A Machine Learning Technique for Prediction of Cold Spray Additive Manufacturing Input Process Parameters to Achieve a Desired Spray Deposit Profile

Cold spray additive manufacturing (CSAM) has recently gained increased attention from the research community and industries due to its several advantages, such as high metal deposition rate, low working temperature, and ability to deposit high-reflectivity materials. Despite these advantages, one of the main limitations of CSAM is poor dimensional accuracy of as-fabricated components. Hence, these components often require costly time-consuming postfabrication machining. Poor dimensional accuracy is mainly attributed to the lack of proper control of spray deposit profiles (SDPs). The shape of the SDP is mainly controlled by a set of input process parameters (IPPs), e.g., spray angle, nozzle standoff distance, and velocity. However, there is no systematic procedure or technique available to determine the IPPs a priori to achieve a desired SDP. In this article, we propose a machine learning-based hybrid multilayer perceptron and particle swarm optimization inverse modeling technique to predict a set of IPPs to achieve a desired SDP. Using a dataset obtained from the literature, with extensive simulation studies, we have provided evidence of accuracy, stability, and reliability of prediction of IPPs in terms of prediction error, convergence characteristics, and consistent results. Considering highly nonlinear complex mapping between the IPPs and the SDP, we believe the performance of the inverse model to be satisfactory.