| Abstract: |
Characterizing thermally sprayed coatings remains challenging due to the interplay between different operating and process parameters. Currently, there is no general framework for accurately predicting the coating characteristics under specific operating conditions. In this work, Artificial Intelligence models were employed to investigate a case study of generating superhydrophobic coatings by suspension plasma spray (SPS), an emerging thermal spray process that can produce coatings with micro-and nano-scale features. The approach aimed to relate key thermal spray process parameters such as plasma torch nozzle diameter, plasma power, sand-off distance, grit-blast effect, and suspension solvent type to different coating characteristics such as water contact angle, sliding angle, and surface roughness. Machine learning (ML) algorithms of both tree-based (ranging from linear regression, random forest, to improved gradient boost) and deep-neural-network models were investigated using a recent dataset of SPS experiments. Following the training of the ML models, selected algorithms were tested on unseen SPS data points at different operating conditions. The ML models were able to predict the sliding angles with good accuracy of over 95%. Finally, a generative adversarial network (GAN) is employed to generate realistic SEM images of SPS coatings with specific sliding angles. In addition, the generated SEM images by GAN are qualitatively and visually satisfactory, paving the way for a machine learning approach to controlling thermally sprayed coating microstructures.
|
|
