Hybrid data-driven feature extraction-enabled surface modeling for metal additive manufacturing

Metal additive manufacturing (AM) has become popular in a large variety of applications due to its excellent capabilities of handling complex geometries and novel materials. However, due to its process complexity, layer-wise surface quality issue is still one of the critical concerns to further broaden adoption of metal AM, because of the impact on products’ property and functionality. The existing experimental studies from literature have shown machine parameters could significantly affect the resulting surface morphology of printing products. Consequently, it is urgently necessary to analyze and model printing surface in metal AM, and thereby printing surface can be further correlated with machine parameters, enabling more appropriate quality assurance applications such as process design and post anomaly detection. However, there are two major practical challenges to realize this goal: (1) the printing surface profiles in metal AM are highly nonlinear; and (2) the measured surface profiles usually have significant outliers, shifts, and porosities. To address these two challenges, this paper models surface profile in a decomposition-based framework and develops a hybrid data-driven feature extraction approach, which integrates a robust convolutional autoencoder-based approach and conventional statistics-based approach. Through the incorporation of supervised machine learning algorithm, the underlying relationship between machine parameters and printing surface can be thereby clearly quantified. To validate effectiveness of the proposed method, both simulation and a real-world case study in laser-engineered net shaping (LENS) were conducted in this work. The results demonstrate that the classification accuracy using the proposed method could achieve 86% in simulation cases and 74% in an actual LENS experiment, which outperforms the benchmark methods with better robustness. Therefore, it demonstrates that the developed approach is very promising for surface morphology analysis and process optimization of metal AM.