Fuzzy and probabilistic techniques applied to problems of the chemical process industries.

摘要

This body of this work addresses two open questions. The first question concerns the validity of both fuzzy and probabilistic controllers. This question arises from the, sometimes heated, debate between the fuzzy logic and statistics communities on the use of probabilistic versus fuzzy methods to solve engineering process control problems. In some cases statements from either community can be interpreted to be simply “Our system is good and yours isn't.” The second question is “Why isn't fuzzy logic (or probabilistic) control used more in the Chemical Process Industries (CPI)”. This is an industry where the complexity and non-linearity of the problems often require human judgment and experience and yet a good technique for applying human judgment, fuzzy logic, is regularly ignored in favor of standard control designed for linear systems.; In order to address both of these questions, we have picked two different difficult control situations from the CPI and solved them with both fuzzy and probabilistic controllers. Where applicable we compare these techniques with proportional-integral (PI) control, the standard technique used in the CPI.; Since the first question is often extended to the statistical process control (SPC) arena, we, in addition, solve both a plant exposure control problem as well as a quality control problem' using both SPC and a fizzy analog.; The major accomplishments of this work are: (1) We have captured the expert knowledge of gentleman with broad experience in oil field clean up and made it the basis for a state-of-the-art fuzzy control system for a three-phase oil field centrifuge. The system includes feedback, and feed forward control, a fuzzy soft sensor and a fuzzy SPC filter, clearly demonstrating the usefulness of fuzzy control in the CPI. (2) We have demonstrated with both the centrifuge problem and several liquid-level control problems that there is essentially no difference between probabilistic and fuzzy control solutions. Where applicable we have demonstrated that both of these systems can outperform standard techniques and can be useful to the CPI.