Modeling and Simulation of Screw Axis Based on PSO-BP Neural Network and Orthogonal Experiment

It is a trouble thing to build theoretical model for stir characteristics of screw axis with variable diameters and different pitches, so a called PSO-BP neural network (NN) model was employed. In this mode, particle swarm optimization (PSO) algorithm is used to train weights and thresholds of artificial neural network instead of BP algorithm, to overcome drawbacks of BP algorithm. To avoid the slow search speed around global optimum in the PSO-BP algorithm, a heuristic way was adopted to give a transition from particle swarm search to gradient descending search. To validate the model, a group of orthogonal experiments were designed and performed, and simulation experiment was carried out with the NN model. Simulation and experiment results indicate that PSO-BP NN is an effective training algorithm, and it provides an effective modeling approach of screw axis with variable diameters and different pitches.