Reshaping Thermal Spraying: Explainable Artificial Intelligence Meets Plasma Spraying

In recent years, Artificial Intelligence (AI) has significantly advanced thermal spraying, particularly in understanding the complex interplay between process parameters and coating properties. However, achieving high accuracy often necessitates using black-box models that provide scarce insight into the learning process, making it difficult to interpret and derive new knowledge about the process itself. Explainable AI (XAI) addresses this issue by providing interpretability to AI models. This study employs an XAI framework to gain insights into Residual Network (ResNet) and Artificial Neural Network (ANN) models trained on both simulations and experimental data to predict deposition efficiency (DE) in Atmospheric Plasma Spraying (APS). SHapley Additive exPlanations (SHAP), an interpretability framework, was then applied to help identify which process parameters have the most significant influence on the DE, revealing how changes in specific parameters affect the DE by elucidating their impact on the model predictions. This increases confidence in our model and enhances our understanding of the process by revealing the rationale behind these predictions. While many studies have integrated AI into thermal spraying, this work offers a new perspective by predicting APS deposition efficiency using ResNet and ANN models along with introducing XAI to bridge the gap between predictive accuracy and interpretability.