Roughness Prediction and Performance Analysis of Data-Driven Superalloy Belt Grinding

Abstract

Abrasive belt grinding has unique advantages in avoiding machining defects and improving surface integrity while grinding hard materials such as superalloys. However, the random distribution of abrasive particles on the abrasive belt surface is uncontrollable, and chatter and machining errors accompany the machining process, leading to unclear mapping relationship between process parameters and surface roughness, which brings great challenges to the prediction of surface roughness of superalloy. Traditional empirical equations are highly dependent on empirical knowledge and the development of scientific theories and can only solve problems with relatively simple and clear mechanisms, but cannot effectively solve complex and mutually coupled problems. The method based on data-driven patterns has a better idea for mining the implicit mapping relationship and eliminating the uncertainty of complex problems. This study presents a data-driven roughness prediction method for GH4169 superalloy. First, a superalloy grinding platform is built. According to the grinding empirical equation, the mapping relationship between process parameters and surface roughness is analyzed, and a prediction model is established based on the error back propagation (BP) algorithm. Second, genetic algorithm (GA) and particle swarm optimization (PSO) algorithm are used to optimize the weights and thresholds of the neural network, and the global optimal solution is obtained. Finally, the prediction performance of different algorithms is compared. The results show that the non-uniform absolute errors of the BP algorithm, GA-BP algorithm, and PSO-BP algorithm are 0.12, 0.085, and 0.078, respectively. The results show that the roughness prediction algorithm based on PSO-BP is more suitable for GH4169 superalloy.