Issues of Uncertainty in Plant Operations

The general objective of this work is the better understanding of the behavior of process systems under uncertainty conditions. This involves quantification of uncertainty, uncertainty propagation, system characterization under variability of conditions, incorporation of uncertainty at the decision making level and the development of the appropriate solution methodologies to address these problems efficiently.

This project focuses first on the development of new ways to quantify the performance of the system under uncertainty, and techniques of uncertainty propagation, and second the application of such uncertainty analysis tools within the plant operations. This encompasses scheduling, planning and supply chain management for which modeling efforts are being currently developed in my group.

Existing approaches for uncertainty analysis either are two expensive computationally (e.g. Monte Carlo techniques) or are too limited in terms of the amount of information that one can obtain (e.g. flexibility analysis). A well-recognized technique that targets on a similar objective is the Six Sigma analysis tool. What this tool achieves is to measure not only the nominal process performance but also its whole distribution that can be quantified by its variance (sigma). It is well known that moving towards the tail of the distribution the probability of fault behavior is getting smaller, thus process that are characterized by a large range of distribution have very small probability for "bad" performance and consequently they are much more profitable.

Our work targets the development of efficient and systematic techniques to quantify the range of variability of systems given uncertainty in process parameters. A comprehensive approach will take into account specific system's behavior and model predictive techniques so as to identify a critical set of required experiments. The work will be extending on developing efficient propagation techniques for complex dynamic systems. Stochastic Surface Response methods will be used as an alternative to traditional Monte Carlo to reduce the enormous computational complexity. Extensions of these techniques to include time and spatial variability explicitly will be also investigated. The result of this work is of
great significance for complex systems. Although each of the units might be tested and designed to operate under Six Sigma specifications, the interactions between the different processes and further input variability should be taken into account to ensure the overall plant performance. Uncertainty propagation techniques would be of great help towards that direction.

The application of the developed stochastic methodologies to handle uncertainty to operations problems such as scheduling, planning and supply chain management will be the focus of the second part of this project. The specific tasks involve:

• the development of a methodology to determine a robust schedule capable of meeting the expected range of uncertain parameter values (i.e. demand, processing times, prices, raw material availability)
• the extension of reactive scheduling methodology to planning and supply chain management problems
• the development of rigorous decomposition based techniques to address process operations problems under uncertainty. This step is very crucial since most of the existing approaches to deal with uncertainty can address very limited size problems.
  

marianthi@sol.rutgers.edu
04/23/02