Grinding roughness prediction model based on evolutionary artificial neural network

Due to the multiple influence factors and difficulty in measure of external cylindrical grinding roughness, an open experimental system with a roughness model was developed. Based on analyzing the Back Propagation(BP) network's disadvantages of low convergence speed and frequently falling into local minimum value, a roughness prediction model of BP neural network integrated with genetic algorithm was proposed'. The total search capability of Genetic Algorithm (GA) was used to optimize initial weight values and threshold values of the BP neural network. The methods for determining every parameter relevant to genetic algorithm and BP neural network were demonstrated in detail. The prediction performances of BP model and GA-BP model were compared under the same network structure. Four different GA-BP network structures were considered according to empirical formula for calculating hidden layer node amount. The best structure was finally determined from prediction accuracy inspection of four models. Experimental results showed that the integration of genetic algorithm and BP network could improve the convergence speed and prediction accuracy of roughness model, and could meet the steep demand of intelligent grinding on prediction efficiency and accuracy.