One of the main characteristics of producing synthetic polymers is that the same process is used for the production of different kind of products (various molecular weights, compositions, etc.). Since the producers are forced to satisfy various demands of various costumers, frequent grade transitions are needed. These grade transitions are expected to be short and effective to avoid the production of so-called off-specification products. It became very popular to apply model predictive controllers (MPCs) to reduce the quantity of off-specification products, however most of them use linear models for prediction. Since polymerization reactions are highly non-linear, using linear models may cause significant difference between the response of the model and of the real plant and this can cause problems e.g. in predictive control. The difference appears mainly during grade transitions, hence it is important to tune the appropriate parameters of the regulators to realize the grade transitions as soon as possible. In this article a novel method - in the field of predictive control - is introduced for parameter tuning, although this method is well known in the field of experiment design. The statistical tools like design of experiments (DoE) permit the investigation of the process via simultaneous changing of factors' levels using reduced number of experimental runs. Through a case study the applicability of full factorial design is going to be examined. It will be proven that full factorial design is appropriate for finding the right tuning parameters of MPC controlled polymerization reactor. The aim of the case study is the reduction the time consumption of grade transitions, so applying the tools of design of experiments as a quasi-APC.
ASJC Scopus subject areas
- Chemical Engineering(all)
- Computer Science Applications