summary of a seminar by John E. Rijnsdorp (emeritus professor University of Twente, Netherlands), to be held on 29^th May, 1998.

Traditionally, control systems are designed for maintaining process variables near given values. But product yield and/or production efficiency can be improved by adapting these values to changes in process conditions. In real-time this is usually realized by a model-based upper control layer, which adjusts set points at the lower traditional control layer. However, by adapting the control structure as well, optimization is faster and more robust for model errors.

For batch processes, this approach leads to "flexible recipes", whereby set point trajectories are approximated by first-order or second-order polynomials. These recipes can extend over several parts of the total production chain, so coordination is required between different local control systems.

Optimization in-the-large is a challenge at production sites, where production plants share utilities. Here on-line linear programming models can be utilized to pinpoint bottlenecks and to adapt plant throughputs to changes in conditions.

In all cases, human factors play a critical role.