Steel material design method based on physical-guided machine learning algorithm

Abstract

The invention provides a steel material design method based on a physics-guided machine learning algorithm, and relates to the technical field of steel material design and machine learning application. The method comprises the steps that at first, data is collected, and divided into a training set and a testing set by adopting a multiple hold-out method; a support vector machine model based on physical metallurgy guidance is established according to the training set; the correlation coefficient, greater than 90%, of the support vector machine model based on physical metallurgy guidance servesas an objective function in a genetic algorithm; optimized composition, process and materials with the best target performance are obtained; a large number of obtained design results are classified and screened through an SVC classifier, and the typical alloy composition, process and target performance are output. Accordingly, a physical metallurgy mechanism is introduced into machine learning, and meanwhile a complete design platform is formed by combining an optimization algorithm, and the design results better conform to the principles of physical metallurgy.

Description

technical field [0001] The invention relates to the technical field of steel material design and machine learning application, in particular to a steel material design method based on a physics-guided machine learning algorithm. Background technique [0002] As material research and development enters the era of big data, the application of new methods to accelerate the research and development of new materials has become the mainstream trend of material research and development. Material research and development is generally developed with the needs of society, and a variety of basic theories and design methods have been formed. The traditional material research and development is based on the traditional trial and error method for development and design. Through a large number of orthogonal experimental researches on the steel grades that meet the actual requirements, the composition and heat treatment process that meet the requirements are determined. However, this exper...

AI Technical Summary

Problems solved by technology

Although some advanced materials have been successfully designed based on physical models, some shortcomings are constantly exposed in further applications: (1) There are a large number of physical and metallurgical parameters in the PM model, which are crucial to the accuracy of the model

However, the acquisition of these important parameters often requires complex microstructure characterization, which greatly increases the amount of experiments required for modeling and reduces the efficiency of PM model development.

(2) Some complex phase transition mechanisms are still controversial in academia, which limits further optimization of the model

(3) As people have a deeper understanding of the physical mechanism in phase transitions, the physical model has been continuously optimized, which greatly increases the complexity of the PM model and limits the universality of the model

However, the current performance prediction and design of materials based on pure machine learning methods is only a purely mathematical process, and few physical metallurgical parameters are involved in the design process, which greatly wastes the unique advantages of physical metallurgy in material design

Images