Neural Network Algorithm Optimization of Casting Performance of New Building Weathering Steel

In order to optimize the casting performance of a new building weathering steel, taking alloy elements, alloy element addition, melting temperature, standing time and pouring temperature as input layer parameters, and corrosion potential as output,tansig function as hidden layer transfer function and purelin function as output layer transfer function, a neural network optimization model for the casting performance of the new building weathering steel with a four-layer topology of 5×30×6×1 was constructed and learning training and prediction verification of the model were carried out. The results show that the model has better prediction ability and higher prediction accuracy, the relative prediction error of the model is between 3.57% and 5.02%, and the average relative prediction error is 4.24%. The new building weathering steel optimized by the model is 09MnCuPTi steel with 0.3% Ce, melting temperature of 1630℃, standing time of 30 min and pouring temperature of 1600℃. Compared with that of 09MnCuPTi building weathering steel, the corrosion potential of the optimized new building weathering steel moves from -676 mV to -543 mV, and shifts positively by 133 mV, and the corrosion resistance is significantly improved.