Instant flow distribution network optimization in liquid composite molding using deep reinforcement learning

Abstract Carbon fibre reinforced plastic (CFRP) manufacturing cycle time is a major driver of production rate and cost for aerospace manufacturers. In vacuum assisted resin transfer molding (VARTM) where liquid thermoset infused into dry carbon reinforcement under pressure, the design distribution network to minimize fill while ensuring preform completely full critical achieving acceptable quality time. Complex networks in composites increase need quick, optimized virtual feedback. Framing problem flow media placement terms learning, we train deep neural agent using 3D Finite Element based process model preforms. Our learns place on thin laminates order avoid starvation reduce total infusion Due knowledge has gained during training variety laminate geometries, when presented with new geometry it able propose good layout less than second. On realistic part complex 12-dimensional network, demonstrate our method reduces by 32% compared an expert designed placement, maintaining same quality.