This project was a closed-loop real time implementation of the neural network thermal compensation on a camshaft grinder.  The software collected data from thermocouples affixed to the grinder body and read the machining error from a calibrated gage after each grinding operation.

The program incorporated a neural network to train on a data set, and after training it communicated with the CNC controller to compensate for the errors in real time.  
The system was able to reduce dimensional errors during machine warm up from several thousands of an inch to a few tenth of thousands of an inch.