Effect of microstructural heterogeneity on fatigue strength predicted by reinforcement machine learning

Abstract

The posterior statistical distributions of fatigue strength are determined using Bayesian inferential statistics and the Metropolis Monte Carlo method. This study explores how structural heterogeneity affects ultrahigh cycle fatigue strength in additive manufacturing. Monte Carlo methods and procedures may assist estimate fatigue strength posteriors and scatter. The acceptable probability in Metropolis Monte Carlo relies on the Markov chain's random microstructure state. In addition to commonly studied variables, the proportion of chemical composition was demonstrated to substantially impact fatigue strength if fatigue lifetime in crack propagation did not prevail due to high threshold internal notches. The study utilizes an algorithm typically used for quantum mechanics to solve the complicated multifactorial fatigue problem. The inputs and outputs are modified by fitting the microstructural heterogeneities into the Metropolis Monte Carlo algorithm. The main advantage here is applying a general-purpose nonphenomenological model that can be applied to multiple influencing factors without high numerical penalty.

Description

Table of contents

Keywords

Additive manufacturing, Machine learning, Markov chain, Monte Carlo simulation, Reinforcement learning, Very high cycle fatigue

Citation