A method for developing predictive models of quality metrics and gas flow variables for 316L PBF-LB/M printed components based on image analysis

Powder bed fusion of metals by a laser beam (PBF-LB/M) is more commonly used than ever in a commercial setting. This requires more focus on product quality, homogeneity, and reproducibility. To obtain this, fast predictive models can be used. These models can give an indication of the approximate properties of the products before they are finished. This study aimed to develop predictive models for selected quality metrics and gas flow variables. The models were developed by analysis of images captured during a build of PBF-LB/M printed components in 316L stainless steel. The images were corrected for lighting inconsistencies followed by component-wise quantification of mean pixel intensity which formed the basis for the predictive models generated using JMP Pro 16. Response screening analyses were done to identify significant correlations between quantification methods and quality metrics or gas flow variables. A correlation between the quality metric bulk O2, which is the oxygen uptake in the bulk material, and the quantification method "mean value bulk"was found. Additionally, a correlation between the gas flow variable oxygen percentage in the build chamber and the "mean value circular"quantification method was found. Thus, two predictive models were developed. The model for bulk O2 could not be validated as further data processing was needed for the remaining components used for validation. This was mainly due to time and equipment limitations. The oxygen percentage model was tested but seemed unusable as the predictions were inaccurate. This result is likely caused by depositions of process by-products on the surface of the studied components.