A penalized autologistic regression with application for modeling the microstructure of dual-phase high-strength steel

Abstract

Recently, dual-phase high-strength steel has attracted increasing attention in the automotive industry due to its prominent physical and mechanical properties. Microstructures of dual-phase high-strength steel have a significant effect on the properties of steel, such as wear resistance and strength, so they have an important role in the quality of steel. Therefore, statistical modeling of the microstructures of steel is of great interest. However, most existing methods require many model parameters due to the complex topological forms of microstructures, which make these models suffer from overfitting and high computational time for parameter estimation. To overcome these challenges, a novel statistical model is proposed to characterize microstructures and select the most effective parameters. Furthermore, an efficient parameter estimation method is developed to estimate the model parameters given a microstructure sample. The developed method is based on a penalized pseudo log-likelihood and the accelerated proximal gradient. A simulation study is conducted to verify the developed methods. The proposed methodology is validated by a real-world example of the microstructures of high-strength steel, and the case study shows the superior performance of the developed model compared with existing methods.