Erosion modelling using Bayesian regulated artificial neural networks

Modelling of the high temperature erosion behaviour of Ni-base alloys using artificial neural networks (ANNs) is presented. Two scenarios have been used: (i) a simple equation-based model and (ii) a comprehensive dataset looking at erosion as a function of particle size, velocity, impact angle and temperature. Common problems associated with ANNs are discussed within the context of erosion modelling. It has been found that the use of multilayer perceptron artificial neural networks for modelling erosion gave unreliable results when trained with traditional algorithms. The more recent Bayesian regularisation algorithm however has proved very successful, yielding both high Pearsonian correlation coefficients (r>0.95) and accuracies averaging better than 90%.