A Transputer Based Neural Network Development System for Industrial Control Applications

Once they have been configured, Neural Networks provide, often the only means of solving difficult control problems. However, the time and computing power required to set-up the network co-efficients can be prohibitive. This paper reviews a project to investigate the use of parallel processing, in particular transputers, to evaluate the parameters for Neural Networks. Three different algorithms are used to parallelise the learning process and their relative advantages are discussed. Back propagation by epoch with the training set distributed between the transputer is found to be most suitable for the types of control problems investigated