A method for predicting material corrosion fatigue crack growth rate
By applying incremental learning strategies in the XGBoost model and updating the initial model to adapt to new data, the problem that traditional models cannot dynamically adapt to new data is solved, and more efficient and flexible corrosion fatigue crack propagation rate prediction is achieved.
Patent Information
- Application Number
- CN202410557105.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-05-07
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2044-05-07
AI Technical Summary
Traditional machine learning models lack dynamic adaptation and continuous learning capabilities in corrosion fatigue crack propagation prediction, and cannot effectively update the model to adapt to new data, resulting in a degradation in prediction performance.
The XGBoost model is combined with the incremental learning strategy, and the initial model is trained through the initial data set, new data is collected for incremental learning, and the model is updated to obtain the evolutionary XGBoost model to achieve adaptive adjustment and real-time cognition of the new data.
Without retraining the entire model, dynamically update the model to improve the real-time and accuracy of corrosion fatigue crack propagation rate prediction, reduce the forgetting rate of the model on the old data, and ensure real-time recognition of the material corrosion fatigue crack propagation behavior Know.
Smart Images

Figure CN118315000B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of material performance evaluation and life prediction, and in particular relates to a method for predicting the growth rate of material corrosion fatigue cracks. Background Art
[0002] Corrosion fatigue failure specifically refers to the common failure form of components under the synergistic effect of corrosive environment and alternating loads. Its complexity far exceeds the simple superposition of corrosion and fatigue effects. In this context, the accurate prediction of corrosion fatigue crack growth rate is of great significance for engineering applications, which is not only related to the service life and structural integrity of materials, but also directly affects the safety and economy of related equipment.
[0003] With the rise of artificial intelligence and the development of machine learning technology, machine learning has been widely used in the prediction of metal corrosion fatigue crack growth behavior. In order to effectively study the damage development of aviation aluminum alloy materials in their service environment, a large number of scientific researchers in my country have conducted in-depth research on the mechanical fatigue and corrosion fatigue behaviors of aviation aluminum alloy materials using data-driven methods to predict the corrosion fatigue crack growth rate and fatigue life of surface crack specimens. The conclusion that traditional machine learning or deep learning models trained with corrosion fatigue crack length data can accurately predict the fatigue crack growth life of specimens has been confirmed.
[0004] However, traditional machine learning methods also face the problem of model updating and adapting to new data in application. In the current research on corrosion fatigue crack growth prediction, although various machine learning algorithms have demonstrated their excellent ability to process complex data and large-scale data sets, and thus achieved efficient prediction performance in multiple fields, most of these traditional machine learning models rely on static data sets for training. The inherent limitation of this training method is that the model lacks the ability to update in real time when facing new data or continuous changes in data, which limits the application effect of the model in a dynamic environment. When faced with new data, existing static models either cannot use new information to improve prediction performance or need to retrain the entire model. Retraining the model not only consumes a lot of time and computing resources, but also seems impractical in rapidly changing application scenarios. In the specific scenario of corrosion fatigue crack growth prediction, due to the continuous accumulation of new experimental data, traditional machine learning models cannot flexibly integrate new information, nor can they maintain real-time cognition of the corrosion fatigue crack growth behavior of materials. This lack of adaptive adjustment ability limits the ability of the model to adapt to the changing trend of corrosion crack growth in a timely manner.
[0005] Therefore, the existing technology has the problem that the model is difficult to update and adapt to new data, which needs to be solved urgently. Summary of the invention
[0006] The purpose of the present invention is to provide a method for predicting the corrosion fatigue crack growth rate of materials, so as to solve the problems in the prior art of lack of dynamic adaptation and continuous learning ability to new data, and the gradual decline of prediction performance as new data accumulates, so as to provide a more efficient and flexible method for corrosion fatigue crack growth prediction.
[0007] To achieve the above object, the present invention provides the following technical solutions:
[0008] A method for predicting the corrosion fatigue crack growth rate of a material comprises the following steps:
[0009] Collect initial data of material corrosion fatigue crack growth and perform preprocessing;
[0010] Construct the initial dataset;
[0011] Train an initial XGBoost model using the initial dataset;
[0012] Collect new material corrosion fatigue crack growth data;
[0013] Based on the newly added material corrosion fatigue crack growth data, an incremental learning strategy is adopted to update the initial XGBoost model to obtain the evolved XGBoost model.
[0014] The evolutionary XGBoost model is used to predict the corrosion fatigue crack growth rate of materials.
[0015] The material corrosion fatigue crack growth rate prediction method of the present invention has the following beneficial effects:
[0016] By combining the XGBoost model with an incremental learning strategy, multi-dimensional influencing factors are taken into account, and new data are flexibly integrated based on the initial XGBoost model. The model can be dynamically updated on the gradually accumulated data without the need to retrain the entire model. The new data can be used to improve the real-time and accuracy of the prediction of the material corrosion fatigue crack growth rate, achieve adaptive adjustment of the new data, and maintain real-time understanding of the material corrosion fatigue crack growth behavior.
[0017] In some embodiments, based on the newly added material corrosion fatigue crack growth data, an incremental learning strategy is adopted to update the initial XGBoost model to obtain an evolved XGBoost model including:
[0018] The newly added material corrosion fatigue crack growth data are preprocessed by adopting the material corrosion fatigue crack growth initial data preprocessing method to obtain the incremental data set;
[0019] Based on the incremental data set, at least two incremental learning strategies are adopted to update the initial XGBoost model respectively;
[0020] Combining various incremental learning strategies, incremental learning of comprehensive strategies is carried out to obtain the evolutionary XGBoost model.
[0021] Therefore, through the use and integration of multiple incremental learning strategies, the advantages of multiple incremental learning strategies can be fully utilized to ensure the adaptability and accuracy of the model and effectively reduce the forgetting rate of the model on old data sets.
[0022] In some implementations, the steps of updating the initial XGBoost model based on the incremental data set and adopting at least two incremental learning strategies include:
[0023] An incremental learning strategy based on structural optimization is used to update the initial XGBoost model;
[0024] Adopt regularization-based incremental learning strategy to update the initial XGBoost model;
[0025] An incremental learning strategy based on experience replay is used to update the initial XGBoost model.
[0026] Therefore, through the implementation of the incremental learning strategy based on structural optimization, the tree depth and the learning rate can be reasonably reduced; the implementation of the incremental learning strategy based on regularization can control the complexity of the model to avoid overfitting; the implementation of the incremental learning strategy based on experience replay, in each incremental update, by combining part of the old data and the new data for training, in order to combat the catastrophic forgetting problem common in incremental learning.
[0027] In some embodiments, the collected initial data of material corrosion fatigue crack growth include but are not limited to:
[0028] Material parameters, including material grade, composition, heat treatment status, initial crack length, sampling direction, and specimen shape;
[0029] Environmental parameters, including environment type, temperature and humidity, medium and concentration, pH value, and air pressure;
[0030] Mechanical parameters, including crack growth rate, stress intensity factor amplitude, stress ratio, loading waveform, loading frequency, elastic modulus, yield strength, ultimate strength, and elongation.
[0031] Therefore, by integrating multi-dimensional material parameters, environmental parameters, and mechanical parameters, the influencing factors of corrosion fatigue crack propagation are considered more comprehensively and fully, thereby fundamentally improving the rationality and accuracy of the model.
[0032] In some embodiments, the method for preprocessing the initial data of material corrosion fatigue crack growth is: performing data cleaning, processing missing values and abnormal values, and performing data normalization, thereby ensuring data quality.
[0033] In some embodiments, the step of training the initial XGBoost model using the initial dataset is as follows:
[0034] An initial XGBoost model is constructed based on the initial data set, machine learning is performed on the initial XGBoost model, and the optimal initial XGBoost model is selected through Bayesian hyperparameter tuning technology.
[0035] In this way, a more reasonable initial XGBoost model is obtained to ensure the accuracy of the prediction.
[0036] In some embodiments, the steps are further included:
[0037] The evolved XGBoost model is evaluated on its prediction accuracy and forgetfulness of old tasks.
[0038] Therefore, by testing the prediction accuracy and the forgetting of old tasks, we can obtain the model's adaptability to new and old data and judge whether the model is practical in various scenarios.
[0039] In some embodiments, the prediction accuracy of the evolved XGBoost model is calculated as:
[0040] The prediction accuracy of the evolved XGBoost model is calculated as:
[0041]
[0042]
[0043] Among them, y pred is the predicted value of the evolved XGBoost model, y true is the actual measured value, E r is the predicted value y pred With the actual value y true Relative error, A c is the prediction accuracy, n represents the total number of samples, and τ is the set threshold.
[0044] In some embodiments, the method for calculating the forgetfulness of old tasks by the evolved XGBoost model is:
[0045]
[0046] represents the accuracy of the model trained at time t1 to master a specific set of knowledge K, is the accuracy of recalling or reproducing the same knowledge K at time t2, where t2>t1, and F is the forgetting rate.
[0047] In some embodiments, the steps are further included:
[0048] Continuously monitor the performance of the evolving XGBoost model;
[0049] If the performance of the evolutionary XGBoost model cannot meet the preset conditions, jump to the step to collect the newly added material corrosion fatigue crack growth data.
[0050] Therefore, as new data continues to arrive, the model is continuously updated incrementally to achieve dynamic adaptability to the ever-increasing new data, ensuring that the model maintains efficient and accurate prediction capabilities throughout its entire life cycle. BRIEF DESCRIPTION OF THE DRAWINGS
[0051] Figure 1 A flow chart of a method for predicting material corrosion fatigue crack growth rate provided in one embodiment of the present invention;
[0052] Figure 2 A flowchart of step S5 of a method for predicting material corrosion fatigue crack growth rate provided in one embodiment of the present invention;
[0053] Figure 3 is a schematic diagram of an incremental learning strategy based on structural optimization in step S51;
[0054] Figure 4 is a schematic diagram of the regularization-based incremental learning strategy in step S51;
[0055] Figure 5 is a schematic diagram of the incremental learning strategy based on experience playback in step S51;
[0056] Figure 6 The content of the data set for testing a certain series of aluminum alloys;
[0057] Figure 7 Schematic diagram of the initial data set and incremental data set for testing;
[0058] Figure 8 A statistical graph showing the prediction accuracy of the tested model for new data;
[0059] Fig. 9 A statistical graph showing the prediction accuracy of the tested model on old data;
[0060] Fig.10 This is a statistical graph of the forgetting rate of the tested model;
[0061] Fig.11 The scatter plot of the predicted value and the experimental value of the model updated based on the comprehensive strategy at each incremental step;
[0062] Fig.12 Scatter plot of the predicted value and experimental value of the model updated based on the comprehensive strategy tested on the old data after each incremental step. DETAILED DESCRIPTION
[0063] The present invention will be further described in detail below in conjunction with the accompanying drawings.
[0064] Figure 1 The figure schematically shows the prediction of material corrosion fatigue crack growth rate according to one embodiment of the present invention. Figure 1 As shown, the method comprises the following steps:
[0065] S1. Collect the initial data of material corrosion fatigue crack growth and perform preprocessing;
[0066] S2, build the initial data set;
[0067] S3, train the initial XGBoost model using the initial dataset;
[0068] S4. Collect new material corrosion fatigue crack growth data;
[0069] S5. Based on the newly added material corrosion fatigue crack growth data, an incremental learning strategy is adopted to update the initial XGBoost model to obtain an evolved XGBoost model.
[0070] S6. Use the evolutionary XGBoost model to predict the corrosion fatigue crack growth rate of materials.
[0071] In a specific embodiment, in step S1, the collected initial data of material corrosion fatigue crack extension include material parameters, environmental parameters, and mechanical parameters, wherein the material parameters include material grade, composition, heat treatment state, initial crack length, sampling direction, sample shape, etc., the environmental parameters include environment category, temperature and humidity, medium and concentration, pH value, air pressure, etc., and the mechanical parameters include crack growth rate (da / dN), stress intensity factor amplitude (ΔK), stress ratio, loading waveform, loading frequency, elastic modulus, yield strength, ultimate strength, elongation, etc. By introducing multi-dimensional features, the influencing factors of corrosion fatigue crack extension are more comprehensively and fully considered, thereby fundamentally improving the rationality and accuracy of the model.
[0072] Specifically, the method for preprocessing the initial data of material corrosion fatigue crack growth is: data cleaning, processing missing values and outliers, and data normalization (Min-Max normalization). The specific formula is as follows:
[0073]
[0074] Where x is the original data point. min(x) is the minimum value in the data set. max(x) is the maximum value in the data set. x′ is the normalized data point. This ensures the quality of the initial data.
[0075] In step S2, an initial data set is constructed from the collected initial data of material corrosion fatigue crack growth, including an initial training set and a test set, for training and evaluating the baseline performance of the initial XGBoost model.
[0076] In a specific embodiment, the method of step S3 is: construct an initial XGBoost model based on the initial data set, perform machine learning on the initial XGBoost model, and select the optimal initial XGBoost model through Bayesian hyperparameter tuning technology. Specifically, the XGBoost model is trained using the initial data set, and the optimal model parameters are selected through Bayesian hyperparameter tuning, including the maximum depth of the tree, the learning rate, the subsample ratio, etc., until the early stopping condition or the maximum number of iterations is reached. The specific formula is as follows:
[0077] Obj=L(θ)+Ω(θ)
[0078] Among them, L(θ) is the loss function of the model on the training data, which is used to measure the deviation between the model prediction value and the true value. Ω(θ) is the regularization term, which is used to control the complexity of the model and avoid overfitting. It usually includes the number of leaf nodes in the model and the weight of the tree.
[0079] like Figure 2 As shown, in a specific embodiment, step S5 includes:
[0080] S51, preprocessing the newly added material corrosion fatigue crack growth data by using the material corrosion fatigue crack growth initial data preprocessing method to obtain an incremental data set;
[0081] S52. Based on the incremental data set, adopt at least two incremental learning strategies to update the initial XGBoost model respectively;
[0082] S53. Combining various incremental learning strategies, incremental learning of comprehensive strategies is performed to obtain an evolutionary XGBoost model.
[0083] In step S51, the newly arrived data is preprocessed in the same way as step S1, that is, data cleaning is performed, missing values and outliers are processed, and data normalization (Min-Max normalization) is performed to obtain an incremental data set, paving the way for subsequent incremental learning.
[0084] Further, step S52 includes:
[0085] Updating the initial XGBoost model using an incremental learning strategy based on structure optimization;
[0086] Updating the initial XGBoost model using an incremental learning strategy based on regularization;
[0087] Updating the initial XGBoost model using an incremental learning strategy based on experience replay.
[0088] As Figure 3 shown, during implementation, first perform incremental learning using an incremental learning strategy based on structure optimization, which mainly includes:
[0089] Reasonably increase the number of trees. The specific formula is as follows:
[0090]
[0091] where, F new (x) is the updated model, F old (x) is the model before update, h i (x) is the newly added i-th tree, representing the learning rate.
[0092] Reasonably reduce the tree depth. The specific formula is as follows:
[0093]
[0094] where, d new is the maximum depth of the reduced tree, d old is the original maximum depth. For a binary tree with a maximum depth of d, the maximum possible number of leaf nodes is 2 d . This indicates that the depth of the tree controls the complexity of the model at an exponential level.
[0095] Reasonably reduce the learning rate. The specific formula is as follows:
[0096]
[0097] where, is the updated model, F t (x) is the current model, is the learning rate, and h t (x) is the decision tree trained based on the residual at the current step.
[0098] When reducing the learning rate (i.e., new < old), the influence of the new tree is weakened, thus ensuring that the model does not quickly discard the information of the old data while absorbing new data. This update strategy helps to achieve a balance between new and old data.
[0099] like Figure 4 As shown, the incremental learning strategy based on regularization is used for incremental learning, mainly through L2 regularization, also known as ridge regression or weight decay, to control the complexity of the model and avoid overfitting. The formula is as follows:
[0100]
[0101] Among them, L(Φ) is the objective function, is the predicted value and i The loss function between the true value, Ω(f k ) is the regularization term of the kth tree, usually defined as:
[0102]
[0103] Here, T is the number of leaves in the tree, ω is the weight of the leaf node, and λ is the regularization parameter. In this way, the initial XGBoost model finds a balance between model complexity (through the structure of the tree) and model fit (through the loss function).
[0104] For a given model, its loss function (such as mean square error) is usually expressed as:
[0105]
[0106] Among them, L(w) is the loss function, y i is the true value of the i-th sample;
[0107] The regularization term in the initial XGBoost model consists of two parts: L1 regularization (Lasso regularization) and L2 regularization (Ridge regularization). The form of the L2 regularization part adopted in this embodiment is as follows:
[0108]
[0109] Here, λ is the regularization parameter that controls the strength of L2 regularization. j is the value of the jth parameter in the model, and T is the total number of model parameters. Through this regularization term, the initial XGBoost model encourages the model parameters to shrink to smaller values to avoid overfitting.
[0110] like Figure 5 As shown in the figure, an incremental learning strategy based on experience replay is used for incremental learning. In each incremental update, a part of the old data and the new data are combined for training to combat the catastrophic forgetting problem that is common in incremental learning. The specific formula is:
[0111]
[0112] D train is the fused training set, Let D be a new dataset containing M samples. new , For the old dataset D old After sampling, a subset containing N′ samples is obtained. Among them, D train It contains M+N′ samples, aiming to improve the generalization ability of the model by combining new and old data and mitigate the impact of catastrophic forgetting.
[0113] The initial XGBoost model adds new trees f in each iteration. t To minimize the enhanced objective function. The tree splitting process is based on maximizing the splitting gain, which can be described in detail as:
[0114] Assume that the sample sets contained in the left and right child nodes of a node are I L and I R , then the split gain Gain can be defined as:
[0115]
[0116] Among them, g i and h i Represent the first and second order derivatives of the loss function with respect to the predicted value, i.e., the gradient and the Hessian value. These derivatives are about the predicted value of the model in the i-th iteration. Through such gain calculation and corresponding tree structure optimization, the initial XGBoost model can effectively adjust the model in each round to adapt to the fused training data D train .
[0117] Then, the three different incremental learning strategies, namely, incremental learning strategy based on structural optimization, incremental learning strategy based on regularization, and incremental learning strategy based on experience replay, are integrated to conduct incremental learning of comprehensive strategies, and the initial XGBoost model is further updated to obtain the evolutionary XGBoost model. This evolutionary XGBoost model combines the advantages of the three incremental learning strategies, can effectively improve the dynamic adaptability to new data, and retain the need for existing knowledge while accepting new data, reducing the forgetting rate on the old data set. Furthermore, the use of the evolutionary XGBoost model to predict the corrosion fatigue crack growth rate of materials can obtain more accurate and reliable prediction values.
[0118] In a specific embodiment, the method further includes step S7: evaluating the prediction accuracy of the evolved XGBoost model and the forgetting of old tasks.
[0119] Specifically, the prediction accuracy of the evolved XGBoost model is evaluated using a uniquely customized accuracy calculation method:
[0120]
[0121]
[0122] Among them, A c is the accuracy, 1 is the indicator function, which is 1 when its internal condition is true, otherwise it is 0; n represents the total number of samples; E r is the predicted value y of the evolved XGBoost model pred Compared with the actual measured value y true Relative error, A c Defined as E r The proportion that does not exceed the given threshold τ, that is, it satisfies the condition E r The proportion of observations with ≤τ to the total observations. The threshold τ defines the upper limit of the acceptable relative error and is a key parameter for evaluating the accuracy of predictions. If the relative error is less than or equal to the threshold, the prediction is considered accurate. In this embodiment, the threshold τ can be set to 5%.
[0123] This custom accuracy metric provides an effective tool to measure the prediction ability of a model within a specific tolerance range. When the threshold τ is set to 0.05, the prediction accuracy of the model can be rigorously evaluated to guide the selection and application of the model in practice.
[0124] More specifically, the forgetting of old tasks of the evolved XGBoost model is evaluated, mainly quantified by the forgetting rate F. The formula for the forgetting rate is as follows:
[0125]
[0126] It represents the accuracy of the model trained at time t1 to master a specific set of knowledge K. At time t2 (where t2>t1) after incremental learning, the model is required to recall or reproduce the same knowledge K and calculate the corresponding accuracy at time t2 The forgetting rate F can be calculated by measuring the degradation of knowledge representation between time points t1 and t2. The closer the forgetting rate is to 0, the more effectively the learning model can retain previously learned knowledge, that is, it forgets less of the learned information while learning new information. On the contrary, if the forgetting rate is close to 1, it means that the model has almost forgotten all previously learned information, which is unacceptable in many practical applications.
[0127] In a specific embodiment, step S8 is also included:
[0128] Continuously monitor the performance of the evolving XGBoost model;
[0129] If the performance of the evolved XGBoost model cannot meet the preset conditions, jump to step S4.
[0130] Specifically, through step S4, newly added material corrosion fatigue crack growth data is collected again and updated again through incremental learning, and the relevant incremental learning strategy can be further optimized so that the evolutionary XGBoost model updated through incremental learning meets the performance requirements. This preset condition can be determined according to actual needs.
[0131] In this way, the performance of the model is continuously monitored, and the model is incrementally updated continuously as new data arrives, ensuring that the model maintains efficient and accurate predictive capabilities throughout its life cycle.
[0132] In a specific embodiment, a certain series of aluminum alloys are used as test objects to collect initial data of material corrosion fatigue crack growth, such as Figure 6 , Figure 7 As shown, the initial data set includes features of 30 dimensions. The initial data set includes 765 data named data. In the first step, 249 new data are added, and the data set name is data_1. In the second step, 236 new data are added, and the data set name is data_2. In the third step, 106 new data are added, and the data set name is data_3.
[0133] The XGBoost model integration model is used to train and predict the corrosion fatigue crack growth behavior of 2024 aluminum alloy to obtain the initial XGBoost model. Then three different types of aluminum alloys (2124, 2E12, 2050) are gradually introduced for continuous batch incremental learning, and three incremental learning strategies, namely, incremental learning strategy based on structural optimization, incremental learning strategy based on regularization, and incremental learning strategy based on experience replay, are used for incremental learning at the same time. Finally, these three strategies are combined for the same incremental learning to obtain the evolutionary XGBoost model, and the prediction accuracy and forgetting rate of the models obtained by various methods are evaluated and compared.
[0134] like Figure 8 As shown in the figure, the prediction accuracy of various models on the initial data set is high. With the addition of incremental data and the advancement of incremental learning, the accuracy of the initial model that has not been updated and the model that is updated with only a single incremental learning strategy decreases significantly. The accuracy of the evolved XGBoost model obtained by adopting a comprehensive strategy always remains at a high level.
[0135] like Fig. 9 , Fig.10As shown in the figure, when tested on the old data, the unupdated initial model and the updated model using the structural optimization and regularized incremental learning strategies have lower accuracy and higher forgetting rates. For example, when testing data_2 on the old data after the third step of incremental learning, the forgetting rate of the unupdated initial model reaches 0.6871, the forgetting rate of the structural optimization incremental learning strategy reaches 0.6582, and the forgetting rate of the regularized incremental learning strategy reaches 0.3240. The forgetting rates based on the experience replay strategy and the comprehensive strategy remain at a low level. In summary, the forgetting rate of the evolutionary XGBoost model at all steps is maintained at a low level, and its forgetting rate is lower than 0.05 in all incremental steps, showing its excellent knowledge retention ability. A good balance has been achieved between learning new tasks and maintaining old knowledge. As Fig.11 , Fig.12 As shown in the figure, the excellent prediction results of the evolutionary XGBoost model can be more intuitively demonstrated through the scatter plot of the predicted value and the experimental value in each incremental step, as well as the scatter plot of the predicted value and the experimental value tested on the old data after each incremental step.
[0136] In addition, in other embodiments, a material corrosion fatigue crack growth rate prediction system is also disclosed, including:
[0137] The initial data acquisition and processing module is used to acquire and pre-process the initial data of material corrosion fatigue crack growth and construct an initial data set;
[0138] A training module, connected to the initial data collection and processing module, for training an initial XGBoost model using an initial data set;
[0139] The incremental data acquisition and processing module is used to acquire the newly added material corrosion fatigue crack growth data and pre-process it in the same way as the initial data pre-processing of the material corrosion fatigue crack growth to obtain the incremental data set;
[0140] The incremental learning module is connected to the training module and the incremental data acquisition and processing module, and is used to update the initial XGBoost model based on the incremental data set by adopting an incremental learning strategy.
[0141] Specifically, the initial data of material corrosion fatigue crack extension is collected through the initial data acquisition and processing module, and preprocessed to construct an initial data set; the initial XGBoost model is trained using the initial data set through the training module; the incremental data acquisition and processing module is used to collect new material corrosion fatigue crack extension data and preprocessed to obtain an incremental data set; the incremental learning module is used to update the initial XGBoost model based on the incremental data set using an incremental learning strategy to obtain a model that is continuously iteratively updated as new data is added, and the new data is used to improve the real-time and accuracy of the prediction of material corrosion fatigue crack extension rate, with dynamic adaptability to always maintain the accuracy of the prediction.
[0142] The embodiment of the present invention also discloses an electronic device, including: a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the above-mentioned material corrosion fatigue crack growth rate prediction method. By executing the above-mentioned material corrosion fatigue crack growth rate prediction method, it can realize the specific functions of the method and provide a corrosion fatigue crack growth model with higher accuracy.
[0143] The embodiment of the present invention also discloses a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above-mentioned material corrosion fatigue crack growth rate prediction method is implemented. By executing the above-mentioned material corrosion fatigue crack growth rate prediction method, the specific functions of the method can be realized, and a corrosion fatigue crack growth model with higher accuracy can be provided.
[0144] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.
[0145] The method according to the present invention described above can be implemented in hardware, firmware, or as software or computer code that can be stored in a recording medium (such as a CD ROM, RAM, floppy disk, hard disk or magneto-optical disk), or as a computer code that is originally stored in a remote recording medium or a non-temporary machine-readable medium downloaded through a network and will be stored in a local recording medium, so that the method described herein can be stored in such software processing on a recording medium using a general-purpose computer, a special-purpose processor or programmable or special-purpose hardware (such as an ASIC or FPGA). It can be understood that a computer, a processor, a microprocessor controller or programmable hardware includes a storage component (e.g., RAM, ROM, flash memory, etc.) that can store or receive software or computer code, and when the software or computer code is accessed and executed by a computer, a processor or hardware, the processing method described herein is implemented. In addition, when a general-purpose computer accesses the code for implementing the processing shown here, the execution of the code converts the general-purpose computer into a special-purpose computer for executing the processing shown here.
[0146] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of the present invention. At the same time, for those skilled in the art, according to the idea of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.
Claims
1. A method for predicting the corrosion fatigue crack growth rate of a material, characterized in that: The method comprises the following steps: Collect initial data of material corrosion fatigue crack growth and perform preprocessing; Construct the initial dataset; Train an initial XGBoost model using the initial dataset; Collect new material corrosion fatigue crack growth data; The newly added material corrosion fatigue crack growth data are preprocessed by adopting the material corrosion fatigue crack growth initial data preprocessing method to obtain the incremental data set; An incremental learning strategy based on structural optimization is used to update the initial XGBoost model; Adopt regularization-based incremental learning strategy to update the initial XGBoost model; Adopting incremental learning strategy based on experience replay to update the initial XGBoost model; Combining various incremental learning strategies, we conduct incremental learning of comprehensive strategies and obtain the evolutionary XGBoost model. The evolutionary XGBoost model is used to predict the corrosion fatigue crack growth rate of materials.
2. The method for predicting material corrosion fatigue crack growth rate according to claim 1, characterized in that: The collected initial data of material corrosion fatigue crack growth include but are not limited to: Material parameters, including material grade, composition, heat treatment status, initial crack length, sampling direction, and specimen shape; Environmental parameters, including environment type, temperature and humidity, medium and concentration, pH value, and air pressure; Mechanical parameters, including crack growth rate, stress intensity factor amplitude, stress ratio, loading waveform, loading frequency, elastic modulus, yield strength, ultimate strength, and elongation.
3. The method for predicting material corrosion fatigue crack growth rate according to claim 1, characterized in that: The method for preprocessing the initial data of material corrosion fatigue crack growth is: Perform data cleaning, handle missing values and outliers, and normalize the data.
4. The method for predicting material corrosion fatigue crack growth rate according to claim 1, characterized in that: The method for training the initial XGBoost model using the initial dataset is: An initial XGBoost model is constructed based on the initial data set, machine learning is performed on the initial XGBoost model, and the optimal initial XGBoost model is selected through Bayesian hyperparameter tuning technology.
5. The method for predicting material corrosion fatigue crack growth rate according to claim 1, characterized in that: Also includes the steps: The evolved XGBoost model is evaluated on its prediction accuracy and forgetfulness of old tasks.
6. The method for predicting material corrosion fatigue crack growth rate according to claim 5, characterized in that: The prediction accuracy of the evolved XGBoost model is calculated as: Among them, y pred is the predicted value of the evolved XGBoost model, y true is the actual measured value, E r is the predicted value y pred With the actual value y true Relative error, A c is the prediction accuracy, n represents the total number of samples, and τ is the set threshold.
7. The method for predicting material corrosion fatigue crack growth rate according to claim 5, characterized in that: The calculation method of the evolutionary XGBoost model for forgetting old tasks is: represents the accuracy of the model trained at time t1 to master a specific set of knowledge K, is the accuracy of recalling or reproducing the same knowledge K at time t2, where t2>t1, and F is the forgetting rate.
8. The method for predicting material corrosion fatigue crack growth rate according to claim 5, characterized in that: Also includes the steps: Continuously monitor the performance of the evolving XGBoost model; If the performance of the evolutionary XGBoost model cannot meet the preset conditions, jump to the step to collect the newly added material corrosion fatigue crack growth data.
