Application of machine learning for the prediction of particle velocity distribution and deposition efficiency for cold spraying titanium powder

This study demonstrates the efficacy of machine learning (ML) techniques, specifically Support Vector Regression (SVR) and Neural Network (NN) models, in predicting the spray plume characteristic particle velocity distribution during cold spraying of Titanium. Considering the complexity of particle velocity distribution, models with the single particle velocity, average particle velocity and particle count in the spray plume have been explored associated with a novel data binning mechanism. The models achieved a root mean square error (RMSE) of approximately 41 m/s when tested for predicting the average particle velocity depending on the lateral position within the spray plume. The models for single particle velocity exhibited inferior performance, ascribed to the stochastic nature of single particles in the spray plume. In addition to predicting particle behaviour, the ML-based models were combined with a semi-empirical method to forecast deposition efficiency (DE) in cold spray operations. The developed DE prediction models showcased promising results, achieving an RMSE of 3.2% DE and 5.5 % DE using SVR and NN, respectively. These findings emphasize the potential of ML approaches in enhancing predictions of the particle velocity distribution and DE in cold spraying titanium which can be leveraged to optimize spray parameters and save raw material and cost.