Method for evaluating corrosion fatigue crack propagation rate of steel member

Abstract

The invention discloses a method for evaluating the corrosion fatigue crack growth rate of a steel member by fusing a physical model and data driving. The method comprises the following steps: firstly, acquiring data through pre-corrosion and fatigue tests, and calculating a stress intensity factor by adopting M-integral; and further, analyzing the characteristic correlation by using a Pearson coefficient and an SHAP-XGBoost method and quantifying the contribution degree of each parameter to crack propagation, thereby proposing a Paris correction formula. Then, a mixed loss function is constructed, data-driven artificial neural network loss and physical model loss based on a Paris correction formula are considered at the same time, and the data-driven artificial neural network loss and the physical model loss are harmonized through a physical influence coefficient; and finally, performing cycle training and hyper-parameter optimization on the mixed model by adopting a Bayesian optimizer until the prediction precision reaches a set threshold value, thereby realizing high-precision crack growth rate prediction. The prediction method has high precision and high reliability, and can be suitable for crack propagation prediction of actual engineering steel structures such as bridges, civil engineering and ocean engineering.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of metal material performance testing, and particularly relates to a method for evaluating corrosion fatigue crack propagation rate of a steel member. BACKGROUND

[0002] Due to reasons such as machining, transportation collision, welding connection, installation and use and maintenance, some defects, that is, initial cracks, are generally present in the steel member. Under the continuous action of cyclic loads (such as wind load, traffic load, wave load and the like), the cracks will gradually propagate, and when the crack size reaches a critical value, brittle fracture or fatigue failure of the member will occur, causing safety accidents and huge economic losses. Therefore, prediction and evaluation of the crack propagation behavior in the steel member are core links for guaranteeing the safety of the structure, realizing predictive maintenance and prolonging the service life.

[0003] Paris formula is widely used in the evaluation of fatigue crack propagation rate, but Paris formula needs to rely on a large amount of fatigue test data, and fatigue test is often time-consuming and costly. In the case of limited samples, the prediction model will have a large error, and Paris formula does not take into account the influence of corrosion rate.

[0004] Therefore, in order to solve the above problems, the method of the present application uses a method of combining test data-fatigue knowledge-Paris model-PINN intelligent algorithm to analyze the fatigue crack propagation of the steel member. SUMMARY

[0005] The present application aims to provide a method for evaluating corrosion fatigue crack propagation rate of a steel member, so as to solve the problems of limited prediction accuracy and lack of consideration of corrosion rate in the existing fatigue crack propagation evaluation method in the case of limited samples.

[0006] To achieve the above-mentioned purpose, the present application provides the following technical scheme:

[0007] A method for evaluating corrosion fatigue crack propagation rate of a steel member, comprising the following steps:

[0008] S1, a steel test piece is prepared, the steel test piece is pre-corroded, and after pre-corrosion, the steel test piece is taken out and subjected to fatigue loading test to obtain fatigue loading test result data;

[0009] S2, the stress intensity factor is calculated by using M-integral:

[0010]

[0011] In the formula, Γ is the counterclockwise integral path around any two crack tips: A and B are the starting points of the two paths, respectively; q is an intermediate variable; and W is a strain energy density factor. is the stress tensor; is the displacement vector along the integral path; is the strain tensor; is the integral arc microelement;

[0012]

[0013] wherein E is the elastic modulus, with the unit of MPa; v is the Poisson's ratio;

[0014] S3, the Pearson coefficient is used to analyze the characteristic correlation, and based on the SHAP-XGBoost method and the Paris model, the contribution degrees of the stress intensity factor amplitude , the stress ratio R and the corrosion degree P to the crack propagation rate are quantified, and a Paris propagation rate correction formula is proposed:

[0015]

[0016] S4, the test data and the characteristic influence parameter values are put into the neural network for training, the test data are put into the ANN for pre-fitting to realize the neural network to alleviate the gradient problem, the weight is initialized, the convergence speed is improved, and the loss function based on data driving is determined

[0017] S5, based on the defect sensitive features and the corrected Paris formula, the physical model loss function is determined, the physical model loss function is composed of the Paris loss function and the defect model loss function , is the offset of the crack propagation speed prediction value based on the offset of the corrected Paris formula, is the stress intensity factor, the corrosion degree and the stress ratio in the fatigue life gradient direction:

[0018]

[0019]

[0020]

[0021] wherein , , respectively represent the loss function of the stress intensity factor, the corrosion degree and the stress ratio under its physical criterion;

[0022] by introducing the physical loss function influence coefficient , the physical model loss function and the loss function based on data driving are harmonized​ The final loss function Loss is obtained:

[0023]

[0024] S6, adopting a Bayesian optimizer to optimize the influence coefficient of the physical loss function and the number of neurons in the hidden layer of the ANN.

[0025] S7, returning to step S5 for training and cross-validation according to the size of the loss function value Loss and the size of the hyperparameters after Bayesian optimization, until the prediction error reaches a satisfactory threshold.

[0026] Further, the loading of the test results in step S1 specifically includes stress ratio, stress amplitude, corrosion degree, crack length, and cycle number.

[0027] Principles and beneficial effects of the technical solution: The method of the present application utilizes a hybrid framework of physical models and data-driven models, and by introducing a physical loss function, the existing Paris model physical law is embedded into neural network training as a constraint, so that the model is more consistent with the physical mechanism, and the prediction accuracy and generalization ability are improved; The environmental factors of corrosion are considered at the beginning of the method, so that the method can be closer to the service conditions of bridge, offshore platform and other actual engineering structures in harsh environments, and the evaluation results have more engineering guiding value. BRIEF DESCRIPTION OF DRAWINGS

[0028] Fig. 1 For the lg is the fitting result of the target.

[0029] Fig. 2 For the lg is the fitting result of the target. DETAILED DESCRIPTION

[0030] The present application will be further described in detail below in combination with the drawings and embodiments:

[0031] In the evaluation of fatigue crack propagation rate, the commonly used calculation formula in the prior art is the Paris formula:

[0032]

[0033] In the formula: a is the crack length (mm); N is the number of load cycles; C and m are material constants, which are determined through material fatigue test; is the stress intensity factor amplitude (Mpa·m 1 / 2 ).

[0034] Paris formula needs to rely on a large number of fatigue test data, and fatigue test is often time-consuming and costly. In the case of limited samples, the prediction model has a large error. Furthermore, Paris formula does not consider the influence of corrosion rate.

[0035] Therefore, in order to solve the above problems, the present application adopts the method of combining test data-fatigue knowledge-Paris model-PINN intelligent algorithm to analyze the fatigue crack propagation of steel members and evaluate the corrosion fatigue crack propagation rate of steel members. The specific steps are as follows:

[0036] Step 1: Make steel test pieces, put them in a salt spray corrosion box for a certain period of time, and take them out after reaching a certain corrosion rate. Perform fatigue loading test to obtain test result data: stress ratio, stress amplitude, corrosion degree, crack length, cycle number.

[0037] Step 2: Calculate the stress intensity factor using M-integral:

[0038]

[0039] In the formula, Γ is the counterclockwise integral path around any two crack tips: , A and B are the starting points of the two paths; q is an intermediate variable; W is the strain energy density factor; is the stress tensor; is the displacement vector on the integral path; is the strain tensor; is the integral arc microelement;

[0040]

[0041] In the formula, E is the elastic modulus, unit: MPa; v is the Poisson's ratio;

[0042] Step 3: Use Pearson correlative coefficirnt to explore the correlation between features, and use SHAP (shapley Additive exPlanations) + XGboost to explore the contribution of feature influence parameters (stress intensity factor amplitude , stress ratio R and corrosion degree P) to crack propagation rate, and propose a modified Paris propagation rate formula:

[0043]

[0044] Step 4: Put the test data and related parameter values into the neural network for training. First, put the test data into the artificial neural network (ANN) for pre-fitting to achieve neural network gradient relief, initialize the weight, improve the convergence speed, and determine the data-driven loss function .

[0045] Step 5: Based on the defect sensitive feature and the corrected Paris formula, determine the physical model loss function , the physical model loss function is composed of the Paris loss function and the defect model loss function , is the crack propagation speed prediction value, which is the offset based on the corrected Paris formula, is the stress intensity factor, corrosion degree and stress ratio in the fatigue life gradient direction offset:

[0046]

[0047]

[0048]

[0049] In the formula, , , respectively represent the loss function of the stress intensity factor, corrosion degree and stress ratio under its physical criterion;

[0050] By introducing the physical loss function influence coefficient

[0051] , the physical model loss function and the data-driven loss function are obtained. The final loss function Loss is:

[0052]

[0053] Step 6: Use the Bayesian optimizer to optimize the physical loss function influence coefficient and the number of neurons in the hidden layer of ANN.

[0054] Step 7: According to the size of the loss function value Loss and the size of the hyperparameters after Bayesian optimization, return to step S5 for training and cross-validation until the prediction value error reaches a satisfactory threshold.

[0055] The specific implementation process is as follows:

[0056] The method of this invention was used to analyze the pre-corrosion fatigue crack propagation rate of a batch of pre-corroded Q690 steel plate specimens with butt welds and coatings, and the modified crack propagation rate formula was obtained:

[0057]

[0058] Its fitting effect is as follows Figs. 1-2 As shown.

[0059] like Fig. 1 The following is for lg The fitting results for the target are shown in the figure. It can be seen that the error of 100% of the data points in the cross-validation set is within 1%. For example... Fig. 2 The figure shows lg The error was amplified to As shown in the figure, the error of the 100% cross-validation set is controlled within 8%, the error of the 86.15% cross-validation set is controlled within 5%, and the error of the 52.31% cross-validation set is controlled within 1%. This demonstrates that the model has high fitting accuracy, and the predicted values ​​are very close to the actual values. Even when the crack propagation rate is restored to 10... -9 ~10 -6 The error between the predicted and actual values ​​is still effectively limited by orders of magnitude. Furthermore, the model incorporates relevant physical criteria and formulas, providing a sound physical interpretation.

[0060] The above descriptions are merely embodiments of the present invention, and common technical solutions or characteristics known in the schemes are not described in detail here. For those skilled in the art, various modifications and improvements can be made without departing from the technical solutions of the present invention, and these should also be considered within the scope of protection of the present invention. These modifications and improvements will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application shall be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A method for evaluating the corrosion fatigue crack propagation rate of steel components, characterized in that, Includes the following steps: S1. Prepare steel specimens, pre-corrode the steel specimens, remove them after pre-corrosion, conduct fatigue loading tests, and obtain fatigue loading test results data. S2. Stress intensity factor is calculated using M-integral: In the formula, Γ represents the counterclockwise integration path around any two crack tips: A and B are the starting points of the two paths, respectively; q is an intermediate variable; W is the strain energy density factor. For stress tensor; The displacement vector along the integration path; For strain tensor; It is an integral arc infinitesimal element; In the formula, E is the elastic modulus, with units of MPa; v is Poisson's ratio; S3. Pearson coefficient analysis was used to analyze the characteristic correlation, and the stress intensity factor amplitude was quantified based on the SHAP-XGBoost method and the Paris model. The contributions of stress ratio R and corrosion degree P to crack propagation rate are analyzed, and a corrected Paris propagation rate formula is proposed. S4. Input the experimental data and the values ​​of the feature-influencing parameters into the neural network for training. Input the experimental data into the ANN for prefitting to alleviate the gradient problem, initialize the weights, improve the convergence speed, and determine the data-driven loss function. ; S5. Based on the defect-sensitive characteristics and the modified Paris formula, determine the physical model loss function. Physical model loss function Paris loss function Composed of the loss function of the defect model composition, The predicted crack propagation rate is based on the offset of the modified Paris formula. Its stress intensity factor, corrosion degree, and stress ratio offsets in the fatigue life gradient direction: In the formula, , , These represent the loss functions of stress intensity factor, corrosion degree, and stress ratio under their physical criteria, respectively. By introducing the influence coefficient of the physical loss function Harmonic physical model loss function and data-driven loss function The final loss function, Loss, is obtained as follows: S6. Using Bayesian optimizer to adjust the influence coefficient of the physical loss function The number of neurons in the hidden layers of an ANN is optimized. S7. Based on the magnitude of the loss function value Loss and the magnitude of the hyperparameters after Bayesian optimization, return to step S5 to perform training and cross-validation until the prediction error reaches a satisfactory threshold.

2. The method for evaluating the corrosion fatigue crack propagation rate of steel components according to claim 1, characterized in that: The loading test results in step S1 specifically include: stress ratio, stress amplitude, corrosion degree, crack length, and number of cycles.

Images