Comparative study of I-optimal design and definitive screening design for developing prediction models and optimization of average surface roughness of PLA printed parts using fused deposition modeling

In fused deposition modeling (FDM), the prediction and optimization of surface roughness distribution by varying the process parameters of the printing is required during the process planning stage. During this stage, the traditional screening design, such as fractional factorial design, is commonly used to identify the process parameters that have the most considerable effect on process outcomes. The screening design is followed by further experiments, including only essential process parameters to develop prediction models and identify the optimal process parameters. Recently developed custom design approaches such as I-optimal design and definitive screening design (DSD) eliminate the need for follow-up experiments by identifying the critical process parameters and optimum process conditions in a single experimental design. Therefore, the present study is intended to compare the performance of I-optimal design and DSD in terms of prediction and optimization of average surface roughness (Ra) of fused deposition modeling (FDM) printed parts from poly lactic acid (PLA). The following process parameters were considered to reduce the prediction error of Ra models: layer thickness, number of contours, infill density, raster angle, printing speed, extrusion temperature, bed temperature, and built orientation. The results revealed that regression models based on the I-optimal design are saturated and complex, with 38 terms (including eight main effects, twenty three interaction effect, and seven quadratic effects) in the final model. However, the DSD model has 12 model terms (including seven main effects, two interaction effects, and three quadratic effects), making it unsaturated and less complex compared to the I-optimal design. Both models have comparable prediction accuracy based on validation tests. Finally, Ra is optimized based on the desirability function. The results of the present study will add knowledge to the existing literature about the performance of I-optimal and DSD methods in predicting and optimizing responses.