Optimization of process parameters for controlling porosity in laser powder bed fusion metal additive manufacturing

Abstract

Additive manufacturing (AM) techniques, particularly laser powder bed fusion (LPBF), offer unparalleled design flexibility, precision, and material efficiency. However, porosity remains a significant challenge, impacting the mechanical properties of LPBF parts. Conversely, porosity can be desirable in some applications, such as biomedical implants and heat exchangers. Therefore, this research investigates a novel approach for optimizing process parameters to control porosity in the LPBF process of Ti6Al4V alloy. The methodology leverages the strengths of three powerful tools: ANSYS Additive Toolbox, MATLAB Deep Learning Toolbox, and the Artificial Bee Colony (ABC) optimization algorithm. ANSYS Additive Toolbox simulates the LPBF process, predicting porosity based on chosen process parameters. The MATLAB Deep Learning Toolbox constructs an Artificial Neural Network (ANN) model, essentially a machine learning representation, of the complex relationship between process parameters and porosity obtained from the ANSYS simulations. This ANN model predicts the porosity of the LPBF part as a function of four printing parameters: laser power, hatch spacing, layer thickness, and scan speed. Finally, the ABC optimization algorithm efficiently searches this ANN model to identify the optimal process parameter combination that achieves the desired porosity level in the parts. Various parametric studies demonstrate that the developed ANN model, utilizing 30 neurons in the hidden layer, effectively captures the correlation between printing parameters and porosity levels. These studies consistently show that porosity decreases as laser power increases for values of scan speed, hatch spacing, and layer thickness. Conversely, for the other three printing parameters, a positive correlation with porosity levels is observed. The correlation among selected printing parameters and porosity levels identified in this study aligns with experimental results and interpretations from the literature. Furthermore, multiple case studies highlight the efficiency of the coupled ANN model and optimization algorithm in accurately predicting porosity levels. Notably, the absolute error percentages remain exceptionally low, never exceeding 0.051% across all case studies, reinforcing the model’s reliability and precision. These results highlight the effectiveness of the proposed framework in optimizing LPBF process parameters for porosity control, offering a faster and more cost-effective alternative to traditional trial-and-error experimentation. This paves the way for producing high-quality LPBF parts with precisely controlled porosity.