Method and system for predicting fatigue limit of engineering component

Abstract

The invention relates to the technical field of engineering structures, in particular to an engineering component fatigue limit prediction method and system, and the method comprises the steps: determining a to-be-tested engineering component, and obtaining a plurality of to-be-tested working conditions corresponding to the to-be-tested engineering component under a plurality of preset stress levels; for any to-be-measured working condition, obtaining an infinite life probability corresponding to the to-be-measured working condition based on a preset life prediction model, and further obtaining a plurality of corresponding infinite life probabilities based on the plurality of to-be-measured working conditions; and based on the plurality of to-be-measured working conditions and the plurality of infinite life probabilities corresponding to the plurality of to-be-measured working conditions, obtaining a probability-stress level curve, further obtaining an infinite life critical probability, further obtaining a stress level critical value corresponding to the infinite life critical probability, and taking the stress level critical value as the fatigue limit of the to-be-measured engineering component. According to the method, the stress level corresponding to the critical probability between the infinite life and the finite life is used as the predicted fatigue limit, so that the accuracy of predicting the fatigue limit is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of engineering structures, and in particular to a method and system for predicting fatigue limits of engineering components. Background Art

[0002] In the field of engineering structure technology, fatigue failure is one of the main forms of failure of engineering components, especially reinforced concrete structures. The failure of many types of load-bearing components is closely related to fatigue. Fatigue failure refers to the phenomenon that the material of an engineering structure is damaged under the strain of alternating stress. Fatigue failure is usually manifested as a process in which local microcracks appear inside the reinforced concrete component under the action of cyclic stress and gradually expand, and the steel bars accumulate damage under repeated stress, which eventually leads to the failure of the reinforced concrete component. Fatigue problems are particularly critical in building structures that are subjected to repeated stress during their service life. For example, bridges, high-rise buildings, engineering components in the ocean, and industrial plants, fatigue problems are directly related to the safety and durability of their structures. Therefore, fatigue life prediction of engineering structures has become an important task in the design of modern building structures.

[0003] At present, in the field of engineering structure technology, the SN curve is widely used to obtain the relationship between the stress of a component and its fatigue life, so as to estimate the fatigue limit of the component and predict the fatigue life of the component. The SN curve is widely used in bridges, machinery, aviation and other fields, but its applicability and accuracy in complex engineering have the following shortcomings:

[0004] 1) The SN curve is usually used in conjunction with the linear cumulative damage model (Miner method), ignoring the nonlinear behavior of local stress changes. The error is significant under multi-level stress loads or random stress loads, and it is difficult to accurately describe the fatigue behavior under multi-axial coupled stress;

[0005] 2) The establishment of the SN curve is highly dependent on a large amount of actual test data. In particular, for components with different materials and geometric characteristics, a lot of time and resources are needed to re-test and obtain data, making it difficult to quickly respond to engineering needs. Moreover, the SN curve tends to be an empirical formula derived from actual test data, which cannot reflect the true fatigue mechanism and requires a large amount of measured data and additional correction factors to optimize its results. Summary of the invention

[0006] The present invention aims to provide a method and system for predicting the fatigue limit of an engineering component to solve the above-mentioned technical problems, so as to more accurately estimate the fatigue limit of the component and predict the fatigue life type of the component, thereby providing a simpler and more accurate quantitative decision-making basis for engineering design.

[0007] In order to achieve the above-mentioned purpose, the first aspect of the present invention provides a method for predicting the fatigue limit of an engineering component, comprising the following steps: determining an engineering component to be tested, and obtaining a number of working conditions to be tested corresponding to the engineering component to be tested under a number of preset stress levels; for any of the working conditions to be tested, obtaining the infinite life probability corresponding to the working condition to be tested based on a preset life prediction model, and then obtaining the corresponding several infinite life probabilities based on the several working conditions to be tested; obtaining a probability-stress level curve based on the several working conditions to be tested and the several infinite life probabilities corresponding to the several working conditions to be tested; obtaining an infinite life critical probability based on the probability-stress level curve, and then obtaining a stress level critical value corresponding to the infinite life critical probability, and using the stress level critical value as the fatigue limit of the engineering component to be tested.

[0008] The above-mentioned method and system for predicting the fatigue limit of an engineering component, through the infinite life probability corresponding to the engineering component to be tested at different stress levels, constructs a probability-stress level curve reflecting the relationship between the infinite life probability and the stress level, and converts the fatigue behavior under multi-axis coupled stress into an infinite life probability problem. Based on the probability-stress level curve, the critical probability between the high probability of infinite life and the high probability of finite life of the engineering component to be tested can be found, and the stress level corresponding to the critical probability is used as the predicted fatigue limit. The predicted fatigue limit can reflect the real fatigue mechanism of the engineering component to be tested and improve the accuracy of the predicted fatigue limit. Furthermore, based on the predicted fatigue limit, the fatigue life type of the engineering component to be tested can be predicted, that is, it is predicted that the engineering component to be tested belongs to the infinite life type or the finite life type under a specific stress level, providing a simpler and more accurate quantitative decision-making basis for engineering design.

[0009] In a possible implementation, for any of the working conditions to be tested, the infinite life probability corresponding to the working condition to be tested is obtained based on a preset life prediction model, and then several corresponding infinite life probabilities are obtained based on several of the working conditions to be tested. The life prediction model preset process includes: generating a training data set based on a preset training component group; performing several rounds of iterative training on a preset binary classification algorithm based on the training data set to obtain a life prediction model.

[0010] In this implementation, the binary classification algorithm is iteratively trained through the training data set, so that the binary classification algorithm learns to judge the fatigue life type of the component, and then enables it to predict the infinite life probability corresponding to the working condition to be tested, and obtain the life prediction model. The traditional equivalent strain energy density or simple stress-strain amplitude relationship is converted into a binary classification prediction of fatigue life, so that the life prediction model can reflect the mechanical response information of complex components, thereby improving the accuracy of predicting the fatigue limit of components.

[0011] In a possible implementation, generating a training data set based on a preset training component group includes: for any training component in the training component group, obtaining a first fatigue life of the training component and a structural performance characteristic of the training component; obtaining a structural performance characteristic set based on all the structural performance characteristics; classifying all the training components based on the first fatigue life according to a preset benchmark fatigue life to obtain a classification label set; and generating a training data set based on the structural performance characteristic set and the classification label set.

[0012] It should be noted that the first fatigue life is obtained by applying a cyclic load of constant amplitude to the training component on a dedicated fatigue testing machine. This type of test method tests the fatigue life of the training component by controlling the cyclic characteristics of the external load or displacement, such as frequency, peak stress, minimum stress and stress ratio. During the test, the fatigue life of the component is recorded by detecting the state and number of cycles from the generation of cracks, crack expansion to final destruction during the loading process. After obtaining the first fatigue life of the training component, the classification category of the training component is determined by comparing the size of the benchmark fatigue life and the first fatigue life. Among them, the benchmark fatigue life is set to 2×106 cycles in this implementation method based on the widely accepted judgment criteria in fatigue life standards and factory time. The structural performance characteristics include the stress-strain state of the steel bars, the stress-strain field of the concrete compression zone and the key parameters of the cross-sectional internal force distribution.

[0013] In this implementation, by integrating the key data of the above-mentioned structural performance feature set and classification label set, the above-mentioned secondary features derived from mechanical calculations and the original variables are combined to form a training data set, thereby enhancing the expression ability of the training data set for the mechanical properties and fatigue life of the components.

[0014] In a possible implementation, all the training components are classified based on the first fatigue life according to a preset benchmark fatigue life to obtain a classification label set, including: for any of the training components, if the first fatigue life corresponding to the training component is less than the benchmark fatigue life, the training component is classified into a finite life category; if the first fatigue life corresponding to the training component is greater than or equal to the benchmark fatigue life, the training component is classified into an infinite life category; and a classification label set is obtained based on the finite life category and the infinite life category.

[0015] In this implementation, the reference fatigue life is determined as 2×106 cycles in this implementation according to the widely accepted discrimination criteria in fatigue life standards and factory time. If the first fatigue life corresponding to the training component is less than 2×106 cycles, the training component is classified into a finite life category, and if the first fatigue life corresponding to the training component is greater than or equal to 2×106 cycles, the training component is classified into an infinite life category. The training component is binary-classified by the above method, that is, the training component is classified into a finite life category and an infinite life category, so that when the preset binary classification algorithm is subsequently iterated for several rounds based on the training data set, the binary classification algorithm can learn to judge the fatigue life category of the component, so that the obtained life prediction model can predict the probability that the engineering component to be tested belongs to the infinite life category and the probability that it belongs to the finite life category, and then obtain the infinite life probability.

[0016] Specifically, in a possible implementation, when the infinite life probability obtained by the life prediction model is higher than 0.5, it is considered that the engineering component to be tested belongs to the infinite life category at this stress level; when the infinite life probability obtained by the life prediction model is less than 0.5, it is considered that the engineering component to be tested belongs to the finite life category at this stress level. Through the above classification prediction process, the fatigue life category and corresponding prediction probability corresponding to the engineering component to be tested at different stress levels can be obtained, which improves the accuracy of the subsequent establishment of the probability-stress level curve and the acquisition of the predicted fatigue limit. Based on the predicted fatigue limit obtained by this method, the fatigue life state of the engineering component to be tested at any stress level can be evaluated. When the actual stress level is less than the predicted fatigue limit, the engineering component to be tested is in the infinite life category. When the actual stress level is greater than the predicted fatigue limit, the engineering component to be tested is in the finite life category, thereby providing a simpler and more accurate quantitative decision-making basis for engineering design.

[0017] In a possible implementation, the preset binary classification algorithm is subjected to several rounds of iterative training based on the training data set to obtain a life prediction model, including: for any round of the iterative training; the binary classification algorithm is repeatedly subjected to layered nested cross-validation training based on the training data set to obtain a current evaluation index and a current feature importance sequence, and then the binary classification algorithm is updated based on the current evaluation index and the feature importance sequence; when the current evaluation index meets a preset prediction error threshold, an optimal feature subset is obtained based on the current feature importance sequence, and then the binary classification algorithm corresponding to the optimal feature subset is used as a life prediction model.

[0018] In this implementation, repeated stratified nested cross-validation is used for model training and evaluation. This method can reflect the evaluation index (prediction error) of the algorithm model during the training process, and calculate the importance score of the feature parameters during the iterative training process, so as to obtain the current feature importance sequence corresponding to the feature parameters; finally, when the evaluation index meets the preset prediction error threshold, that is, the evaluation index reaches the minimum prediction error, based on the cumulative importance scores of the features in all repeated validations, the feature importance ranking is generated and the optimal feature subset with the highest feature importance is determined to obtain the life prediction model, thereby ensuring the reliability of the feature parameter selection and improving the accuracy of the life prediction model.

[0019] In one possible implementation, for any round of the iterative training, the binary classification algorithm is repeatedly trained on the training data set to obtain a current evaluation index and a current feature importance sequence, and then the binary classification algorithm is updated based on the current evaluation index and the feature importance sequence, including: obtaining a feature parameter sequence based on the current feature importance sequence, and then eliminating feature parameters in the feature parameter sequence that meet preset feature elimination rules.

[0020] In this implementation, a recursive feature elimination method is introduced in the model training process. Specifically, based on the feature importance ranking and the preset feature elimination rules, the feature parameters with the lowest importance are gradually removed in each round of iterative training, and the model is retrained using the remaining feature parameters, so that the feature parameters with low predictive ability can be eliminated, and the feature subset with the most predictive ability can be identified and screened, and finally the optimal feature subset with the smallest prediction error is obtained, which improves the accuracy of the final feature subset and life prediction model.

[0021] In a possible implementation, the determining of the engineering component to be tested and obtaining a number of working conditions to be tested corresponding to the engineering component to be tested at a number of preset stress levels include: obtaining the geometric information of the engineering component to be tested; for any of the stress levels, obtaining the internal force state and strain field characteristic index of the engineering component to be tested at the stress level; constructing a characteristic matrix based on the internal force state corresponding to the stress level, the strain field characteristic index corresponding to the stress level, and the geometric information of the engineering component to be tested, and using the characteristic matrix as the working condition to be tested corresponding to the engineering component to be tested at the stress level.

[0022] In this implementation, a series of stress levels are defined for the engineering component to be tested, and the internal mechanical response calculation is performed for the engineering component to be tested at each stress level. The internal mechanical response calculation includes the internal force state and strain field characteristic indicators of the engineering component to be tested at the stress level, such as the distribution of bending moment, shear force and normal stress on the cross section, the maximum principal strain and strain energy density. The above key characteristic variables at a stress level are extracted to form a characteristic matrix corresponding to the stress level, and the characteristic matrix is ​​used as the working condition to be tested corresponding to the engineering component to be tested at the stress level, as the input of the life prediction model. The above parameters cover the nonlinear characteristics that reflect the coupling between the geometric characteristics and material properties of the engineering component to be tested, and improve the accuracy of the life prediction model in predicting the fatigue limit of the engineering component to be tested.

[0023] In a possible implementation, the method for predicting fatigue limits of engineering components also includes: determining a number of the engineering components to be tested, and obtaining a number of the fatigue limits corresponding to the engineering components to be tested; for any of the engineering components to be tested, obtaining a fatigue life type corresponding to the engineering component to be tested based on the fatigue limit corresponding to the engineering component to be tested, and then obtaining a number of corresponding fatigue life types based on the engineering components to be tested; constructing a feature space based on the engineering components to be tested and the fatigue life types corresponding to the engineering components to be tested; performing boundary analysis on the feature space to obtain a fatigue design equation.

[0024] In this implementation, in order to obtain a universal fatigue equation that reflects the relationship between stress and fatigue life, the sample space is expanded. Specifically, several workpieces to be tested are also obtained, and the fatigue life type of each workpiece to be tested is determined using the above method. In the parameter analysis stage, through the projection and mapping of the multidimensional feature space, a feature space is constructed in the Cartesian coordinate system based on several of the engineering components to be tested and several of the fatigue life types corresponding to the engineering components to be tested, and then the boundary analysis of the feature space is performed to determine the boundary surface that can effectively distinguish the finite life category and the infinite life category of the engineering components to be tested. The boundary surface contains the coupling relationship between the key parameters that affect the fatigue life, and provides a theoretical basis for the construction of the fatigue design equation. Finally, the fatigue design equation is obtained based on the boundary surface. The fatigue design equation can reflect the relationship between stress and fatigue life, improve the accuracy of predicting the fatigue limit, and provide a simpler and more accurate quantitative decision-making basis for engineering design.

[0025] Specifically, the feature space is constructed in the Cartesian coordinate system based on the plurality of engineering components to be measured and the plurality of fatigue life types corresponding to the plurality of engineering components to be measured, including constructing the feature space based on the feature parameters related to the engineering components to be measured and the fatigue life types corresponding to the engineering components to be measured. According to different selected feature parameters, different feature spaces can be established to reflect the relationship between the fatigue life types of the engineering components to be measured and different feature parameters.

[0026] In a possible implementation, an optimal feature subset of the engineering component to be measured is obtained based on the life prediction model, and several most important feature parameters of the engineering component to be measured are selected based on the optimal feature subset as feature parameters for establishing a feature space.

[0027] The second aspect of the present invention provides a fatigue limit prediction system for an engineering component, comprising a working condition acquisition module, an infinite life probability prediction module, a curve acquisition module and a fatigue limit prediction module, wherein: the working condition acquisition module is used to determine the engineering component to be tested, and obtain a number of working conditions to be tested corresponding to the engineering component to be tested under a number of preset stress levels; the infinite life probability prediction module is used to obtain the infinite life probability corresponding to the working condition to be tested for any of the working conditions to be tested based on a preset life prediction model, and then obtain the corresponding several infinite life probabilities based on the several working conditions to be tested; the curve acquisition module is used to obtain a probability-stress level curve based on the several working conditions to be tested and the several infinite life probabilities corresponding to the several working conditions to be tested; the fatigue limit prediction module is used to obtain the infinite life critical probability based on the probability-stress level curve, and then obtain the stress level critical value corresponding to the infinite life critical probability, and use the stress level critical value as the fatigue limit of the engineering component to be tested.

[0028] In a possible implementation, in the infinite life probability prediction module, for any of the working conditions to be tested, the infinite life probability corresponding to the working condition to be tested is obtained based on a preset life prediction model, and then several corresponding infinite life probabilities are obtained based on several of the working conditions to be tested. The life prediction model preset process includes: generating a training data set based on a preset training component group; performing several rounds of iterative training on a preset binary classification algorithm based on the training data set to obtain a life prediction model.

[0029] A fatigue limit prediction method and system for an engineering component provided by the present invention has at least the following advantages over the prior art:

[0030] The present invention utilizes the flexibility of the binary classification algorithm in feature selection and combination, and iteratively trains the binary classification algorithm so that the obtained life prediction model can predict the infinite life probability of the engineering component to be tested under different stress levels, and obtain the fatigue limit of the engineering component to be tested in combination with the probability threshold, and converts the traditional equivalent strain energy density or simple stress-strain amplitude relationship into a binary classification prediction of fatigue life, so that the life prediction model can reflect the mechanical response information of complex components, and improves the accuracy of predicting the fatigue limit of components; the present invention also expands the sample space, and constructs a feature space in a Cartesian coordinate system based on a number of the engineering components to be tested and a number of fatigue life types corresponding to the engineering components to be tested, and then performs boundary analysis on the feature space to determine the boundary surface that can effectively distinguish the finite life category and the infinite life category of the engineering component to be tested, so as to construct a new fatigue design equation through high-level data projection and mapping, thereby improving the accuracy of predicting fatigue limit, and providing a simpler and more accurate quantitative decision-making basis for engineering design. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a schematic diagram of a process flow of a method for predicting fatigue limit of an engineering component provided by an embodiment of the present invention;

[0032] Figure 2 It is a schematic diagram of a process of repeated hierarchical nested cross-validation training provided by an embodiment of the present invention;

[0033] Figure 3 It is a flowchart of repeated hierarchical nested cross-validation training combined with recursive feature elimination provided by an embodiment of the present invention;

[0034] Figure 4 It is a structural schematic diagram of a fatigue limit prediction system for an engineering component provided by an embodiment of the present invention;

[0035] Figure 5 is a stereoscopic diagram of an FRP reinforced reinforced concrete RC beam provided by an embodiment of the present invention;

[0036] Figure 6 It is a bottom view of an FRP reinforced reinforced concrete RC beam provided by an embodiment of the present invention;

[0037] Figure 7 is an F1 score frequency distribution diagram provided by an embodiment of the present invention;

[0038] Figure 8 is a schematic diagram of probability-stress level curves of four engineering components to be tested provided by an embodiment of the present invention;

[0039] Fig. 9is a schematic diagram of the average relative error between the predicted fatigue limit and the actual test result provided by the embodiment of the present invention;

[0040] Fig.10 It is a schematic diagram of boundary analysis in a feature space provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0041] The present invention will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that the embodiments and features in the embodiments of the present invention can be combined with each other without conflict.

[0042] The following detailed descriptions are all exemplary descriptions, and are intended to provide further detailed descriptions of the present invention. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments, and are not intended to limit the present application; the terms "including" and "having" in the specification and claims of the present application and the above-mentioned drawings and any variations thereof are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0043] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0044] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0045] Before describing the present application in detail with reference to the accompanying drawings and in combination with the embodiments, the terms and application scenarios involved in the present application are first described.

[0046] In the field of engineering structures, reinforced concrete (RC) structures are the main widely used structural form. In order to improve the bearing capacity and fatigue resistance of RC structures, the technology of reinforcing them with fiber reinforced plastics (FRP) has gradually been widely accepted, among which external FRP reinforcement technology is frequently used as an efficient reinforcement method. Especially under fatigue loads, FRP reinforcement can effectively improve the stress distribution and fatigue resistance of RC structures.

[0047] The data used in the current fatigue design specifications are mainly derived from pure steel or unreinforced RC beams. These test data cannot reflect the true performance of FRP-reinforced RC beams under fatigue loads. Since FRP reinforcement significantly changes the stress distribution of RC beams and inhibits crack propagation, their fatigue performance shows great differences. The existing specifications directly apply the data of pure steel or unreinforced RC beams for design, lacking adjustments and optimizations for FRP-reinforced RC beams, which may lead to overly conservative designs or increased failure risks.

[0048] Most of the existing SN fatigue life curves are based on single variable modeling, that is, directly linking the stress range with the fatigue life, and fail to fully consider the comprehensive influence of multivariate factors (such as material properties, loading conditions, environmental factors, etc.) on fatigue performance. In addition, due to the high cost of fatigue testing and the difficulty of sampling, the experimental data scale of the SN curve currently used is small, the sample coverage is limited, and it is difficult to fully reflect the various working conditions that may be encountered in actual engineering. This situation has caused the accuracy and applicability of fatigue performance prediction to be seriously affected in FRP strengthened beams.

[0049] In order to solve the above technical problems, see Figure 1 The first aspect of the present invention provides a method for predicting fatigue limit of an engineering component, comprising the following steps:

[0050] S101, determining an engineering component to be tested, and obtaining a plurality of working conditions to be tested corresponding to the engineering component to be tested under a plurality of preset stress levels;

[0051] S102, for any of the working conditions to be tested, obtaining an infinite life probability corresponding to the working condition to be tested based on a preset life prediction model, and then obtaining a number of corresponding infinite life probabilities based on a number of the working conditions to be tested;

[0052] S103, obtaining a probability-stress level curve based on a plurality of the working conditions to be tested and a plurality of the infinite life probabilities corresponding to the working conditions to be tested;

[0053] S104, obtaining the infinite life critical probability based on the probability-stress level curve, and then obtaining the stress level critical value corresponding to the infinite life critical probability, and using the stress level critical value as the fatigue limit of the engineering component to be tested.

[0054] The above-mentioned method and system for predicting the fatigue limit of an engineering component constructs a probability-stress level curve reflecting the relationship between the infinite life probability and the stress level through the infinite life probability corresponding to the engineering component to be tested at different stress levels. Based on the probability-stress level curve, the critical probability between the high probability of infinite life and the high probability of finite life of the engineering component to be tested can be found, so that the stress level corresponding to the critical probability is used as the predicted fatigue limit, which improves the accuracy of predicting the fatigue limit and provides a simpler and more accurate quantitative decision-making basis for engineering design. The above-mentioned method for predicting fatigue limit based on probability can not only accurately estimate the fatigue limit, but also evaluate the fatigue life state of the engineering component to be tested at any stress level, providing a quantitative decision-making basis for engineering design.

[0055] In a possible embodiment, for any of the working conditions to be tested, the infinite life probability corresponding to the working condition to be tested is obtained based on a preset life prediction model, and then several corresponding infinite life probabilities are obtained based on several of the working conditions to be tested. The life prediction model preset process includes: generating a training data set based on a preset training component group; performing several rounds of iterative training on a preset binary classification algorithm based on the training data set to obtain a life prediction model.

[0056] Specifically, this implementation adopts a multi-algorithm parallel strategy, selects multiple mainstream machine learning algorithms including support vector machine (SVM), random forest (RF), XGBoost and deep neural network, presets binary classification algorithms and performs iterative training, and finally obtains multiple life prediction models. The one with the highest accuracy is selected from the multiple life prediction models as the final optimal classifier.

[0057] In this embodiment, the binary classification algorithm is iteratively trained through a training data set, so that the binary classification algorithm learns to determine the fatigue life type of the component, and then enables it to predict the infinite life probability corresponding to the working condition to be tested, and obtain a life prediction model, which converts the traditional equivalent strain energy density or simple stress-strain amplitude relationship into a binary classification prediction of fatigue life, so that the life prediction model can reflect the mechanical response information of complex components, thereby improving the accuracy of predicting the fatigue limit of the component.

[0058] In a possible embodiment, generating a training data set based on a preset training component group includes: for any training component in the training component group, obtaining a first fatigue life of the training component and a structural performance characteristic of the training component; obtaining a structural performance characteristic set based on all the structural performance characteristics; classifying all the training components based on the first fatigue life according to a preset benchmark fatigue life to obtain a classification label set; and generating a training data set based on the structural performance characteristic set and the classification label set.

[0059] It should be noted that the first fatigue life is obtained by applying a cyclic load of constant amplitude to the training component on a dedicated fatigue testing machine. This type of test method tests the fatigue life of the training component by controlling the cyclic characteristics of the external load or displacement, such as frequency, peak stress, minimum stress and stress ratio. During the test, the fatigue life of the component is recorded by detecting the state and number of cycles from the generation of cracks, crack expansion to final destruction during the loading process. After obtaining the first fatigue life of the training component, the classification category of the training component is determined by comparing the size of the benchmark fatigue life and the first fatigue life. Among them, the benchmark fatigue life is determined to be 2×106 cycles in this embodiment based on the widely accepted judgment criteria in fatigue life standards and factory time. The structural performance characteristics include the stress-strain state of the steel bars, the stress-strain field of the concrete compression zone and the key parameters of the cross-sectional internal force distribution.

[0060] Specifically, the cross-sectional internal force distribution includes axial force, bending moment and shear force. In the process of applying a cyclic load of a constant amplitude to the training component on a dedicated fatigue testing machine, it is necessary to obtain the geometric information of the training component, including the component size, such as the cross-sectional width, height and component length; obtain the material mechanical properties of the training component, including the elastic modulus, Poisson's ratio, tensile and compressive strength, stress-strain curve or constitutive model of concrete, the yield strength, ultimate strength, elastic modulus, stress-strain relationship of steel bars, elastic modulus, ultimate stress-strain value of FRP materials and the interface characteristics between it and concrete / bonding layer and the mechanical parameters of other auxiliary materials (structural steel bars, bonding materials, etc.); obtain the load and boundary conditions of the training component, including: the type and size of the applied external load, the loading method and loading frequency, the boundary support form, and possible environmental factors. Among them, the external load type includes static load, dynamic load and fatigue load; the boundary support form includes hinged support, fixed support and simple support; the environmental factors include the influence of temperature, corrosion and humidity on the stiffness and strength of the training component material, and the environmental factors can be used as correction parameters.

[0061] In this embodiment, by integrating the key data of the above-mentioned structural performance feature set and classification label set, the above-mentioned secondary features derived from mechanical calculations and the original variables are combined to form a training data set, thereby enhancing the expression ability of the training data set for the mechanical properties and fatigue life of the components.

[0062] In a possible embodiment, all the training components are classified based on the first fatigue life according to the preset reference fatigue life to obtain a classification label set, including: for any of the training components, if the first fatigue life corresponding to the training component is less than the reference fatigue life, the training component is classified into the finite life category; if the first fatigue life corresponding to the training component is greater than or equal to the reference fatigue life, the training component is classified into the infinite life category; and a classification label set is obtained based on the finite life category and the infinite life category.

[0063] In this embodiment, the reference fatigue life is determined to be 2×10^6 cycles according to the fatigue life standard and the widely accepted discrimination criterion in the factory time. If the first fatigue life corresponding to the training component is less than 2×10^6 cycles, the training component is classified into the finite life category; if the first fatigue life corresponding to the training component is greater than or equal to 2×10^6 cycles, the training component is classified into the infinite life category. By classifying the training components through the above method, that is, classifying the training components into the finite life category and the infinite life category, when performing several rounds of iterative training on the preset binary classification algorithm based on this training data set, the binary classification algorithm can learn to judge the fatigue life category of the components, so that the obtained life prediction model can predict the probability that the engineering component to be tested belongs to the infinite life category and the probability that it belongs to the finite life category, and further obtain the infinite life probability.

[0064] Specifically, in a possible embodiment, when the infinite life probability obtained by the life prediction model is higher than 0.5, it is considered that the engineering component to be tested belongs to the infinite life category at this stress level; when the infinite life probability obtained by the life prediction model is less than 0.5, it is considered that the engineering component to be tested belongs to the finite life category at this stress level. Through the above classification and prediction process, the fatigue life category and the corresponding prediction probability of the engineering component to be tested under different stress levels can be obtained, improving the accuracy of establishing the probability-stress level curve and obtaining the predicted fatigue limit subsequently.

[0065] Based on the predicted fatigue limit obtained by this method, the fatigue life state of the engineering component to be tested at any stress level can be evaluated. When the actual stress level is less than the predicted fatigue limit, the engineering component to be tested is of the infinite life category; when the actual stress level is greater than the predicted fatigue limit, the engineering component to be tested is of the finite life category, thus providing a simpler and more accurate quantitative decision-making basis for engineering design.

[0066] In a possible embodiment, the preset binary classification algorithm is subjected to several rounds of iterative training based on the training data set to obtain a life prediction model, including: for any round of the iterative training; the binary classification algorithm is repeatedly subjected to layered nested cross-validation training based on the training data set to obtain a current evaluation index and a current feature importance sequence, and then the binary classification algorithm is updated based on the current evaluation index and the feature importance sequence; when the current evaluation index meets a preset prediction error threshold, an optimal feature subset is obtained based on the current feature importance sequence, and then the binary classification algorithm corresponding to the optimal feature subset is used as the life prediction model.

[0067] Specifically, see Figure 2 , Figure 2 It is a flowchart of a repeated stratified nested cross-validation training provided by an embodiment of the present invention. In this embodiment, the feature set (structural performance feature set) and the classification label (classification label set) are first integrated to form a training data set, and a stratified sampling strategy is used to ensure that the sample ratio of each classification category in the data set is balanced, laying the foundation for the subsequent cross-validation process and avoiding the influence of uneven sample distribution on the model training effect.

[0068] Among them, the repeated stratified nested cross-validation training includes an outer loop and an inner loop, the outer loop is also the outer cross-validation, and the inner loop is also the inner cross-validation. In the outer loop, the training set is randomly divided into 10 parts to ensure that the category ratio of each subset is consistent; one of them is taken as the test set each time, and the remaining nine are taken as the training set, and the optimal hyperparameters are used for training, and then the average score of all test sets is calculated, and the importance of each feature parameter and the importance of the feature subset are obtained based on the average score. The outer loop is repeated 50 times, and the impact of randomness is reduced by the average results of multiple verifications to improve the stability and reliability of the results.

[0069] For each training set obtained from the outer loop, a 10-fold stratified cross-validation is performed in the inner loop. The cycle is repeated 50 times, and the average performance of each feature subset in the inner loop is counted. The hyperparameters are adjusted based on the statistical results. Then, based on the hyperparameter adjustment results obtained from the inner loop, the optimal feature combination and the corresponding model parameters are selected and returned to the test set of the outer loop for independent evaluation, so as to cumulatively record the number of times each feature parameter is selected and the importance score.

[0070] Finally, based on the accumulated importance scores of parameter features in all repeated stratified nested cross-validation training, a feature importance ranking is generated and the optimal feature subset is determined, and the binary classification algorithm corresponding to the optimal feature subset is used as a life prediction model.

[0071] In this embodiment, repeated stratified nested cross-validation is used for model training and evaluation. This method can reflect the evaluation index (prediction error) of the algorithm model during the training process, and calculate the importance score of the feature parameters during the iterative training process, so as to obtain the current feature importance sequence corresponding to the feature parameters; finally, when the evaluation index meets the preset prediction error threshold, that is, when the evaluation index reaches the minimum prediction error, based on the cumulative importance scores of the features in all repeated validations, the feature importance ranking is generated and the optimal feature subset with the highest feature importance is determined to obtain the life prediction model, thereby ensuring the reliability of the feature parameter selection and improving the accuracy of the life prediction model. The above-mentioned double-layer validation framework of repeated stratified nested cross-validation training effectively prevents model overfitting and data leakage problems.

[0072] In a possible embodiment, for any round of the iterative training; the binary classification algorithm is repeatedly trained with layered nested cross-validation based on the training data set to obtain a current evaluation index and a current feature importance sequence, and then the binary classification algorithm is updated based on the current evaluation index and the feature importance sequence, including: obtaining a feature parameter sequence based on the current feature importance sequence, and then eliminating feature parameters in the feature parameter sequence that meet preset feature elimination rules.

[0073] Specifically, see Figure 3 , Figure 3 It is a flow chart of a repeated stratified nested cross-validation training combined with recursive feature elimination provided by an embodiment of the present invention. In this embodiment, recursive feature elimination is introduced in the process of repeated stratified nested cross-validation training, and the feature parameters with the lowest importance are gradually removed according to the importance ranking of the feature parameters obtained in each round of repeated stratified nested cross-validation training, until the repeated stratified nested cross-validation training obtains the optimal feature subset, that is, the optimal performance is obtained, and the binary classification algorithm corresponding to the optimal feature subset is used as the life prediction model.

[0074] In this embodiment, a recursive feature elimination method is introduced in the model training process. Specifically, based on the feature importance ranking and the preset feature elimination rules, the feature parameters with the lowest importance are gradually removed in each round of iterative training, and the model is retrained using the remaining feature parameters, so that the feature parameters with low predictive ability can be eliminated, and the feature subset with the most predictive ability can be identified and screened, and finally the optimal feature subset with the smallest prediction error is obtained, which improves the accuracy of the final feature subset and life prediction model.

[0075] In a possible embodiment, the determining of the engineering component to be tested and obtaining a number of working conditions to be tested corresponding to the engineering component to be tested at a number of preset stress levels include: obtaining geometric information of the engineering component to be tested; for any of the stress levels, obtaining the internal force state and strain field characteristic index of the engineering component to be tested at the stress level; constructing a characteristic matrix based on the internal force state corresponding to the stress level, the strain field characteristic index corresponding to the stress level, and the geometric information of the engineering component to be tested, and using the characteristic matrix as the working condition to be tested corresponding to the engineering component to be tested at the stress level.

[0076] In this embodiment, a series of stress levels are defined for the engineering component to be tested, and the internal mechanical response calculation is performed for the engineering component to be tested at each stress level, and the internal mechanical response calculation includes the internal force state and strain field characteristic indicators of the engineering component to be tested at the stress level, such as the distribution of bending moment, shear force and normal stress on the cross section, the maximum principal strain and strain energy density. The above key characteristic variables at a stress level are extracted to form a characteristic matrix corresponding to the stress level, and the characteristic matrix is ​​used as the working condition to be tested corresponding to the engineering component to be tested at the stress level, as the input of the life prediction model. The above parameters cover the nonlinear characteristics that reflect the coupling between the geometric characteristics and material properties of the engineering component to be tested, and improve the accuracy of the life prediction model in predicting the fatigue limit of the engineering component to be tested.

[0077] Specifically, the acquisition of several working conditions to be tested corresponding to the engineering component to be tested under several preset stress levels is a process of mapping the engineering component to be tested to different stress levels by setting a series of stress gradient values. The above process includes acquiring the geometric dimensions of the engineering component to be tested, such as the cross-sectional dimensions and span; acquiring the steel bar configuration of the engineering component to be tested, such as the steel bar diameter, the number of steel bars and the steel bar position; acquiring the FPR reinforcement layer parameters of the engineering component to be tested, such as the thickness, width and laying position of the engineering component to be tested; acquiring the material mechanical properties of the engineering component to be tested, including the strength and elastic modulus of the engineering component to be tested; acquiring the boundary conditions and loading information of the engineering component to be tested, including the support method of the engineering component to be tested, the loading method of the engineering component to be tested and the preset load gradient sequence; acquiring the mechanical analysis model of the engineering component to be tested, including the cross-sectional analysis model or finite element model of the engineering component to be tested; acquiring the stress-strain relationship of the engineering component to be tested; and acquiring the load-stress conversion relationship of the engineering component to be tested.

[0078] In a possible embodiment, based on the above process, 100 stress gradient values ​​are set for the engineering component to be tested, which are distributed in sequence from 0 to the ultimate load range of the engineering component to be tested; through this process, 100 working conditions to be tested corresponding to different stress levels of the engineering component to be tested can be formed, thereby constructing an extended sample set to provide data support for subsequent feature extraction and prediction.

[0079] In a possible embodiment, the working condition to be tested is input into the life prediction model to determine whether the engineering component to be tested belongs to the infinite fatigue life category under the working condition to be tested. Specifically, the life prediction model will output the probability that the engineering component to be tested belongs to the finite life category and the probability that it belongs to the infinite life category, and then obtain the fatigue life classification results and infinite life prediction probabilities corresponding to 100 groups of working conditions to be tested. Then, the infinite life prediction probability P and the corresponding stress level S of the engineering component to be tested are plotted as a scatter plot in the Cartesian coordinate system, and the data points are smoothed by interpolation or curve fitting to generate a smooth and continuous probability-stress level curve (PS curve). Through the probability-stress level curve, the regularity of the fatigue performance of the component with the load level can be intuitively displayed. The stress level corresponding to the critical probability is found on the probability-stress level curve, and it is defined as the fatigue limit of the engineering component to be tested. The data point corresponding to the critical probability represents the critical load at which the engineering component to be tested belongs to the infinite fatigue life category under a given stress level. In this embodiment, the value of the critical probability is specifically P=0.5.

[0080] Finally, the conditions of the engineering component to be tested under different actual working conditions can be determined based on the obtained fatigue limit. The peak load of the engineering component to be tested in the actual working condition is compared with the fatigue limit. If the peak load is less than the fatigue limit, the engineering component to be tested has an infinite fatigue life; if the peak load is greater than the fatigue limit, the engineering component to be tested has a finite fatigue life. In addition, by combining the probability-stress level curve, it can also provide designers with the probability distribution of the component life state under different stress levels, providing a quantitative reference for parameter selection and safety margin design in engineering decision-making.

[0081] In a possible embodiment, the method for predicting fatigue limits of engineering components also includes: determining a number of the engineering components to be tested, and obtaining a number of the fatigue limits corresponding to the engineering components to be tested; for any of the engineering components to be tested, obtaining a fatigue life type corresponding to the engineering component to be tested based on the fatigue limit corresponding to the engineering component to be tested, and then obtaining a number of corresponding fatigue life types based on the engineering components to be tested; constructing a feature space based on the engineering components to be tested and the fatigue life types corresponding to the engineering components to be tested; performing boundary analysis on the feature space to obtain a fatigue design equation.

[0082] Specifically, by obtaining enough engineering components to be tested, the probability-stress level curve corresponding to each engineering component to be tested can be drawn one by one, and then several fatigue limits corresponding to several engineering components to be tested can be obtained. After comparing the original stress level of the engineering component to be tested with its fatigue limit, the fatigue life type of the engineering component to be tested can be determined, that is, whether the engineering component to be tested belongs to the finite life category or the infinite life category. Furthermore, the data component feature space is combined with multiple parameter dimensions, and a boundary surface or dividing line that can effectively distinguish between finite life components and infinite life components is extracted in the space. The boundary surface or dividing line comprehensively reflects the coupling relationship between multi-dimensional parameters, and provides an important basis for fatigue design equations.

[0083] Specifically, the feature space of the data component is constructed by combining multiple parameter dimensions, that is, in a Cartesian coordinate system, based on a number of the engineering components to be tested and a number of the fatigue life types corresponding to the engineering components to be tested, including constructing a feature space based on the characteristic parameters related to the engineering components to be tested and the fatigue life types corresponding to the engineering components to be tested. The multiple parameter dimensions refer to multiple characteristic parameters; different characteristic spaces can be established according to different selected characteristic parameters, reflecting the relationship between the fatigue life types of the engineering components to be tested and different characteristic parameters.

[0084] In a possible embodiment, an optimal feature subset of the engineering component to be measured is obtained based on the life prediction model, and several most important feature parameters of the engineering component to be measured are selected based on the optimal feature subset as feature parameters for establishing a feature space.

[0085] In a possible embodiment, a multi-source data fusion strategy is used to expand the sample space. Through numerical simulation, such as nonlinear finite element analysis, deep learning generation model and physical rule-based sample generation method, a high-quality virtual sample set is constructed to replace several engineering components to be tested. These virtual samples have consistent data statistical characteristics with the real engineering components to be tested, and their physical constraints meet actual conditions.

[0086] In this embodiment, in order to obtain a universal fatigue equation that reflects the relationship between stress and fatigue life, the sample space is expanded. Specifically, several workpieces to be tested are also obtained, and the fatigue life type of each workpiece to be tested is determined using the above method. In the parameter analysis stage, through the projection and mapping of the multidimensional feature space, a feature space is constructed in a Cartesian coordinate system based on several of the engineering components to be tested and several of the fatigue life types corresponding to the engineering components to be tested, and then the boundary analysis of the feature space is performed to determine the boundary surface that can effectively distinguish the finite life category and the infinite life category of the engineering components to be tested. The boundary surface contains the coupling relationship between the key parameters that affect the fatigue life, and provides a theoretical basis for the construction of the fatigue design equation. Finally, the fatigue design equation is obtained based on the boundary surface. The fatigue design equation can reflect the relationship between stress and fatigue life, improve the accuracy of predicting the fatigue limit, and provide a simpler and more accurate quantitative decision-making basis for engineering design.

[0087] In a possible embodiment, the above-mentioned data component feature space is combined with multiple parameter dimensions, and the parameter dimension is designed as the stress amplitude Δf of the steel bar in the use state. s , so that the fatigue design equation can reflect Δf s and fatigue life, if Δf s If it does not exceed the allowable value, it means that the component meets the fatigue requirements; otherwise, it is necessary to take measures such as increasing the cross-section, changing the reinforcement or using higher grade steel bars.

[0088] In actual application, different parameter dimensions can be designed according to engineering requirements to obtain multiple fatigue design equations for fatigue life and different parameter relationships. If the remaining life or reinforcement scheme needs to be evaluated, this equation will also be used to test its fatigue safety under service conditions. If the stress amplitude is too large, it indicates that the component may face a high fatigue risk, and reinforcement or load reduction measures should be taken in time. Through such a clear stress amplitude control criterion, designers can quickly determine whether the selected reinforcement scheme is safe under common fatigue conditions, so that the component is not prone to fatigue cracks or damage during its service life, effectively ensuring the long-term service and safety of the structure.

[0089] See also Figure 4 , Figure 4It is a structural schematic diagram of a fatigue limit prediction system for an engineering component provided by an embodiment of the present invention. The embodiment of the present invention provides a fatigue limit prediction system for an engineering component, including a working condition acquisition module 100, an infinite life probability prediction module 200, a curve acquisition module 300 and a fatigue limit prediction module 400, wherein: the working condition acquisition module 100 is used to determine the engineering component to be tested, and obtain a number of working conditions to be tested corresponding to the engineering component to be tested under a number of preset stress levels; the infinite life probability prediction module 200 is used to obtain the infinite life probability corresponding to the working condition to be tested based on a preset life prediction model for any of the working conditions to be tested, and then obtain the corresponding several infinite life probabilities based on the several working conditions to be tested; the curve acquisition module 300 is used to obtain a probability-stress level curve based on the several working conditions to be tested and the several infinite life probabilities corresponding to the several working conditions to be tested; the fatigue limit prediction module 400 is used to obtain the infinite life critical probability based on the probability-stress level curve, and then obtain the stress level critical value corresponding to the infinite life critical probability, and use the stress level critical value as the fatigue limit of the engineering component to be tested.

[0090] In a possible embodiment, in the infinite life probability prediction module 200, for any of the working conditions to be tested, the infinite life probability corresponding to the working condition to be tested is obtained based on a preset life prediction model, and then several corresponding infinite life probabilities are obtained based on several of the working conditions to be tested. The life prediction model preset process includes: generating a training data set based on a preset training component group; performing several rounds of iterative training on a preset binary classification algorithm based on the training data set to obtain a life prediction model.

[0091] See also Figure 5 and Figure 6 , Figure 5 is a stereoscopic diagram of an FRP reinforced reinforced concrete RC beam provided by an embodiment of the present invention, Figure 6It is a top view of an FRP reinforced reinforced concrete RC beam provided by an embodiment of the present invention. In a possible embodiment, the FRP reinforced reinforced concrete RC beam is used as the engineering component to be tested, and the fatigue performance data of the FRP reinforced reinforced concrete RC beam under three-point or four-point bending loading conditions are collected through experiments. In the experiment, the response of the reinforced RC beam under cyclic load becomes the basis for analysis, and the specific parameters measured and recorded include mid-span bending moment, steel bar strain, FRP strain and other contents. At the same time, the relevant mechanical behaviors are calculated through the component cross-section analysis model, and the key mechanical parameters of the component under bending load are further extracted, such as the maximum deflection at mid-span, the stress-strain characteristics of steel bars and FRP, the stress state of the concrete compression zone, and the internal force distribution of the cross section, including bending moment, shear force and axial force. These parameters obtained by experimental measurement and theoretical calculation are consistent with the material performance characteristics of the component, including concrete compressive strength, steel bar yield strength, FRP elastic modulus and thickness; and geometric characteristics, including beam section size, steel bar configuration parameters, FRP reinforcement thickness and bonding length. The above data are integrated together to form the input feature set of the working condition to be tested for the engineering component to be tested. According to the fatigue life standard and based on the widely accepted judgment criteria in engineering practice, the fatigue life of the beam is classified according to whether the number of cycles exceeds 2×106 times: f Components with less than 2×106 cycles are defined as limited life components and are assigned label 1, indicating that they have a risk of fatigue failure in a shorter life; f Components that reach or exceed 2×106 cycles are defined as infinite life components and assigned label 0, indicating that they have long-term service safety under specified loads and working conditions.

[0092] See also Figure 7 , Figure 7 It is an F1 score frequency distribution diagram provided by an embodiment of the present invention. In a possible embodiment, four commonly used machine learning algorithms, namely support vector machine (SVM), decision tree (DT), random forest (RF) and extreme gradient boosting (XGBoost), are selected to conduct in-depth mining and analysis of the data in view of the nonlinear relationship characteristics of the experimental data. A repeated hierarchical nested cross-validation framework is adopted in the model training and evaluation process, in which the outer cross-validation is mainly used to evaluate the generalization performance of the model, and the inner cross-validation is used for model selection and hyperparameter optimization to ensure a robust performance model under limited data conditions. At the same time, in order to further improve the prediction accuracy and interpretability of the model, the recursive feature elimination (RFE) method is used to screen the feature set of the data, and gradually eliminate the features that contribute less to the model, thereby optimizing the final feature combination. In terms of model performance evaluation, the F1 score is selected as an indicator to comprehensively consider the precision and recall of the classification results. From the experimental results, Figure 7The frequency distribution of F1 scores of the four algorithms after 50 repeated experiments is shown. Among them, the XGBoost model shows the best performance, with a higher average F1 score and a smaller performance fluctuation range, indicating that its ability to fit multidimensional nonlinear feature relationships is better than SVM, DT and RF. Therefore, XGBoost was finally selected as the optimal binary classifier for the fatigue life prediction task. After recursive feature elimination, the best feature subset selected contains five variables, namely: steel bar strain range, ratio of steel bar strain range to yield strain, stress ratio, maximum mid-span bending moment, and stiffness ratio of steel bar to FRP. This feature combination not only has a strong physical meaning, but also ensures the efficiency and interpretability of the model, and can fully reflect the key influencing factors of the fatigue life of reinforced components.

[0093] See also Figure 8 , Figure 8 It is a schematic diagram of the probability-stress level curve of four engineering components to be tested provided by an embodiment of the present invention. In a possible embodiment, four engineering components to be tested are selected, and a series of different stress gradient values ​​S and stress levels are set for each engineering component to be tested to simulate the fatigue performance of the component at different stress levels. The trained optimal XGBoost model is used to predict the sample to be tested to obtain the probability output P of the fatigue life type to which the sample belongs at each stress level. Among them, the XGBoost model has been fully trained with the above-mentioned high-quality and sufficient data, and its input features can effectively capture the impact of stress level on fatigue life. In particular, in this embodiment, the probability P represents the possibility that the sample belongs to the finite life category at the current stress level S, and its technical effect corresponds to the technical effect of obtaining the possibility that the engineering component to be tested belongs to the infinite life category at the corresponding stress level as described above. It should be understood that after obtaining the probability output P of the fatigue life type to which the sample belongs at each stress level, the possibility of the sample belonging to the finite life category and the possibility of belonging to the infinite life category at the corresponding stress level are obtained accordingly. The stress level S is fitted with the predicted probability P, and the following is established: Figure 8 The probability-stress PS curve shown in Figure 1 is a curve that intuitively describes the law of how the fatigue performance of the test sample changes with the stress level: as the stress level increases, the probability that the component has a limited life gradually increases. The curve can be expressed as the following formula (1):

[0094]

[0095] A, B, C, and D are fixed parameter values ​​that each control a different characteristic of the function (amplitude, steepness, center position, and baseline shift), and these parameters are determined by fitting.

[0096] Specifically, the derivation of formula 1 includes the following process:

[0097] First, before explaining the derivation process of Formula 1, necessary explanations are given to the background technology of the present invention. In high-cycle fatigue problems, the fatigue performance of materials is usually described by the relationship between the stress level and fatigue life during cyclic loading, such as the relationship between parameters such as stress amplitude or maximum stress and fatigue life; the above relationship is called SN curve or stress-life relationship. When studying the fatigue performance of reinforced concrete (RC) beams reinforced with FRP (fiber reinforced polymer), a key issue is the relationship between stress level and fatigue life. The stress level is usually defined as the ratio of peak load to ultimate bearing capacity, expressed as S=P max / P u In each set of experiments, the researchers kept the stress ratio or valley load constant and tested the fatigue life of the specimens at multiple stress levels that were gradually increased. By fitting the data points, it can be observed that the stress level (S) and the logarithm of the fatigue life (log N f ) show a linear relationship, showing a high degree of fit. This relationship can be expressed by formula 2:

[0098] S=AlogN f +B (Formula 2)

[0099] Among them, A and B are constants, and the value range of S is (0,1).

[0100] According to Formula 2, as the stress level decreases, the fatigue life shows a strictly monotonically increasing trend. Under the existing experimental parameters, corresponding to 2 million cycles (log N f The stress level of ≈6.301) is the fatigue limit of FRP reinforced RC beam. Therefore, the relationship between fatigue life (Nf) and stress level (S) can be expressed as the following formula 3:

[0101]

[0102] According to Formula 3, an increase in stress level will lead to a corresponding decrease in fatigue life.

[0103] Based on the above technical background, the embodiment of the present invention uses the trained optimal XGBoost model to predict the sample to be tested to obtain the probability output P of the fatigue life type to which the sample belongs at each stress level. Specifically, an increase in stress level will lead to a corresponding decrease in fatigue life, thereby increasing the probability that the sample is classified as having a life of more than 2 million cycles. The relationship between the model classification probability (P(y=1|S)) and the stress level (S) can be expressed as the following formula 4:

[0104] P(y=1|S)∝S (Formula 4)

[0105] Furthermore, the non-linear prediction process of the XGBoost model is described by the function g(θ), where θ represents the input parameter. The output of the model is denoted as g(S), corresponding to various stress levels S. According to the characteristics of the XGBoost algorithm, the model output is transformed through an S-shaped function to obtain the classification probability of the model, representing the probability that the sample life is less than 2 million cycles. Thus, the value of P(y = 1|S) in Equation 4 is determined by Equation 5 below:

[0106]

[0107] Since Equation (5) is a monotonically increasing function, the relationship shown in Equation 6 below can be obtained:

[0108] g(S) ∝ S (Equation 6)

[0109] It can be seen from Equation 6 that for an ideal XGBoost model, the function g(S) is monotonically increasing in the interval (0, 1). Refer to Figure 8 , Figure 8 which shows the prediction results of the XGBoost model for the stress levels and classification probabilities of FRP-strengthened RC beams under the current test parameters. The relationship between the probability P and the stress level S is described by a function similar to the standard S-shaped curve. However, its specific shape is affected by the function g(S). When the value of S is very small or very large, the value of P approaches 0 or 1, but in the middle region, the curve undergoes a rapid transition stage. It can be observed in Figure 8 that the distribution of data points approximates the standard S-shaped curve, and the data points on both sides are generally aligned with the lines representing the probabilities of 0 or 1. However, in the middle region, the continuity of the data points is lower than that on both sides. As the value of S increases, the growth trend of the P value lacks theoretical monotonicity and shows oscillatory behavior, which is because Equation 6 may not hold for non-ideal models. To enhance the stability of the model classification performance and maintain monotonicity in the probability distribution, the P-S curve of the FRP-strengthened RC beam under the current test parameters can be obtained by fitting the data points through Equation 1, as shown in Figure 8 . The probability threshold P = 0.5 corresponding to the stress level is the estimated value of the fatigue limit (SL,est).

[0110] To accurately evaluate the fatigue limit of the component, through the calibration of experimental data, the probability threshold P = 0.73 is set as the evaluation criterion, and the corresponding critical stress level of each sample is determined accordingly, which is defined as the fatigue limit of the component.

[0111] Refer to Fig. 9 , Fig. 9It is a schematic diagram of the average relative error between the predicted fatigue limit and the actual test result provided by the embodiment of the present invention. The final analysis results show that the average relative error (MAPE) between the predicted fatigue limit value and the actual test result is only 5.71%. This low error further verifies the high accuracy of the method and can provide a reliable decision-making basis for engineering design and fatigue assessment.

[0112] See also Fig.10 , Fig.10 It is a schematic diagram of boundary analysis in a feature space provided by an embodiment of the present invention. In a possible embodiment, a large-scale virtual sample set containing 20,000 samples is constructed by a virtual data generation method based on physical rules. During the generation process, it is strictly ensured that the statistical characteristics of the virtual samples are consistent with the actual experimental data and conform to the rationality of the physical constraints to ensure the reliability and engineering applicability of the virtual samples in fatigue life prediction. Using the aforementioned method for obtaining the PS curve for the engineering component to be tested, the PS curve corresponding to each virtual sample is obtained, and then the fatigue limit of all virtual samples is predicted, and finally the fatigue life category of the virtual sample is determined according to the probability threshold P = 0.73. Among them, the corresponding fatigue life N f Virtual samples with less than 2×106 cycles are assigned to the finite life category and are assigned label 1. f Virtual samples with ≥2×106 cycles are assigned to the infinite life category and are assigned the label 0. Subsequently, combined with the design ideas based on steel bar stress control that have been widely proven to be reliable in fatigue design, two key parameters closely related to steel bar performance were selected for in-depth analysis: "steel bar strain range" and "ratio of steel bar strain range to yield strain". Further, these two variables were converted into "steel bar stress range" and "ratio of steel bar stress range to yield stress". The influence of these two transformed variables on the fatigue life category distribution of the samples was studied in the Cartesian coordinate system. By analyzing the virtual samples in the following conditions: Fig.10 The distribution law in the feature space shown in the figure can clearly be observed in the finite life sample (N f <2×106) shows an obvious lower boundary in the variable space, which is marked by the black dotted line in the figure. This boundary not only reveals the coupling relationship between fatigue life and steel bar variables, but also can be used as an engineering basis for the infinite fatigue life design of FRP-reinforced RC beams. On this basis, the fatigue design equation reflecting the relationship between the steel bar stress range and fatigue life is further derived, as shown in Equation 7 below:

[0113]

[0114] Where Δf s is the steel bar strain range, f yis the yield strain of the steel bar. The fatigue design equation provides a guiding theoretical basis and calculation method for the long-life fatigue design of FRP-reinforced RC beams.

[0115] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0116] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope recorded in this specification.

[0117] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be construed as limiting the scope of the present application. It should be noted that, for a person of ordinary skill in the art, several improvements and substitutions may be made without departing from the concept of the present application, and these improvements and substitutions shall also be regarded as the protection scope of the present invention. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for predicting fatigue limit of engineering components, characterized in that: include: Determine the engineering component to be tested, and obtain a number of working conditions to be tested corresponding to the engineering component to be tested under a number of preset stress levels; For any of the working conditions to be tested, obtaining the infinite life probability corresponding to the working condition to be tested based on a preset life prediction model, and then obtaining a corresponding number of infinite life probabilities based on a number of the working conditions to be tested; Based on the plurality of working conditions to be tested and the plurality of infinite life probabilities corresponding to the plurality of working conditions to be tested, obtaining a probability-stress level curve; Based on the probability-stress level curve, the critical probability of infinite life is obtained, and then the critical value of stress level corresponding to the critical probability of infinite life is obtained, and the critical value of stress level is used as the fatigue limit of the engineering component to be tested.

2. A method for predicting fatigue limit of an engineering component according to claim 1, characterized in that: In the process of obtaining the infinite life probability corresponding to any of the working conditions to be tested based on a preset life prediction model, and then obtaining the corresponding several infinite life probabilities based on several of the working conditions to be tested, the life prediction model preset process includes: Generate a training data set based on a preset training component group; The preset binary classification algorithm is iteratively trained for several rounds based on the training data set to obtain a life prediction model.

3. A method for predicting fatigue limit of an engineering component according to claim 2, characterized in that: The generating of the training data set based on the preset training component group comprises: For any training component in the training component group, obtaining a first fatigue life of the training component and a structural performance characteristic of the training component; Obtaining a set of structural performance characteristics based on all of the structural performance characteristics; According to a preset benchmark fatigue life, classifying all the training components based on the first fatigue life to obtain a classification label set; A training data set is generated based on the structural performance feature set and the classification label set.

4. A method for predicting fatigue limit of an engineering component according to claim 3, characterized in that: The step of classifying all the training components based on the first fatigue life according to the preset reference fatigue life to obtain a classification label set includes: For any of the training components, if the first fatigue life corresponding to the training component is less than the reference fatigue life, the training component is classified into a finite life category; if the first fatigue life corresponding to the training component is greater than or equal to the reference fatigue life, the training component is classified into an infinite life category; A classification label set is obtained based on the finite lifespan category and the infinite lifespan category.

5. A method for predicting fatigue limit of an engineering component according to claim 2, characterized in that: The method of performing several rounds of iterative training on the preset binary classification algorithm based on the training data set to obtain a life prediction model includes: For any round of the iterative training; performing repeated hierarchical nested cross-validation training on the binary classification algorithm based on the training data set to obtain a current evaluation index and a current feature importance sequence, and then updating the binary classification algorithm based on the current evaluation index and the feature importance sequence; When the current evaluation index meets the preset prediction error threshold, the optimal feature subset is obtained based on the current feature importance sequence, and then the binary classification algorithm corresponding to the optimal feature subset is used as the life prediction model.

6. A method for predicting fatigue limit of an engineering component according to claim 5, characterized in that: For any round of the iterative training; performing repeated hierarchical nested cross-validation training on the binary classification algorithm based on the training data set to obtain a current evaluation index and a current feature importance sequence, and then updating the binary classification algorithm based on the current evaluation index and the feature importance sequence, including: A feature parameter sequence is acquired based on the current feature importance sequence, and then feature parameters in the feature parameter sequence that meet preset feature elimination rules are eliminated.

7. A method for predicting fatigue limit of an engineering component according to claim 1, characterized in that: The step of determining the engineering component to be tested and obtaining a plurality of working conditions to be tested corresponding to the engineering component to be tested under a plurality of preset stress levels includes: Acquiring geometric information of the engineering component to be tested; For any of the stress levels, obtaining the internal force state and strain field characteristic index of the engineering component to be tested at the stress level; A characteristic matrix is ​​constructed based on the internal force state corresponding to the stress level, the strain field characteristic index corresponding to the stress level and the geometric information of the engineering component to be tested, and the characteristic matrix is ​​used as the working condition to be tested corresponding to the engineering component to be tested at the stress level.

8. A method for predicting fatigue limit of an engineering component according to claim 1, characterized in that: Also includes: Determine a number of the engineering components to be tested, and obtain a number of the fatigue limits corresponding to the number of the engineering components to be tested; For any of the engineering components to be tested, the fatigue life type corresponding to the engineering component to be tested is obtained based on the fatigue limit corresponding to the engineering component to be tested, and then the corresponding fatigue life types are obtained based on several of the engineering components to be tested; Constructing a feature space based on a plurality of the engineering components to be tested and a plurality of the fatigue life types corresponding to the engineering components to be tested; A boundary analysis is performed on the characteristic space to obtain a fatigue design equation.

9. A fatigue limit prediction system for engineering components, characterized in that: It includes working condition acquisition module, infinite life probability prediction module, curve acquisition module and fatigue limit prediction module, among which: The working condition acquisition module is used to determine the engineering component to be tested, and obtain a number of working conditions to be tested corresponding to the engineering component to be tested under a number of preset stress levels; The infinite life probability prediction module is used to obtain the infinite life probability corresponding to any of the working conditions to be tested based on a preset life prediction model, and then obtain a number of corresponding infinite life probabilities based on a number of the working conditions to be tested; The curve acquisition module is used to acquire a probability-stress level curve based on a plurality of the working conditions to be tested and a plurality of the infinite life probabilities corresponding to the working conditions to be tested; The fatigue limit prediction module is used to obtain the critical probability of infinite life based on the probability-stress level curve, and then obtain the critical value of stress level corresponding to the critical probability of infinite life, and use the critical value of stress level as the fatigue limit of the engineering component to be tested.

10. The fatigue limit prediction system for engineering components according to claim 9, characterized in that: In the infinite life probability prediction module, for any of the working conditions to be tested, the infinite life probability corresponding to the working condition to be tested is obtained based on a preset life prediction model, and then a plurality of corresponding infinite life probabilities are obtained based on a plurality of the working conditions to be tested. The life prediction model preset process includes: Generate a training data set based on a preset training component group; The preset binary classification algorithm is iteratively trained for several rounds based on the training data set to obtain a life prediction model.

Images