A Meta-Learning Based Surface Roughness Prediction During SiCp/Al Composites Machining

Abstract SiCp particle-reinforced Al (SiCp/Al) metal matrix composites offer excellent mechanical properties but are prone to surface defects during machining due to microstructure non-uniformity. Accurate surface roughness prediction with corresponding surface feature identification is therefore essential to prevent surface degradation and performance loss. This article introduces a novel predictive model that uses surface topography images to identify surface features and predict roughness parameters. Our approach employs a prototypical network, a meta-learning method to decouple and classify multimodal features from machined surfaces. These extracted features are then linked to surface roughness values using empirical Bayesian learning. The model combines the capability of a prototypical network in the presence of a limited sample size with multiple classes and the strength of empirical Bayesian learning in handling high-dimensional data. Surface topography images capture machining uncertainties influencing surface quality, and their encoded features serve as intermediate response variables, enabling the model to generalize across datasets from varying conditions. These eliminate the need for prior knowledge of processing settings, making it ideal for dynamic industrial environments. The feasibility of the model has been validated through SiCp/Al composites milling experiments with topography data acquired via an optical ultra-depth microscope.