(Dr. Steven Liang, advisor)

"Neural Network" Based Process Monitoring

Abstract

The pressure on manufacturing industries to improve yield, mass-production and above all increase profits makes it essential to increase the efficiency of manufacturing process. Ways to monitor and control the manufacturing processes are constantly investigated. Monitoring on-line avoid turning off the machine and reduce the machine inactivity and thus increase the productivity. While some are developing models that in a certain time become non-valid due to machine wear, other work on evolutionary model based on self-learning paradigm such as neural networks. This seven-month study shows a new application for these networks: how to measure the out-of-roundness of a work piece during machining in turning thanks to a single accelerometer. The first period of work was dedicated to create a piece of software in Visual Basic to simulate a neural network using the back-propagation algorithm. Afterwards the network was presented numerical values to carry out a convergence on a mathematical function. Finally the above manufacturing issue was studied and solved with an average error around 7% which really open a possible set up in manufacturing industry.