A multi-criteria framework for optimizing drilling conditions in AFRP composites based on a BPNN-GA intelligent framework

Aramid fiber-reinforced polymer (AFRP) composites are increasingly used in lightweight engineering structures due to their high specific strength, corrosion resistance, and impact tolerance. However, drilling AFRP remains challenging because interactions among tool geometry, spindle speed, feed rate, fiber reinforcement, and polymer matrix can generate excessive thrust force and torque, as well as delamination damage. This study proposes an integration of backpropagation neural network (BPNN) with genetic algorithm (GA) for multi-response optimization of AFRP drilling. Drilling experiments were conducted using an L18 Taguchi orthogonal array with two replications by varying tool type, spindle speed, and feed rate. Four parameters were evaluated, namely thrust force (F_z), torque (M_z), peel-up delamination factor (F_de), and push-out delamination factor (F_do). Independent BPNN models were developed for each response and subsequently used as prediction engines in the GA-based optimization. The developed models showed reliable predictive capability, with average prediction errors of 4.251%, 2.134%, 1.892%, and 1.009% for F_z, M_z, F_de, and F_do, respectively. The GA identified the optimum drilling condition as a twist drill at 1500 rpm and 25 mm/min, with predicted values of 37.82 N (F_z), 78.44 N·mm (M_z), 1.181 (F_de), and 1.087 (F_do). Confirmation experiments closely matched the predicted results, producing average values of 37.83 N, 78.45 N·mm, 1.182, and 1.088, respectively. These results validate the proposed BPNN-GA framework as an effective approach for determining damage-minimized drilling parameters in AFRP composite machining.