Prediction of drilling performance of Eggshell–Iron dust reinforced hybrid epoxy composites using deep neural network

High filler content hybrid polymer composites create complicated drilling behavior, which is attributed to stiffness dissimilarity, fracturing of brittle particles, and spatially non-uniform stress transfer. This research article presents a report on the performance of drilling and the formulation of predictive equations of an epoxy matrix that is reinforced with 15 wt.% eggshell-derived CaCO ₃ and 35 wt.% iron dust (total filler: 50 wt%; matrix phase: 50 wt% epoxy resin-hardener system, summing to 100 wt% in total). The study of drilling trials was modeled based on the Taguchi L25 orthogonal array that allowed independent variation of spindle speed, feed rate, and measurement of the outcomes of drilling, delamination factor, circularity error, and cylindricity deviation. The results of multivariate statistical analysis showed that feed rate was the most influential variable, as it explained more than 60% of the variance in delamination and greater than 55% of the variance in geometric deviation. SEM of machined surfaces indicated a homogeneous distribution of fillers and an absence of interfacial voids, which supports damage mechanisms based on the cracking of the matrix, particle pull-out, and brittle CaCO ₃ particle fracture. A deep neural network (DNN) was later developed to reflect the nonlinear process-response associations, with an excellent predictive accuracy (R ² ) over standard regression and shallow artificial neural network designs. Confirmation experiments at the optimal condition of 1800 rpm spindle speed and 0.05 mm/rev feed rate yielded prediction deviations within ±3% for all responses. The experimental-computational method that has been developed herein is a viable approach to quality-based optimization of the drilling of sustainable hybrid composites containing high amounts of filler.