Application of Machine Learning Techniques in the Prediction of Surface Geometry

Area of Science:

• Materials Science
• Computational Science
• Surface Engineering

Background:

• Tungsten carbide-cobalt-aluminum oxide (WC-Co-Al2O3) coatings offer superior hardness and abrasion resistance.
• Existing statistical models struggle with the complex, non-linear nature of surface topography data.
• Digital representation of surface layers is crucial for predicting material performance and optimizing manufacturing.

Purpose of the Study:

• To develop an advanced method for generating digital surface representations of WC-Co-Al2O3 coatings.
• To integrate machine learning (ML) with statistical approaches for modeling stochastic surface geometries.
• To compare the proposed ML-stochastic hybrid model with traditional methods and emerging techniques.

Main Methods:

• Profilometric analysis to collect experimental surface data.
• Development of a hybrid model combining Recurrent Neural Networks (RNNs) and Monte Carlo simulation.
• Review of Generative Adversarial Networks (GANs) and Expectation-Maximization (EM) algorithms for stochastic simulation and parameter estimation.

Main Results:

• The ML-stochastic hybrid model effectively captures both deterministic and random features of WC-Co-Al2O3 surface geometries.
• Demonstrated the capability to generate series of digital surfaces with similar geometric parameters.
• Highlighted the limitations of traditional models (ARMA/ARIMA, HMMs) in handling complex surface data.

Conclusions:

• Machine learning-stochastic hybrids offer a powerful approach for modeling and generating complex surface geometries.
• The study confirms the effectiveness of the proposed RNN-Monte Carlo model for WC-Co-Al2O3 coatings.
• Future research should focus on physics-informed ML and explainable AI to address computational demands and interpretability challenges.