Learning a nonlinear model of a manufacturing process using multilayer connectionist networks

摘要: Control of a manufacturing process can be very risky when the is incompletely understood. The risk making adjustments deceased by building model and experimenting with changes to controls rather than those actual process. A connectionist (neural) network learns nonlinear observing simulated in operation. objective use estimate effects different control strategies, removing experimentation from Previously it was demonstrated that linear, single-layer learn as accurately conventional linear regression technique, advantage processes data they are sampled. Here, experiments multilayer extension presented. >