Integrating Machine Learning and In-Flight Particle Diagnostics to Enhance Coating Quality in Atmospheric Plasma Spraying

Abstract Atmospheric Plasma Spraying (APS) remains a cornerstone of thermal spraying, yet its complex interdependence between in-flight particle dynamics and coating quality often limits process optimization. In this study, we integrate machine learning and in-flight particle diagnostics to develop a predictive framework to enhance coating quality. Using NiCrAlY powder, 12 controlled spraying experiments were performed, with 25 spatial measurements collected for each condition, yielding 300 measurement points for particle velocity, temperature, particle number flow rate, and diameter. These datasets were used to train feedforward neural network (FFNN) models, which demonstrated strong predictive capability with coefficients of determination (R 2 ) ranging from 0.75 to 0.976. Building on this, an empirical Plume Quality Score (PQS) was formulated by combining the dominant parameters governing particle behavior. This data-driven metric enables the identification of optimal plume conditions to improve coating characteristics. Experimental deposition showed that selected high-scoring conditions produced substantially lower porosity (as low as 1.28 ± 0.21%) than the low-scoring experimental condition (17.36 ± 1.43%). The results demonstrate a proof-of-concept process-screening framework for the investigated NiCrAlY feedstock, spray system, and operating domain; broader use requires independent material-, system-, and multi-pass validation.