A PROCEDURE FOR JUSTIFYING NEURAL NETWORK BASED ADVANCED CONTROLAPPLICATIONS

摘要

For a number of years, papers have been presented concerning the successful application of model based softrnsensors, model predictive control, fuzzy logic control, rule based control, etc. However, with all of the manyrnsuccesses presented, very little attention has been given to the process of evaluating and justifying advanced controlrnprojects. Reasons given are: “this is a new technology, the savings are indirect and difficult to identify,” and manyrnothers.rnThis paper presents a methodology for calculating the potential savings for an advanced control project based uponrnclearly statistical methods. The savings addressed are real and directly related to real business issues: improvingrnfirst-pass first-quality yield, increasing throughput, increasing up time and reducing cycle time. These direct,rnauditable savings are derived by first identifying process variability and process capability and then calculating thernpotential savings that will result from the direct application of advanced modeling and control techniques.