Smart process mapping of powder bed fusion additively manufactured metallic wicks using surrogate modeling

Powder bed fusion is an innovative additive manufacturing (AM) technique to achieve metallic wick structures for efficient two-phase thermal management systems. However, a technical challenge lies in the lack of standard process maps as it currently relies on an expensive trial and error approach. In this study, five types of surrogate models for classification analysis (i.e., naïve Bayes, logistic regression, random forest, support vector machine, and Gaussian process classification) were constructed and compared to efficiently unlock the relations between five process parameters (i.e., laser power, scan speed, hatch spacing, spot diameter, and effective laser energy) and wick manufacturability. The models were trained using data from a total of 187 AM wick manufacturability experiments. Using four process parameter (PP) model (five PP model without effective laser energy), the Gaussian process classification (GPC) showed the maximum median prediction accuracy (PA) of 93%, while it further improved to 99.7% using support vector machine (SVM) and five process parameter model. Also, the median PAs of the SVM and GPC remains above 98.5% with only 60% of the total experimental data using five PP model. The sensitivity analysis showed that the hatch spacing was the most sensitive parameter for the wick manufacturability using four PP model, while the effective laser energy is the most sensitive one using five PP model. This study provides insights into the smart selection of optimal process parameters for the desired metallic AM wicks.