Model of grinding-induced line/area roughness and subsurface damage in brittle material based on genetic algorithm and deep neural network

Post processes are usually needed to improve the quality and performance of ground brittle materials, and their low efficiency and high cost are greatly determined by grinding-induced roughness and subsurface damage (SSD). This raises an urgent demand to accurately predict various roughness and SSD depth. In this paper, grinding experiments are conducted on K9 glass samples with different processing parameters, including abrasive grain diameter, grinding depth, wheel speed, and feed speed. The line roughness Ra, area roughness Sa, and SSD depth are measured. Based on genetic algorithm (GA) and deep neural network, a relationship model among processing parameters, Ra, Sa, and SSD depth, is established. The model is accurate and reliable with a mean absolute percentage error MAPE < 10% and a correlation coefficient R > 0.94. The research is valuable in the evaluation of surface and subsurface integrity for ground brittle materials.