THE PREDICTION OF LASER CLAD PARAMETERS BASED ON NEURAL NETWORK

Because laser cladding is a multi-variable coupling process, the relation between clad bead and process parameters is non-linear with multi-objects and multi-variables. Moreover, it is very difficult to find out an exact analytic model to express this relation. In this study, a neural network model, which is often used for non-linear problems, is developed to explain the relationship between process variables and clad parameters (the width and height of cladding bead). Some samples obtained in experiments are used to train the network model to form the perfect map relation between input and output, and other samples are employed to test the network model. A normal BP algorithm and an amend BP one are trained, and it is found that the amend BP algorithm has advantages over the normal BP one. The experimental results agreed well with those calculated with the neural network model, which indicates that the developed BP neural network prediction model is feasible and valid in theory and in practice.