Prediction of Diameter Error of Workpiece in Turning Process Using Neural Network

Abstract

Multi-layer perceptron feed-forward neural network is adopted to predicate the diameter error of workpiece in turning process on the basis of the characteristics of diameter error. Turning experiment is designed to obtain the original training data and testing data. After analyzing the advantages and disadvantages of gradient descent algorithm and traditional genetic algorithm, gradient descent algorithm is incorporated into traditional genetic algorithm to constitute the hybrid genetic algorithm. Using training data, a multi-layer perceptron feed-forward neural network is trained by gradient descent algorithm, traditional genetic algorithm and hybrid genetic algorithm respectively, the convergence effect of hybrid genetic algorithm is better than that of gradient descent algorithm and traditional genetic algorithm, neural network that is trained by hybrid genetic algorithm is tested with testing data, the result is reasonable. The study turns out that neural network that is based on hybrid-genetic-algorithm is reliable to predicate diameter error of workpiece in turning process.