Surface Roughness Prediction for Aluminum Alloy Wheel Surface Polishing Using a PSO-Based Multilayer Perceptron

Reducing surface roughness is one of the most important aims for aluminum alloy wheel polishing. To predict the final surface roughness, a multilayer perceptron (MLP) is introduced into it. The prediction model using MLP based on particle swarm optimization (PSO) algorithm is implemented in order to avoid the local infinitesimal defect and slow constringency in the classical back propagation (BP) algorithm. Both the activation function parameters in the hidden layer and connection weights are optimized by PSO algorithm. Inputs to the MLP consist of normal polishing force, feed rate, the peripheral velocity of polishing tool, effective radius, and polishing times. The output is only the final surface roughness and only one hidden layer is used. The Sigmoid activation function with a variable parameter in the hidden layer is adopted. The prediction result shows that the MLP based on PSO can fit the testing samples well.