Neural network directed steel annealing
Patrick H. Garrett · 2018
Steel recrystallization annealing is demonstrated for continuous strip ductility reconstitution. This is achieved via a neural network ex situ planner, anneal net, that has been trained offline from previous strip coupons representing material hardness properties. Thermal annealing for restoration of cold-reduced steel strip ductility is a common steel production process amenable to improvement by means of computationally intelligent processing. Steel recrystallization annealing requires product property modeling capable of defining steel strip ductility from postprocessing sample testing. Anneal net is a feedforward network, with one hidden layer employing backpropagation for training weight values, w, that incorporate a momentum factor, alpha, to accelerate learning. The annealing process apparatus accommodates a cold-reduced steel strip in motion for thermal modification of its deformed crystalline grain microstructure. Regulation of in situ strip temperatures is consequently achieved by ten environmental feedback control loops for gas flow and one for strip speed adjustment.