FEM-based neural network modeling of laser-assisted bending

Abstract

An artificial neural network (ANN) model of laser-assisted bending is developed based on the data obtained from a finite element method (FEM) model. FEM model is validated with the experiments. In the experimental setup, the sheet is clamped like a cantilever beam with mechanical load on the free end. A laser beam scans the bend line for reducing the flow stress of the material locally. ANN considers four process parameters viz., laser power, mechanical load, distance of the scan-line from the free end, and scan speed. The ANN predicts the most likely, upper, and lower estimates of the bend angle. Commercial package MATLAB® is used for developing the ANN model. First, a multilayer perceptron (MLP) neural network is trained and tested using the data from the FEM model. Using the best fit model of the MLP, additional data are generated for training a radial basis function (RBF) neural network. The RBF neural network is tested and validated against experimental data and subsequently used for obtaining the most likely, lower, and upper estimates of bend angle as a function of the considered process parameters. A parametric study is also carried out based on the RBF neural network model. Finally, it is shown that the neural network can be used for the inverse prediction using the bisection method.