Machine learning regression models for predicting mechanical properties of coatings deposited via HiPIMS in DOMS mode

Abstract

High-Power Impulse Magnetron Sputtering (HiPIMS) in Deep Oscillation Magnetron Sputtering (DOMS) mode is an advanced technique for depositing thin films with tailored mechanical properties. Predicting and optimizing film thickness, hardness, and Young's modulus are crucial for enhancing material performance in demanding applications. This study develops machine learning regression models to simultaneously predict these properties of coatings deposited via HiPIMS in DOMS mode.

A dataset comprising 66 data points from previous studies was compiled, including deposition condition and target parameters. Support Vector Machines (SVM), Gradient Boosted Trees (GBT), and Gaussian Process Regression (GPR) were employed to build predictive models. The GPR model demonstrated superior performance, effectively capturing complex, non-linear relationships. Key factors influencing mechanical properties were identified. Deposition pressure significantly affected all properties. For film thickness, deposition pressure, deposition rate, and target density were most influential. In predicting hardness, deposition rate, pressure, and nitrogen content were significant. The target material's Young's modulus had a strong impact on predicting the film's Young's modulus, indicating dependence on the target's intrinsic properties.

To validate the models, titanium films were deposited under varying peak power conditions. The titanium data, not included in the training set, served as an independent test. The GPR model accurately predicted the mechanical properties of these films, confirming its applicability to new materials.

This study demonstrates the effectiveness of machine learning models, particularly GPR, in predicting mechanical properties of coatings. The models provide valuable insights into optimizing deposition processes, contributing to the development of advanced coatings with improved performance.