Estimation of coating thickness in electrostatic spray deposition by machine learning and response surface methodology

Abstract To improve the quality and productivity of the process or system before resorting to expensive and laborious experimental tests, it is essential to model and predict the system performance concerning its operational parameters. Predictive modeling and parameter optimization through machine learning techniques has been the most advantageous process and are the best alternative to the conventional statistical tools. In this work, carbide cutting tool inserts were coated with molybdenum disulfide (MoS2) solid lubricant utilizing the electrostatic spray deposition (ESD) process. The optimum artificial neural network (ANN) model with 3-6-6-1 architecture includes 0.6 momentum term and 0.3 learning rate with attained mean squared error (MSE), absolute error in prediction (AEP) of trained and test data are 0.000334, 0.197, and 0.543, respectively. The support vector machine (SVM) hyperplane parameters are optimized using the Bayesian optimization technique, and after 90 evaluations, the model with the least error is used for predicting ESD coating thickness. The coating thickness predictions from ANN and SVM models were related to the response surface methodology (RSM) model predictions. From the results presented, the correlation coefficient (R-value) between experimental results and model predictions for ANN and SVM are 0.979 and 0.991, respectively, whereas, for RSM, it is 0.919. In addition, a genetic algorithm (GA) has been employed to establish the optimum conditions for the ESD deposition parameters. The presented SVM and GA method would support rapid and precise estimate and optimization of coating thickness in the ESD process.