Multi-sensor integration using neural networks for predicting quality characteristics of end-milled parts: Effects of training parameters and learning rules

The current intense international and domestic market competition has forced manufacturers to look towards the automation of manufacturing systems as a means of increasing the quality and rate of production of manufactured goods. For the successful realization of the unmanned factory of the future, it will be necessary to achieve automated real time monitoring and control of manufacturing processes. It is therefore necessary to develop a real time machining monitoring and diagnostic system that will integrate quality assurance with manufacturing processes. This research investigates the use of artificial neural networks as a means of integrating multiple sensor signals for on-line real time monitoring of manufacturing processes. The networks are part of a proposed intelligent machining monitoring and diagnostic system. The research involves investigating the effects of the learning parameters and learning rules on the performance of backpropagation networks, and the prediction of quality characteristics of machined parts using neural networks. The Objective of the research is to determine the optimum combination of these parameters that will optimize the performance of the trained networks. In the first part of this work the effect of varying one of the learning parameters, while the other parameters are kept constant, on the performance of backpropagation networks is systematically explored and documented for end-milled 4140 Steel parts. In the second part of this work the interaction effects of varying more than one of these factors at a time is examined for the same 4140 Steel parts. The third part of this work involves a demonstration of the actual application of the results, of the first two parts of this work, for predicting the surface roughness and bore tolerance of end-milled 6061-T6 Aluminum parts --Abstract, page ii