Prediction of internal surface roughness in drilling using three feedforward neural networks - a comparison

The internal surface roughness in drilling has long been realised as one of the essential quality characteristics to be precisely controlled and monitored. As an artificial intelligent tool, feedforward neural networks are purposed to quantitatively predict internal surface roughness. In this paper, comparison of the novel network, the optimisation layer by layer (OLL), with common feedforward networks, backpropagation and radial function basis networks were carried out on this application. Three variations of network architectures were systematically optimised and benchmarked on the basis of accuracy and speed. RMS error and computing time were used as yardstick indexes to select the optimal feedforward neural network. The selected OLL network has shown superior and numerical results in the test stage which confirms the predictive capability of the neural network based approach.