A Thermal Displacement Prediction System With an Automatic LRGTVAC-PSO Optimized Branch Structured Bidirectional GRU Neural Network

Abstract

Considering technology’s rapid development, traditional manufacturing methods are insufficient to achieve the high accuracy demanded by aerospace, national defense, and numerous leading-edge engineering projects. Thermal displacement is a significant source of manufacturing errors, and accurately correcting such errors is difficult or even impossible using traditional machining methods. This article proposes a machine learning method for high-accuracy error prediction that nonprofessionals can easily implement. An optimized automatic logistic random generator time-varying acceleration coefficient particle swarm optimization (LRGTVAC-PSO) method is proposed to optimize a branch structured bidirectional gated recurrent unit (GRU) neural network. The accuracy of the proposed method (with a three-axis average of 0.945) is superior to that of the other optimized algorithms evaluated in this study. The method serves as a means not only of accurately predicting thermal displacement but also of autotuning the hyperparameters of machine learning algorithms.