Detecting secondary-phase in bainite microstructure through deep-learning based single-shot approach

TL;DR

This paper introduces a deep-learning single-shot object detection method to identify secondary phases in bainite microstructures, enabling automated and scalable microstructure analysis crucial for understanding material properties.

Contribution

A novel regression-based deep-learning algorithm is developed for single-shot detection of secondary phases in bainite, outperforming conventional classification methods.

Findings

Effective detection of MA islands in bainite microstructures

Single-shot detection framework maintains micrograph dimensions

Automated analysis improves understanding of microstructure-property relationships

Abstract

Relating properties and processing conditions to multiphase microstructures begins with identifying the constituent phases. In bainite, distinguishing the secondary phases is an arduous task, owing to their intricate morphology. In this work, deep-learning techniques deployed as object-detection algorithms are extended to realise martensite-austenite (MA) islands in bainite microstructures, which noticeably affect their mechanical properties. Having explored different techniques, an extensively trained regression-based algorithm is developed to identify the MA islands. This approach effectively detects the secondary phases in a single-shot framework without altering the micrograph dimensions. The identified technique enables scalable, automated detection of secondary phase in bainitic steels. This extension of the detection algorithm is suitably prefaced by an analysis exposing the inadequacy of conventional classification approaches in relating the processing conditions and composition to the bainite microstructures with secondary phases.