The neural network project was sponsored by ARPA to determine the feasibility of using artificial intelligence to compensate for thermally induced machining errors.  The test setup consisted of a 30 inch copper tube affixed with a couple dozen thermocouples.  Fiber optic non-contact probes measured the displacement of the tube relative to a fixed Invar rod while being randomly heated by computer controlled industrial heat guns.  The displacement and temperatures were used to train a 3 layer back propagation neural network and the same network was then used to predict future displacements under similar conditions.  The same concept was then used to outfit a CNC lathe and the machining errors were predicted as a function of the temperature profile of the lathe.  The network was able to correctly predict errors in both setups to about a 95% accuracy.