriging models, characterized in that: It includes the following steps: S1. Ultrasonic guided wave signal acquisition: Paste piezoelectric ceramics on the surface of the metal specimen to be tested, excite them through a signal generator, connect the output signal to the sensor through a signal amplifier, and then perform signal acquisition to form an ultrasonic guided wave signal dataset; S2. Build a complete non-destructive ultrasonic guided wave propagation two-dimensional simulation model based on the COMSOL MULTIPHYSICS commercial software; S3. Calibrate the mechanical properties of the simulation model; Introduce the Bayesian optimization algorithm to optimize the non-destructive guided wave propagation two-dimensional simulation model, providing a basis for constructing a corrosion damage signal library; S4. Identification of the corrosion damage path of the metal specimen. The equivalent thickness EI of the metal specimen is introduced as the identification index, and the propagation velocity v corresponding to the equivalent thickness of the metal specimen is obtained by solving g , and the equivalent thickness index EI corresponding to the metal specimen is obtained. The identification index is expressed as: where d damage is the damage propagation distance on this path, d intact is the non-damage propagation distance on this path, v g1 is the propagation speed corresponding to the thickness of the damaged metal specimen, v g2 is the propagation speed corresponding to the thickness of the non-damaged metal specimen, v g is the propagation speed corresponding to EI; S5. Solve the width and depth information of the corrosion damage of the metal specimen on different paths; S51. Build a mapping model between the corrosion damage information of the metal specimen and the ultrasonic guided wave signal, introduce the Kriging model to increase the fineness of the damage information; S52. Through the established Kriging model, predict the unsampled points on the metal specimen, realize the inverse solution of the corrosion width-depth combination with the minimum RMSE value, and solve the optimal corrosion width-depth combination of the metal specimen under the damage path; S53. Repeat step S52 for each damage path on the metal specimen to solve the damage information of each path; S6. Use the minimum distance method to locate the damage center of the metal specimen and solve the point with the minimum sum of distances to each damage path in the entire geometric space; where x i , y i are the coordinates of each point in the scanning space, d() represents the distance from any point to all damage paths, and argmin() represents the minimum value of all combinations; S7. Draw a perpendicular line from the damage center of the metal specimen to the propagation path, draw a circle with the foot of the perpendicular as the center and the damage length as the diameter, solve the intersection points of the circle and the damage path, and obtain the boundary points of the damage; S8. Fit all the boundary points of the metal specimen damage by the least squares method.

2. The metal corrosion damage evaluation inversion method based on the Bayesian and Kriging models according to claim 1, characterized in that Step S3 specifically includes the following steps: S31. Evaluate the objective function, and use the difference between the output signal of the non-destructive guided wave propagation two-dimensional simulation model and the experimental signal as the objective function, expressed as: where x represents a d-dimensional decision vector, represents the decision space, and signal observed is the signal output by the simulation model, and signal measured is the experimental signal; S32. Build a prior model of the objective function through the initially selected elastic constants, and determine the next evaluation point of the elastic modulus combination by maximizing the acquisition function; S33. Add the newly evaluated points to the historical observation dataset, update the elastic modulus parameter combination of the non-destructive guided wave propagation two-dimensional simulation model, solve the elastic modulus corresponding to the minimum value of the objective function, and narrow the distance between the non-destructive guided wave propagation simulation model and the actual physical entity model.

3. The Bayesian and Kriging model-based metal corrosion damage evaluation inversion method according to claim 1, wherein Step S4 specifically includes the following steps: S41. Based on the optimized non-destructive guided wave propagation two-dimensional simulation model, perform parametric scanning on the plate thickness of the metal specimen and solve the ultrasonic guided wave signals corresponding to different thicknesses; S42. Compare the signals output by the corrosion damage detection experiment with the signal set output by the non-destructive guided wave propagation two-dimensional simulation model, measure the approximation degree of the two groups of signals with the RMSE value as an index, and obtain the equivalent thickness index EI; EI = RMSE(intact signa llibrary -signal experimental ) (2) Among them, intact signal library represents the signal library corresponding to different thicknesses constructed by the intact guided wave propagation simulation model, signal experimental represents the experimental signal, and RMSE() is the root mean square error, which represents the square root of the ratio of the sum of the squares of the differences between the predicted value and the true value to the number of observations.

4. The metal corrosion damage evaluation inversion method based on the Bayesian and Kriging models according to claim 1, wherein, Step S51 specifically includes the following steps: S511. Build a damage model according to the calibrated mechanical properties of the metal specimen material; S512. Perform parametric scanning within the specified damage parameter space, solve the ultrasonic guided wave signals corresponding to different corrosion damage widths and damage depths, and build an ultrasonic guided wave signal library; S513. Calculate the RMSE values of each signal in the ultrasonic guided wave signal library and the target experimental signal, and construct a data set DI of the corrosion damage information of the metal specimen and the RMSE index; DI i = RMSE(damage signal library(i) - signal experimental ) (3) Among them, damage signal library represents the signal library corresponding to different corrosion depth-widths constructed by the damage guided wave propagation simulation model, and signal experimental represents the experimental signal; S514. Introduce the Kriging model to construct a mapping model between the corrosion damage information and the ultrasonic guided wave signal, and increase the fineness of the damage information.

5. The metal corrosion damage evaluation inversion method based on the Bayesian and Kriging models according to claim 4, characterized in that The implementation process of step S514 of introducing the Kriging model to construct a mapping model between the corrosion damage information and the ultrasonic guided wave signal and increasing the fineness of the damage information is as follows: Construct a mapping relationship between the corrosion damage information of the metal specimen and the signal difference index; the model consists of two parts: the residual and the Gaussian process prior; f(X) = X T ω, Y = f(X) + ∈ (4) where X = {x1, x2,..., x n} is the given corrosion damage width and depth of the metal specimen as the input data set, Y = {DI1, DI2,..., DI n}, represents the set of RMSE indicators, ω is the weight coefficient, ∈ is the residual, and f(X) represents the prior function following the Gaussian process; given the learning samples, it is assumed that the regression residuals follow an independent and identically distributed Gaussian distribution, that is For f(X), with the help of the Gaussian distribution, it is expressed as: Among them, represents a Gaussian process, and the Gaussian distribution is determined by the mean m(x) and the covariance . Among them, m(x) is set to 0, (X, X′) represents any two different sample points in the parameter space, and the covariance represents the covariance matrix between the dimensions of two vectors, that is, the kernel function, which is further expressed as: Predict the new sample x that is not sampled in the damage information dataset of metal specimens using the Kriging model * Then there is y * = f(x * ) + ∈ (7) It follows the Gaussian conditional distribution: where x * is the input value, i.e., two variables of the length and width of the corrosion damage, is the output mean value, i.e., the RMSE value solved by the simulation signal and the experimental signal, cov(y * ) is the output of the variance and there is: That is the predicted combination of corrosion damage width and depth.

Images