Study On Evaluation Of Corrosion Resistance And Mechanical Properties Of Low-alloy Steel Based On Machine Learning Method

Low-alloy steel with excellent mechanical properties and low cost is widely used in bridges,railroads,construction,offshore oil and gas platforms and other fields.With the further exploitation of marine resources and the rapid development of marine engineering,the service performance of traditional low-alloy steel in the harsh marine environment is significantly reduced,and there is an urgent need to develop new low-alloy steel with better corrosion resistance and mechanical properties to meet the demand for long-term safe service.The alloying elements and their ratios,complex environmental factors and long service cycles determine the service performance of new low-alloy steels.In recent years,material informatics methods based on machine learning have shown great potential in predicting material properties and alloy composition screening with their powerful data fitting and knowledge mining capabilities.Therefore,the organic combination of machine learning methods to carry out efficient service performance evaluation and prediction technology and alloy composition optimization technology is one of the important ways to achieve efficient research and development of new low-alloy steels.In this dissertation,the key factors affecting corrosion resistance,tensile strength and elongation of low-alloy steels in marine atmospheric and seawater environment are investigated using machine learning methods,and a key factor screening method based on the combination of feature correlation and importance is proposed.The accurate prediction of corrosion resistance,tensile strength and elongation of low-alloy steels provides a theoretical basis and methodological guidance for the design and development of low-alloy steels with excellent performance.The main research results of this thesis are as follows:(1)A model for predicting the marine atmospheric corrosion rate of low-alloy steel is developed by the random forest algorithm.Based on the results of the random forest feature importance ranking,it is concluded that the atmospheric environmental corrosion determinants affecting the open environment and the sheltered environment are significantly different.In addition,for the specific atmospheric exposure test site,the key environmental factors affecting the corrosion rate in the open environment are closely related to the climatic characteristics of the site,while the corrosion rate in the sheltered environment is not sensitive to changes in climatic characteristics.(2)Based on the gradient boosting decision tree feature importance ranking and Kendall correlation,the key chemical composition factors and environmental factors affecting the corrosion rate of low-alloy steel in the seawater environment are identified.In order to avoid the limitation of model input alloying element types,a feature construction method that converts alloying element content into elemental physical and chemical feature is proposed.A seawater corrosion rate prediction model is established with the converted elemental features and environmental factors as input variables,which significantly improves the applicability of the machine learning model.(3)The key chemical composition and process condition factors affecting tensile strength and elongation are identified by the random forest algorithm,and the machine learning prediction models for tensile strength and elongation are developed using the key influencing factors,respectively.The synergistic optimization of a pair of negatively correlated mechanical properties is achieved by fusing feature engineering techniques,machine learning algorithms and global optimization algorithms,and the low-alloy steel composition with the best comprehensive mechanical property is successfully predicted.(4)Combining Pearson correlation,gradient boosting decision tree feature importance ranking and exhaustive enumeration method,the key elemental physical and chemical features affecting the tensile strength and elongation of lowalloy steel are identified.The analysis of the physical and chemical features of the key elements reveals that the effective nuclear charge number and stoichiometric properties are key indicators for increasing the elongation without affecting the tensile strength,which can contribute to guide the design of comprehensive mechanical property optimization of low-alloy steels.