Intelligent prediction method and system for fatigue life of rusted steel plate

Abstract

The invention relates to the technical field of fatigue life prediction, in particular to an intelligent prediction method and system for the fatigue life of a rusted steel plate, and the method comprises the steps: measuring the three-dimensional shape of the rusted steel plate to be measured to obtain the surface data of the rusted steel plate, and then inputting the data into a trained crack fatigue initiation life prediction model, the model comprises a crack initiation life prediction module and a stress concentration coefficient prediction module, so that the fatigue initiation life of the crack is obtained. Then corrosion detail information generated by cracks is analyzed to obtain corrosion defect characteristics, and then defects are input into the fatigue crack propagation life prediction model to obtain corresponding life. And finally, calculating the total service life of the rusted steel plate according to the fatigue initiation service life and the fatigue crack propagation service life of the crack, thereby further improving the crack prediction precision.

Description

Technical Field

[0001] The present invention relates to the technical field of fatigue life prediction, and particularly to an intelligent prediction method and system for the fatigue life of rusted steel plates. Background Art

[0002] Steel structures are widely used in large-span structures, light steel structures, high-rise buildings, bridges and other fields due to their high strength and excellent seismic performance. However, components in engineering structures inevitably face fatigue problems during long-term use, that is, materials are gradually damaged under repeated or cyclic loads, eventually leading to failure. Although the strength of a single load is lower than the yield strength of the material, cracks gradually initiate in the material under repeated loads, and these cracks continue to expand, eventually resulting in material fracture. When there is water and air in the environment, serious corrosion will occur to steel. The influence of rust on fatigue is extremely significant and is the main driving force for accelerating the fatigue process. Rust pits and rough areas of varying depths are formed on the surface of the steel structure, and these irregular surface features greatly increase the stress concentration, making cracks more likely to initiate rapidly and expand faster in the rusted area. Rust will significantly shorten the fatigue life of components, causing the structure to reach failure in a relatively short time.

[0003] Therefore, how to accurately predict the fatigue life of rusted components and take targeted preventive and reinforcement measures has become an important part of the service performance evaluation and improvement of existing structures in a corrosive environment.

[0004] In the study of fatigue life, the fatigue life is usually divided into crack initiation life and crack propagation life. By studying these two stages separately and taking different analysis methods and measures for different mechanical behaviors, the overall fatigue life of materials or structures can be predicted more accurately.

[0005] Crack initiation refers to the time stage from the beginning until the appearance of initial micro-cracks on the surface or inside of the material. This stage is mainly affected by the surface state of the material. The rusted surface usually has a complex and irregular morphology, and the surface topography characteristics of the rusted steel plate can be accurately obtained through three-dimensional topography scanning. Fatigue crack initiation often originates from the defects or the maximum stress points of the specimen. Stress concentration leads to an increase in local stress, making it easier for the local area to exceed the fatigue limit of the material and enter the fatigue crack initiation stage in advance. The higher the stress concentration coefficient, the shorter the crack initiation life. The complexity of the rusted surface and stress concentration are important reasons for the variability of the fatigue crack initiation life of rusted steel plates.

[0006] The crack propagation life refers to the time period from the appearance of an initial crack to the final failure caused by crack propagation, which is mainly affected by factors such as the stress intensity factor at the crack tip. Through the observation of the rusty surface, it is found that there are different local defects at the bottom of the rust pit, which will lead to changes in the stress intensity factor at the bottom of the rust pit. Therefore, it can be considered that the local defects at the bottom of the rust pit are the main reason for the large variability in the crack propagation life of the rusty steel plate.

[0007] The crack initiation and propagation of rusty components are affected by various factors, making it difficult to systematically analyze their complex relationships. Traditional fatigue life models mainly rely on material properties and test conditions, with low universality, numerous parameters, and large computational amounts, thus having limited engineering application value. Summary of the Invention

[0008] The technical problem to be solved by the present invention is to provide an intelligent prediction method and system for the fatigue life of rusty steel plates in view of the deficiencies in the above-mentioned prior art, so as to solve the technical problem of the low accuracy of the existing fatigue life prediction of rusty steel plates.

[0009] The object of the present invention is achieved by the following technical solutions: In a first aspect, the present invention provides an intelligent prediction method for the fatigue life of rusty steel plates, including: S1. Measure the three-dimensional morphology of the rusty steel plate to be measured and obtain the surface data of the rusty steel plate to be measured; S2. Input the surface data of the rusty steel plate to be measured into the trained crack fatigue initiation life prediction model to obtain the crack fatigue initiation life of the rusty steel plate to be measured; the crack fatigue initiation life prediction model includes a crack initiation life prediction module and a stress concentration coefficient prediction module; the stress concentration coefficient prediction module is used to obtain the stress concentration coefficient according to the input surface data of the rusty steel plate to be measured, and the crack initiation life prediction module is used to obtain the crack initiation life according to the stress concentration coefficient; S3. Analyze the rust details information of the crack initiation to obtain the rust defect characteristics; S4. Input the rust defects of the rusty steel plate to be measured into the trained fatigue crack propagation life prediction model to obtain the fatigue crack propagation life of the rusty steel plate to be measured; S5. Obtain the total life of the rusty steel plate according to the crack fatigue initiation life and the fatigue crack propagation life of the rusty steel plate to be measured.

[0010] As a further improvement of the present invention, the measurement and analysis of the three-dimensional morphology of the rusty steel plate to be measured to obtain the surface data of the rusty steel plate to be measured specifically include: Use a three-dimensional non-contact optical profiler to record the morphology of the corroded surface of the steel plate to be measured, perform three-dimensional morphology measurement on the surface of the steel plate to be measured, and obtain the surface data of the steel plate to be measured. The surface data of the steel plate to be measured includes the distribution, depth, and size information of the surface corrosion pits.

[0011] As a further improvement of the present invention, inputting the surface data of the steel plate to be measured into the trained crack fatigue initiation life prediction model to obtain the crack fatigue initiation life of the steel plate to be measured specifically includes: Input the surface data of the corroded steel plate to be measured into the trained stress concentration coefficient prediction module to obtain the predicted maximum surface stress concentration coefficient; Multiply the predicted maximum surface stress concentration coefficient by the applied stress and then input it into the trained crack initiation life prediction module to obtain the crack fatigue initiation life of the steel plate to be measured; The training processes of the stress concentration coefficient prediction module and the crack initiation life prediction module respectively include: Import the collected surface data of the steel plate to be measured into finite element software, construct a solid model of the corroded steel plate according to the finite element software, perform finite element analysis on the corroded surfaces with different morphologies based on the solid model of the corroded steel plate, and calculate the stress concentration coefficient on the surface of the steel plate based on the solid model of the corroded steel plate; construct a database of the stress concentration coefficient on the surface of the corroded steel plate according to the geometric characteristics of the corroded surface and the corresponding stress concentration coefficients; Train the stress concentration coefficient prediction module according to the database of the stress concentration coefficient on the surface of the corroded steel plate; Determine the fatigue life of the uncorroded steel plate under different stress amplitudes and the proportion of the crack initiation life in the total fatigue life, and construct a database of the fatigue crack initiation life of the uncorroded steel plate; Train the crack initiation life prediction module according to the database of the fatigue crack initiation life of the uncorroded steel plate. During the training process of the crack initiation life prediction module, use the loss function corresponding to the neural network in the crack initiation life prediction module of the uncorroded steel plate to optimize the crack initiation life prediction module.

[0012] As a further improvement of the present invention, the crack initiation life prediction module adopts a physics-informed neural network model; the physical information corresponding to the crack initiation life prediction module is obtained according to the S-N curve. The physical information includes that the relationship between the crack initiation life and the stress amplitude applied to the corroded steel plate shows a monotonically decreasing trend, and the slope of the S-N curve gradually increases.

[0013] As a further improvement of the present invention, the loss function corresponding to the neural network in the crack initiation life prediction module of the uncorroded steel plate specifically includes:

[0014]

[0015]

[0016]

[0017] In the formula, y true is the actual value of the crack initiation life, and y pred is the predicted value of the crack initiation life, and Penaltys s (1) is used to penalize the case where the prediction function and the first derivative of the stress are positive, and Penaltys s (2) is used to penalize the case where the prediction function and the second derivative of the stress are negative.

[0018] As a further improvement of the present invention, a detailed analysis of the fatigue crack characteristics of the corroded steel plate is carried out, specifically including: Determine the rust pits corresponding to the crack initiation of the corroded steel plate to be measured according to the maximum stress concentration coefficient, obtain the two-dimensional cross-sectional view of the rust pits, and characterize the macroscopic characteristics of the rust pits on the surface of the corroded steel plate and the mesoscopic defects at the bottom of the rust pits; Establish a correlation model between the macroscopic geometric characteristics of the rust pits and the mesoscopic defect characteristics at the bottom of the rust pits through regression analysis.

[0019] As a further improvement of the present invention, the fatigue crack propagation life prediction model is trained using a fatigue crack propagation life database, and the training process of the fatigue crack propagation life prediction model includes: According to the correlation model, simplify the mesoscopic defects at the bottom of the rust pits on the surface of the corroded steel plate, simplify the irregular rust pits into elliptical regular concave pits, and simplify the mesoscopic defects at the bottom of the rust pits into multiple initial cracks in the crack propagation stage, and obtain a simplified model of the fatigue crack propagation of the corroded steel plate according to the defect characteristics; Use the finite element model to perform parametric analysis of the stress state of the simplified model of the corroded steel plate, and use FRANC3D to simulate the crack propagation behavior under cyclic loads under different conditions; The crack propagates until the stress intensity factor of the corroded steel plate exceeds the fracture toughness and fracture occurs, and record the number of cyclic loads at this time as the crack propagation life; Collect the crack propagation simulation results of different corroded steel plates, record the rust pit geometric characteristics, crack characteristics, cyclic load characteristics and crack propagation life corresponding to each model, construct a fatigue crack propagation life database, and train a fatigue crack propagation life prediction model based on the fatigue crack propagation life database.

[0020] As a further improvement of the present invention, the fatigue crack growth life prediction model adopts a physics-informed neural network model optimized based on the physical information in the Paris formula, where the physical information includes the influence of the changes in the maximum stress and the initial crack length on the crack growth life.

[0021] As a further improvement of the present invention, the loss function of the fatigue crack growth life prediction model is:

[0022] where

[0023] In the formula, P1 and P2 are respectively used to penalize the case where the first-order derivatives of the prediction function with respect to the maximum stress and the initial crack length are positive.

[0024] In a second aspect, the present invention provides an intelligent prediction system for the fatigue life of a corroded steel plate, which is used to implement the above-mentioned intelligent prediction method for the fatigue life of a corroded steel plate, and includes: A data acquisition unit that measures the three-dimensional morphology of the to-be-tested corroded steel plate and obtains the surface data of the to-be-tested corroded steel plate; A crack fatigue initiation life prediction unit that inputs the surface data of the to-be-tested corroded steel plate into the trained crack fatigue initiation life prediction model to obtain the crack fatigue initiation life of the to-be-tested corroded steel plate; the crack fatigue initiation life prediction model includes a crack initiation life prediction module and a stress concentration coefficient prediction module; the stress concentration coefficient prediction module obtains the stress concentration coefficient, and the crack initiation life prediction module is used to obtain the crack initiation life according to the stress concentration coefficient; A data analysis unit that analyzes the corrosion detail information of the crack initiation to obtain the corrosion defect characteristics; A fatigue crack growth life prediction unit that inputs the corrosion defects of the to-be-tested corroded steel plate into the trained fatigue crack growth life prediction model to obtain the fatigue crack growth life of the to-be-tested corroded steel plate; A total life prediction unit for the corroded steel plate that obtains the total life of the corroded steel plate according to the crack fatigue initiation life and the fatigue crack growth life of the to-be-tested corroded steel plate.

[0025] The beneficial effects of the present invention are as follows: The intelligent prediction method for the fatigue life of the corroded steel plate of the present invention obtains surface data by measuring the three-dimensional morphology of the corroded steel plate to be measured, which can comprehensively and accurately reflect the actual corrosion condition of the steel plate. Compared with the traditional single-dimensional measurement method, this three-dimensional measurement can capture more detailed information, such as the depth, shape, distribution of corrosion pits, etc., providing more reliable basic data for subsequent life prediction. The separate application of the crack fatigue initiation life prediction model and the fatigue crack propagation life prediction model makes the evaluation of the life of the corroded steel plate more refined. The crack initiation life prediction module combines with the stress concentration coefficient prediction module, considering the key influence of stress concentration on crack initiation, and can accurately predict when and where cracks will start to appear. The fatigue crack propagation life prediction model focuses on the propagation stage after the cracks appear. The separate prediction of the two stages can more accurately grasp the whole process of the corroded steel plate from the beginning of the problem to the final failure. Analyzing the corrosion detail information of the area where the crack initiation is located to obtain the corrosion defect characteristics helps to deeply understand the specific impact of corrosion on the steel plate structure. The present invention specifically predicts and analyzes the life of the corroded steel plate, which can avoid unnecessary premature replacement or over-maintenance, thus saving costs. By accurately understanding the remaining life of the steel plate, maintenance or replacement can be carried out at the most appropriate time, avoiding waste of resources.

[0026] Furthermore, compared with the traditional contact measurement method, the non-contact measurement will not damage the surface of the corroded steel plate, ensuring the authenticity and reliability of the measurement results. At the same time, it can avoid the deformation of corrosion pits or measurement errors caused by contact pressure, further improving the measurement accuracy.

[0027] Furthermore, by inputting the surface data of the corroded steel plate to be measured into the specially trained stress concentration coefficient prediction module and crack initiation life prediction module, the accurate prediction of the crack fatigue initiation life can be realized. The stress concentration coefficient is an important parameter for evaluating the possibility of crack initiation in a structure. Accurately predicting the stress concentration coefficient provides a reliable basis for subsequent calculation of the crack fatigue initiation life. At the same time, determining the crack initiation helps to take targeted maintenance measures in advance to prevent further damage to the structure. Combining the prediction results of the two modules comprehensively considers the surface characteristics and mechanical properties of the corroded steel plate, improving the prediction accuracy. Compared with a single prediction method, this step-by-step prediction method can more carefully analyze the influence of corrosion on the fatigue performance of the steel plate. Description of the Drawings

[0028] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0029] Figure 1 is a flowchart for intelligent prediction of the fatigue life of corroded steel plates; Figure 2 is a structural diagram of an artificial neural network for stress concentration coefficient; Figure 3 is a scanning result diagram of the three-dimensional surface topography of corroded steel plates; Figure 4 is a stress nephogram of the surface of the physical model of corroded steel plates; Figure 5 is a physical information neural network diagram for predicting the crack initiation life of uncorroded steel plates; Figure 6 is a simplified model diagram of irregular corrosion pits and mesoscopic defects at the bottom of corrosion pits of corroded steel plates; Figure 7 is a simplified result diagram obtained from the simplified model of irregular corrosion pits and mesoscopic defects at the bottom of corrosion pits of corroded steel plates; Figure 8 shows the propagation and fusion of multiple cracks at the bottom of corrosion pits during the crack propagation process. Detailed implementation manners

[0030] In order to make the objectives and technical solutions of the present invention clearer and easier to understand, the following will further elaborate on the present invention in combination with the drawings and embodiments. The specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0031] The following will clearly and completely describe the technical solutions of the present invention in combination with the drawings and specific embodiments. Among them, the described embodiments are only some embodiments of the present invention, rather than all embodiments.

[0032] Embodiment 1 As Figures 1-8As shown in the figure, this embodiment provides an intelligent prediction method for the fatigue life of corroded steel plates. The fatigue life of corroded steel plates is divided into crack initiation life and crack propagation life. The life prediction in the crack initiation stage of corroded steel plates includes stress concentration analysis based on the three-dimensional topography characteristics of the corroded surface, prediction of the stress concentration coefficient of the corroded surface based on a deep artificial neural network, and prediction of the crack initiation life based on a physics-informed neural network. The life prediction in the crack propagation stage includes simulation of the crack propagation life based on FRANC3D and prediction of the crack propagation life based on a physics-informed neural network. Based on the predictions in the two stages, the accuracy of the fatigue life prediction of corroded steel plates and the interpretability of the artificial neural network are greatly improved. The following are the specific implementation manners.

[0033] S1. Measure the three-dimensional topography of the to-be-tested corroded steel plate to obtain the surface data of the to-be-tested corroded steel plate; in this embodiment, a three-dimensional non-contact optical profiler is used to record the topography of the corroded surface of the to-be-tested corroded steel plate, perform three-dimensional topography measurement on the surface of the to-be-tested corroded steel plate, and obtain the surface data of the to-be-tested corroded steel plate. The surface data of the to-be-tested corroded steel plate includes corrosion pit details, and the corrosion pit details include the distribution, depth, size, etc. of the surface corrosion pits.

[0034] Specifically, a three-dimensional non-contact optical profiler (3D Profiling) produced by NANOVER Company of the United States is used to perform high-precision three-dimensional scanning on a large number of actually corroded steel plates, capture the geometric characteristics of the surface corrosion, including the depth, width, distribution shape, etc. of the rust pits. The results are as Figure 3 shown; import the results of the three-dimensional scanning into Abaqus to establish a solid model of the corroded steel plate, accurately reflecting its surface corrosion characteristics. Through numerical simulation calculations, study the stress concentration effect of different corroded surfaces. The results are as Figure 4 shown.

[0035] S2. Input the surface data of the to-be-tested corroded steel plate into the trained crack fatigue initiation life prediction model to obtain the crack fatigue initiation life of the to-be-tested corroded steel plate. The crack fatigue initiation life prediction model includes a crack initiation life prediction module and a stress concentration coefficient prediction module; the stress concentration coefficient prediction module obtains the stress concentration coefficient, and the crack initiation life prediction module is used to obtain the crack initiation life according to the stress concentration coefficient. Specifically, it includes: Input the surface data of the to-be-tested corroded steel plate into the trained stress concentration coefficient prediction module to obtain the predicted stress concentration coefficient; multiply the predicted stress concentration coefficient by the applied stress and then input it into the trained crack initiation life prediction module to obtain the crack fatigue initiation life of the to-be-tested corroded steel plate.

[0036] The training processes of the stress concentration coefficient prediction module and the crack initiation life prediction module include: Import the surface data of the rusted steel plate to be measured into the finite element software. Construct a solid model of the rusted steel plate according to the finite element software. Conduct finite element analysis on the rusted surfaces with different morphologies based on the solid model of the rusted steel plate, and calculate the stress concentration coefficient on the steel plate surface based on the solid model of the rusted steel plate. Specifically, conduct multiple finite element analyses on a large number of rusted surface models with different morphologies. These finite element models cover various parameters such as rust depth, pit shape, distribution, and density.

[0037] Obtain a stress concentration coefficient database for the rusted steel plate surface based on the geometric characteristics of the rusted surface and the corresponding stress concentration coefficients. This database contains the geometric characteristics of the rusted surface and the corresponding stress concentration coefficients, providing an adequate data basis for the training of the subsequent data-driven artificial neural network for the stress concentration coefficient of the rusted surface.

[0038] Train a stress concentration coefficient prediction module based on a deep neural network architecture according to the stress concentration coefficient database for the rusted steel plate surface to obtain a trained stress concentration coefficient prediction module. Among them, the stress concentration coefficient prediction module in this embodiment uses a physics-informed neural network model. The trained physics-informed neural network model can quickly obtain the corresponding stress concentration coefficient by inputting any rusted surface characteristics, without the need to perform finite element simulations again each time, greatly improving the calculation efficiency. Generate a database of "rusted surface geometric characteristics and stress concentration coefficients", recording the corresponding relationships between various rusted characteristics (such as the shape and size of rust pits) and the corresponding stress concentration coefficients; through the artificial neural network model as shown in Figure 2 Train the database to establish a module for quickly predicting the stress concentration coefficient based on the scanned rusted characteristics for the subsequent prediction of the crack initiation life of the rusted steel plate.

[0039] According to the loss function corresponding to the neural network in the crack initiation life prediction module of the unrusted steel plate, determine the fatigue life of the unrusted steel plate under different stress amplitudes and the proportion of the crack initiation life in the total fatigue life, and then establish a fatigue crack initiation life database for the unrusted steel plate; Train the crack initiation life prediction module according to the fatigue crack initiation life database of the unrusted steel plate to obtain a trained crack initiation life prediction module.

[0040] Among them, in the physics-informed neural network for the fatigue crack initiation life of the unrusted steel plate, the physical quantity introduced in the loss function is the stress amplitude applied to the component. According to the physical information of the S-N curve, the physical information includes that the relationship between the crack initiation life and the corresponding stress amplitude applied to the rusted steel plate shows a monotonically decreasing trend, and the slope of the S-N curve gradually increases. Therefore, the formula for the physical information loss function adopted by this network is as follows:

[0041]

[0042]

[0043]

[0044] where y true is the actual value of the crack initiation life, and y pred is the predicted value of the crack initiation life, and Penaltys s (1) is used to penalize the case where the prediction function and the first derivative of the stress are positive, and Penaltys s (2) is used to penalize the case where the prediction function and the second derivative of the stress are negative.

[0045] Through the fatigue crack initiation life database of unrusted steel plates, a physics-informed neural network model is trained to enable it to predict the crack initiation life of unrusted steel plates based on the input material property characteristics, load characteristics, etc.

[0046] Specifically, based on the existing experimental data, the crack initiation life of unrusted steel plates under different stress amplitudes is determined, and then a fatigue crack initiation life database of unrusted steel plates is established, including the loading regime (stress amplitude, stress ratio, etc.) and the crack initiation life; a physics-informed neural network model (PINN) is constructed, as Figure 5 shown. This model combines physical laws to obtain the relationship between stress amplitude and crack initiation life from the S-N curve. Based on the crack initiation life database, the neural network model is trained to obtain a tool that can quickly predict the crack initiation life according to different load conditions, etc. The process of predicting the crack initiation life of a specific rusted steel plate using the above module is as Figure 6 shown.

[0047] S3. Analyze the corrosion detail information in the area where the crack initiation occurs to obtain the corrosion defect characteristics.

[0048] Determine the rust pit corresponding to the crack initiation of the rusted steel plate to be measured according to the maximum stress concentration factor. Use three-dimensional scanning technology to finely analyze the bottom of the rust pit corresponding to the crack initiation of the rusted steel plate to be measured, and extract the two-dimensional cross-sectional view of the rust pit. Characterize all the defects at the bottom of the rust pit on the surface of the rusted steel plate, and analyze the quantity, size, distribution, etc. of the local defects at the bottom of the rust pit. According to different rust characteristics, conduct a regression analysis on "the quantity, size, etc. of the local defects at the bottom of the rust pit and the macroscopic characteristics of the rust pit (such as the depth and width of the rust pit)". Factors such as the shape, size, and distribution of the mesoscopic defects at the bottom of the rust pit will make the fatigue crack propagation life show great uncertainty. Among them, mesoscopic defects refer to various discontinuous or abnormal structures that are difficult to directly and clearly distinguish with the naked eye at the bottom of the rust, but are visible at the microscopic scale. These defects will affect the mechanical properties of the steel plate, such as reducing strength and toughness, increasing the degree of stress concentration, thus becoming potential starting points for crack initiation and propagation when the steel plate is loaded, and having an important impact on the fatigue life of the steel plate, etc.

[0049] Establish a correlation model between the macroscopic geometric characteristics of the rust pit and its mesoscopic defect characteristics (such as quantity, size, etc.) through regression analysis.

[0050] According to the correlation model, simplify the mesoscopic defects at the bottom of the rust pit on the surface of the rusted steel plate. Simplify the irregular rust pit into an elliptical regular concave pit, and simplify the mesoscopic defects at the bottom of the rust pit into multiple initial cracks in the crack propagation stage. Obtain a simplified model of the fatigue crack propagation of the rusted steel plate according to the defect characteristics, and obtain the defect characteristics according to the simplified model of the rusted steel plate. The results of the simplified defect characteristics are as Figure 7 shown. Use the finite element model to conduct parametric analysis of the stress state of the simplified model, and use FRANC3D to simulate the crack propagation behavior under cyclic loads under different conditions; the crack propagates until the stress intensity factor of the rusted steel plate exceeds the fracture toughness and fractures, and record the number of cyclic loads at this time as the crack propagation life; collect the crack propagation simulation results of different rusted steel plates, record the rust pit geometric characteristics, crack characteristics, cyclic load characteristics, and crack propagation life corresponding to each model, construct a fatigue crack propagation life database, and train a fatigue crack propagation life prediction model based on the fatigue crack propagation life database.

[0051] S4. Input the rust defects of the rusted steel plate to be measured into the trained fatigue crack propagation life prediction model to obtain the fatigue crack propagation life of the rusted steel plate to be measured.

[0052] Among them, the fatigue crack propagation life prediction model adopts a physics-informed neural network model constructed based on the Paris formula, and the physical information includes the influence of the changes in the maximum stress and the initial crack length on the crack propagation life. The training process of the fatigue crack propagation life prediction model includes: In the area of the corroded steel plate corresponding to the maximum stress concentration factor, obtain the two-dimensional cross-sectional view of the corrosion pit, and determine the corrosion characteristics according to the two-dimensional cross-sectional view. Perform a regression analysis on the corrosion characteristics, simplify the microscopic defects at the bottom of the corrosion pits on the surface of the corroded steel plate, simplify the irregular corrosion pits into elliptical regular pits, and simplify the microscopic defects at the bottom of the corrosion pits into multiple initial cracks in the crack propagation stage to obtain a simplified model of the corroded steel plate and different defect characteristics. Among them, import the simplified model into the Franc3D software, and set the necessary material parameters including elastic modulus, Poisson's ratio, fracture toughness, etc. according to the actual mechanical properties of the material. In addition, set the loading conditions required for fatigue crack propagation analysis according to the working conditions of the fatigue load.

[0053] Use the finite element model to perform parametric analysis of crack propagation on the defect characteristics of the corroded steel plate, and use the Paris formula to simulate the propagation behavior of cracks under cyclic loads under different conditions. In addition, this embodiment also focuses on the change of the stress intensity factor (SIF) and the propagation path of cracks at the bottom of the corrosion pit, and the propagation and fusion of multiple cracks at the bottom of the corrosion pit are as Figure 8 shown.

[0054] The crack propagates until the stress intensity factor of the corroded steel plate exceeds the fracture toughness and fracture occurs. Record the number of cyclic loads at this time as the crack propagation life. Repeat the above simulation process to establish a database including "loading regime, initial crack length, macroscopic characteristics of corrosion pits, etc. and crack propagation life".

[0055] Collect the crack propagation simulation results of different corroded steel plates, record the geometric characteristics of the corrosion pits, crack characteristics, cyclic load characteristics, and crack propagation life corresponding to each model, construct a fatigue crack propagation life database, and train a fatigue crack propagation life prediction model based on the fatigue crack propagation life database.

[0056] Among them, according to the Paris formula, the crack propagation rate is related to the stress intensity factor (ΔK), and the stress intensity factor depends on the maximum stress and crack length. Therefore, the model should learn that for a given initial crack length, an increase in the maximum stress will lead to a decrease in the crack propagation life; for a given maximum stress, an increase in the initial crack length will also lead to a decrease in the crack propagation life; in the fatigue crack propagation life prediction model for predicting the crack propagation life, the physical quantities introduced in the loss function include the cyclic load stress applied to the component and the length of the initial crack at the bottom of the corrosion pit; the formula of the physical information loss function adopted by the fatigue crack propagation life prediction model is as follows:

[0057] Among them,

[0058] In the formula, P1 and P2 are respectively used to penalize the cases where the first-order derivatives of the prediction function with respect to the maximum stress and the initial crack length are positive.

[0059] S5. Obtain the total life of the corroded steel plate according to the crack fatigue initiation life and the fatigue crack propagation life of the corroded steel plate to be measured. Specifically, add the fatigue crack initiation life and the propagation life of the corroded steel plate predicted based on the neural network to obtain the total life of the corroded steel plate.

[0060] For a certain existing corroded steel plate, after obtaining its corrosion characteristics through three-dimensional scanning, predict the crack initiation life and the crack propagation life of the steel plate under different loading conditions respectively, and give the total fatigue life of the corroded steel plate.

[0061] In summary, by measuring the three-dimensional morphology of the corroded steel plate to be measured with a three-dimensional non-contact optical profiler, detailed information including the distribution, depth, size, etc. of the corrosion pits can be obtained accurately, providing an accurate data basis for subsequent analysis. Combining with finite element software to construct a solid model and conduct analysis further improves the understanding of the surface characteristics and stress concentration effect of the corroded steel plate. By measuring the three-dimensional morphology of the corroded steel plate to be measured with a three-dimensional non-contact optical profiler, detailed information including the distribution, depth, size, etc. of the corrosion pits can be obtained accurately, providing an accurate data basis for subsequent analysis. Combining with finite element software to construct a solid model and conduct analysis further improves the understanding of the surface characteristics and stress concentration effect of the corroded steel plate. Analyze the corrosion detail information in the area where the crack initiation occurs, use three-dimensional scanning technology to extract the two-dimensional cross-sectional view of the corrosion pit, characterize and perform regression analysis on the defects at the bottom of the corrosion pit, establish a correlation model and simplify the defect characteristics. This detailed analysis helps to more accurately understand the influence of corrosion defects on the fatigue crack propagation life, providing a reliable basis for the subsequent prediction of the fatigue crack propagation life.

[0062] This embodiment adopts advanced three-dimensional scanning technology and a physics-informed neural network model, greatly improving the prediction efficiency. Without conducting a large number of traditional experiments and complex calculations, only by obtaining the surface data of the corroded steel plate through three-dimensional scanning and inputting it into the trained model, the prediction results of the crack fatigue initiation life, the fatigue crack propagation life, and the total life can be quickly obtained.

[0063] Embodiment 2 This embodiment provides an intelligent prediction system for the fatigue life of a corroded steel plate, which is used to implement the intelligent prediction method for the fatigue life of the corroded steel plate in the above Embodiment 1. The system includes: a data acquisition unit, a crack fatigue initiation life prediction unit, a data analysis unit, a fatigue crack propagation life prediction unit, and a total life prediction unit of the corroded steel plate. Each unit specifically includes: The data acquisition unit measures the three-dimensional morphology of the corroded steel plate to be measured and obtains the surface data of the corroded steel plate to be measured; The crack fatigue initiation life prediction unit inputs the surface data of the rusted steel plate to be measured into the trained crack fatigue initiation life prediction model to obtain the crack fatigue initiation life of the rusted steel plate to be measured; the crack fatigue initiation life prediction model includes a crack initiation life prediction module and a stress concentration coefficient prediction module; the stress concentration coefficient prediction module obtains the stress concentration coefficient, and the crack initiation life prediction module is used to obtain the crack initiation life according to the stress concentration coefficient; The data analysis unit analyzes the rust details information of crack initiation to obtain the rust defect characteristics; The fatigue crack propagation life prediction unit inputs the rust defects of the rusted steel plate to be measured into the trained fatigue crack propagation life prediction model to obtain the fatigue crack propagation life of the rusted steel plate to be measured; The total life prediction unit of the rusted steel plate obtains the total life of the rusted steel plate according to the crack fatigue initiation life and the fatigue crack propagation life of the rusted steel plate to be measured.

Claims

1. An intelligent prediction method for fatigue life of corroded steel plates, characterized in that: include: S1. Measure the three-dimensional morphology of the corroded steel plate to be tested and obtain surface data of the corroded steel plate to be tested; S2, inputting the surface data of the corroded steel plate to be tested into the trained crack fatigue initiation life prediction model to obtain the crack fatigue initiation life of the corroded steel plate to be tested; the crack fatigue initiation life prediction model includes a crack initiation life prediction module and a stress concentration factor prediction module; the stress concentration factor prediction module is used to obtain the stress concentration factor according to the input surface data of the corroded steel plate to be tested, and the crack initiation life prediction module is used to obtain the crack initiation life according to the stress concentration factor; S3, analyzing the corrosion details of the crack initiation to obtain the corrosion defect characteristics; S4, inputting the corrosion defects of the corroded steel plate to be tested into the trained fatigue crack growth life prediction model to obtain the fatigue crack growth life of the corroded steel plate to be tested; S5. According to the crack fatigue initiation life and fatigue crack growth life of the corroded steel plate to be tested, the total life of the corroded steel plate is obtained.

2. The fatigue life intelligent prediction method of corroded steel plate according to claim 1 is characterized in that: The measuring and analyzing of the three-dimensional morphology of the corroded steel plate to be tested to obtain the surface data of the corroded steel plate to be tested specifically includes: A three-dimensional non-contact optical profilometer is used to record the morphology of the corroded surface of the rusted steel plate to be tested, and the three-dimensional morphology of the surface of the rusted steel plate to be tested is measured to obtain the surface data of the rusted steel plate to be tested, wherein the surface data of the rusted steel plate to be tested includes the distribution, depth and size information of the surface rust pits.

3. The fatigue life intelligent prediction method of corroded steel plate according to claim 1 is characterized in that: The method of inputting the surface data of the corroded steel plate to be tested into the trained crack fatigue initiation life prediction model to obtain the crack fatigue initiation life of the corroded steel plate to be tested specifically includes: The surface data of the corroded steel plate to be tested is input into the trained stress concentration factor prediction module to obtain the predicted maximum surface stress concentration factor; After multiplying the predicted maximum surface stress concentration factor by the applied stress, the result is input into the trained crack initiation life prediction module to obtain the crack fatigue initiation life of the corroded steel plate to be tested; The training process of the stress concentration factor prediction module and the crack initiation life prediction module includes: Import the collected surface data of the corroded steel plate to be tested into the finite element software, construct a solid model of the corroded steel plate according to the finite element software, perform finite element analysis on corroded surfaces of different morphologies according to the solid model of the corroded steel plate, and calculate the maximum surface stress concentration factor of the steel plate based on the solid model of the corroded steel plate; construct a database of surface stress concentration factors of corroded steel plates according to the geometric characteristics of the corroded surface and the corresponding maximum surface stress concentration factor; The stress concentration factor prediction module is trained based on the stress concentration factor database of the corroded steel plate surface; Determine the fatigue life of uncorroded steel plates at different stress amplitudes and the proportion of crack initiation life to total fatigue life, and build a fatigue crack initiation life database for uncorroded steel plates; A crack initiation life prediction module is trained according to a fatigue crack initiation life database of uncorroded steel plates. During the training process, the crack initiation life prediction module optimizes the crack initiation life prediction module by using a loss function corresponding to a neural network in the crack initiation life prediction module.

4. The fatigue life intelligent prediction method of corroded steel plate according to claim 3 is characterized in that: The crack initiation life prediction module adopts a physical information neural network model; the physical information corresponding to the crack initiation life prediction module is obtained according to the SN curve, and the physical information includes that the relationship between the crack initiation life and the corresponding stress amplitude applied to the corroded steel plate shows a monotonically decreasing trend, and the slope of the SN curve gradually increases.

5. The fatigue life intelligent prediction method of the corroded steel plate according to claim 3 is characterized in that: The loss function corresponding to the neural network in the crack initiation life prediction module of the uncorroded steel plate includes: In the formula, y true is the actual value of the crack initiation life, y pred is the predicted value of crack initiation life, Penaltys s (1) Penaltys is used to penalize the prediction function when the first-order derivative of stress is positive. s (2) Used to penalize the case where the prediction function and the second derivative of stress are negative.

6. The fatigue life intelligent prediction method of corroded steel plate according to claim 1, characterized in that: Detailed analysis of fatigue crack characteristics of corroded steel plates, including: The rust pit corresponding to the crack initiation of the corroded steel plate to be tested is determined according to the maximum stress concentration factor, a two-dimensional cross-sectional image of the rust pit is obtained, and the macroscopic characteristics of the rust pit on the surface of the corroded steel plate and the microscopic defects at the bottom of the rust pit are characterized; The correlation model between the macroscopic geometric characteristics of rust pits and the microscopic defect characteristics at the bottom of rust pits was established through regression analysis.

7. The fatigue life intelligent prediction method of corroded steel plate according to claim 6, characterized in that: The fatigue crack growth life prediction model is obtained by training with a fatigue crack growth life database. The fatigue crack growth life prediction model training process includes: According to the correlation model, the mesoscopic defects at the bottom of the rust pits on the surface of the corroded steel plate are simplified, the irregular rust pits are simplified into regular elliptical pits, and the mesoscopic defects at the bottom of the rust pits are simplified into multiple initial cracks in the crack propagation stage, and a simplified model of fatigue crack propagation of the corroded steel plate is obtained according to the defect characteristics; The simplified model is subjected to parametric analysis of the stress state using a finite element model, and FRANC3D is used to simulate the crack propagation behavior under cyclic loads under different conditions; The crack propagates until the stress intensity factor of the corroded steel plate exceeds the fracture toughness and fracture occurs. The number of cyclic loads at this time is recorded as the crack propagation life. The crack propagation simulation results of different corroded steel plates were collected, and the rust pit geometric characteristics, crack characteristics, cyclic load characteristics and crack propagation life corresponding to each model were recorded. A fatigue crack propagation life database was constructed, and a fatigue crack propagation life prediction model was trained based on the fatigue crack propagation life database.

8. The fatigue life intelligent prediction method of corroded steel plate according to claim 1, characterized in that: The fatigue crack growth life prediction model adopts physical information based on the Paris formula to optimize the physical information neural network model, and the physical information includes the influence of the maximum stress and the change of the initial crack length on the crack growth life.

9. The fatigue life intelligent prediction method of corroded steel plate according to claim 8, characterized in that: The fatigue crack growth life prediction model loss function is: in, Where P1 and P2 are used to penalize the positive first-order derivatives of the prediction function at the maximum stress and initial crack length, respectively.

10. An intelligent prediction system for fatigue life of corroded steel plates, characterized in that: The method for intelligently predicting fatigue life of a corroded steel plate according to any one of claims 1 to 9 comprises: A data acquisition unit measures the three-dimensional morphology of the corroded steel plate to be tested and acquires surface data of the corroded steel plate to be tested; A crack fatigue initiation life prediction unit inputs the surface data of the corroded steel plate to be tested into a trained crack fatigue initiation life prediction model to obtain the crack fatigue initiation life of the corroded steel plate to be tested; the crack fatigue initiation life prediction model includes a crack initiation life prediction module and a stress concentration factor prediction module; the stress concentration factor prediction module obtains the stress concentration factor, and the crack initiation life prediction module is used to obtain the crack initiation life according to the stress concentration factor; The data analysis unit analyzes the corrosion details of the crack initiation and obtains the corrosion defect characteristics; The fatigue crack growth life prediction unit inputs the corrosion defects of the corroded steel plate to be tested into the trained fatigue crack growth life prediction model to obtain the fatigue crack growth life of the corroded steel plate to be tested; The total life prediction unit of the corroded steel plate obtains the total life of the corroded steel plate according to the crack fatigue initiation life and fatigue crack growth life of the corroded steel plate to be tested.

Images