LET THE PRODUCT BE THE JUDGE
In April, at a Sematech conference in Dallas, I attended a lecture by Robin Barnes of Bycom Solutions, St. Peters, Mo. The talk touched on some philosophical considerations relevant to control of the manufacturing process. For one, Barnes suggested that there's more to control than the latest PID algorithm.
She outlined several possible approaches to better product and process, in particular, predictive control involving modeling and nonlinear relationships. To me, most interesting was consideration of control system configurations that used product feedback as the primary loop, which process feedback serving to stabilize local loops.
Perhaps some definitions are in order. Four basic control and parameter considerations. Production control is are the measurement of the process attributes. Last, and unfortunately too often ignored, are product parameters, which link product attributes back to the manufacturing steps.
Product feedback gives managers some sense of process quality. Time-to-market and quality are but two possible areas that can be improved by better product feedback. Improvement happens when information relevant to the hierarchical control scheme for each production step, focusing the impact of each process step on the final product. In addition, values such as temperature, flow and force should be directly related to their product values - color, consistency and hardness, for example.
Predictive modeling examines for aspects of a process - those that are controllable and observable; those controllable but not observable; those observable but not controllable; and those unobservable and uncontrollable. Classical process control only deals with linear aspects of the controllable and observable, and usually only by means of feedback. In other words, an error must exist before corrective action can be taken. Classical control normally involves linear approximations of nonlinear processes, since most process are nonlinear. This can lead to large product and process errors. (I can see the letters coming in from vendors already. Before going any further I should say, there are products that already do modeling and nonlinear control, but too often only as patch to the classical system.) Looking at all possible processes, classical systems apply only to a small subset. Even the ubiquitous PID loops use linear algorithms.
Models are better suited for dealing with the unobservable. Models predict future deviations. Wouldn't it be better to make the starting point for process control efforts a model of the true characteristics of the process, one that allows for nonlinear control schemes? The classical approach would thus be a subset of the mode inclusive system.
Recent trends of using modeling in real-time control are promising. Modeling proves most useful for processes that have multiple inputs and outputs nonlinear aspects, that are intolerance of any error, and that must link product parameters. Modeling and product feedback provide great benefits, especially when simply reacting to deviations is not enough.
Localized process optimization can distort overall production values, leading to poor products. So it's really not overly obvious to emphasize that good products are the whole point. Modeling ensures the entire process is the issue, not some group of loops that's become someone's private obsession.
In general, many of the precompetitive consortia attending the conference seemed interested in process control. They are also interested in setting standards for open systems. Most, however, are only interested in the standard concepts of linear loop control. The majority of the attendees talked about that which all our efforts are aimed at: better products.
As appeared in Manufacturing Systems Magazine June 1993 Page 10
References - Table of Contents
|R. Morley Incorporated
614 Nashua Street, Suite 56
Milford, NH 03055-4992 USA
Tel: 603-878-4365 FAX: 603-878-4385