The design of alloys with tailored tribological response is not trivial given that intricate material and system properties combined with surface-driven mechanocatalytic interactions heavily impact the resultant friction behavior. Thus, fully mapping the performance space of a material system requires resource-intensive experiments spanning multiple modalities and system variables. In this work, we address this challenge by developing a Bayesian optimization-based approach to efficiently and effectively determine the structural characteristics and manufacturing parameters of Pt-Au films that yield a desired friction behavior. A critical component of the developed framework is an accurate and computationally efficient multimodal surrogate model, based on an encoder-decoder regression architecture. The model, trained on both experimental and simulated data, utilizes nanoindentation-based reduced modulus and hardness measurements, X-ray fluorescence (XRF) spectra, and SimTra analysis to accurately predict the friction behavior of a thermally stable nanocrystalline alloy system (Pt-Au) used as a tribological coating. Adequately integrating this model with Bayesian optimization enables one to effectively identify the Pt-Au alloy film parameters (i.e., material and processing properties) that will exhibit dynamic friction behavior closely matching desired specifications. While ensuring physical consistency with experimental data, the Bayesian optimization framework significantly accelerates the material design process.
Brown et al. (Mon,) studied this question.