COLD SPRAY OPTIMIZATION VIA MACHINE LEARNING

Abstract

Additive manufacturing can be leveraged by the Department of Defense to reduce the financial and temporal cost of logistics. Cold spray has been shown to produce high-quality, cost-effective parts in a limited time frame. Cold spray is a developing technology and has not been fully optimized. Machine learning can be used to optimize cold spray parameters.To create a machine learning model, many cold spray prints were performed using varied parameters. These parameters included print parameters like temperature and pressure as well as properties of the powder like yield strength and particle size. Some of these prints were performed using composites of alumina powder and AA7075. Other prints were pure AA6061 or AA7075. Highest deposition efficiencies for the materials varied from 3.63% using AA7075 (80%) and Al_2 O_3 (20%) to 43.62% using pure AA6061. Pure AA7075 reached 33.72% deposition efficiency. The selected parameters for every material and print were fed into a multilayer perceptron. Once trained, a model was created to predict which adjustable parameters were most likely to produce the highest relative deposition efficiency for a given material. This model is highly accurate for the tested materials, reaching a 99.43% proper correlation, and can be used with confidence on similar aluminum powders. The collected dataset can be expanded upon to include new materials or different cold spray machines.