Optimizing Cold Spray Deposition on Thermoplastics: A Machine Learning Approach Focused on Powder Properties

Abstract The cold spray (CS) process offers an advanced method for metallizing thermoplastic polymers, providing a low-temperature solution to overcome the limitations of traditional coating techniques. However, optimizing the cold spray process for metallizing thermoplastic polymers is a complex task due to the numerous interacting parameters that influence coating quality. As traditional trial-and-error approaches are time-consuming and costly, machine learning (ML) could offer a solution to these challenges by providing further insights into the process and enabling more efficient optimization. The aim of this work is to identify the most relevant input parameters for ML models, with a particular focus on powder characteristics, to predict two critical outcomes: particle flattening and penetration depth. Two distinct datasets were created for this study: one focused on particle yield strength and the other on powder density, each combined with further input parameters like impact velocity and substrate yield strength. These datasets were constructed using experimental data and finite element modeling (FEM) simulations, with materials including copper, aluminum, titanium, and others, applied to thermoplastic substrates like polyether ether ketone (PEEK), acrylonitrile butadiene styrene (ABS), and polyamide 66 (PA66). Several ML algorithms, including decision trees, neural networks, and Gaussian process regression, were tested to predict coating behavior, and the effects of Z-score normalization were evaluated for improving model stability and prediction accuracy. The results show that particle yield strength is crucial for flattening, while particle density primarily governs penetration depth. This study demonstrates that ML, when combined with a solid understanding of the process, offers an effective framework for optimizing CS deposition on polymers.