Use of machine learning in micro-plasma metal additive manufacturing process for optimum deposition geometry

Abstract

Micro-plasma metal additive manufacturing (μ-PMAM) is a highly precise and energy-efficient process to fabricate the metallic components. But, achieving consistent deposition geometry remains challenging due to the complex interactions among process parameters and dynamics of different deposition layers. This thesis presents noble approach of employing machine learning (ML) and deep learning (DL) respectively to predict and optimize geometry of single-layer and multi-layer depositions fabricated by the μ-PMAM process. A high dynamic range (HDR) camera was used to record videos of single-layer depositions of Ti6Al4V on the same material base plate and single-layer and multi-layer depositions of SS 316L on a mild streel base plate for different combinations of μ-PMAM process parameters (such as μ-plasma power, feedstock powder flow rate, worktable feed rate for single-layer deposition and additional parameters namely stand-off-distance, deposition layer index, height and width of previously deposited layer, and cumulative height for multi-layer depositions). Images were extracted from each recorded video at a rate of 30 frames per second. The extracted images were annotated and feature scaling was performed for the single-layer depositions and the Histogram based Multi‑Mode method was used for multi-layer depositions to generate the datasets.