On-line optimization of the Tennessee Eastman challenge problem

At the fall AlChE annual meeting In 1990, Downs and Vogel presented a complex control and optimization problem for continuous chemical processes based on a Tennessee Eastman company recycle reactor process. The abbreviation TE will be used to refer to Tennessee Eastman throughout this thesis. The Tennessee Eastman (TE) process is an open loop unstable process consisting of four process units: an exothermic two-phase reactor, a flash separator, a reboiled sthpper, and a recycle compressor. The process is operated either at a fixed production rate or maximum production rate for three different product grades. A number of known and unknown process disturbances affect the feed compositions, feed flows, and reaction properties.

The TE process is open-loop unstable due to the presence of exothermic reactions and a recycle loop. Open loop step tests used for process identification also cause the process to become unstable. Most multivahable process identification techniques require the use of open loop responses. Due to the number of available measurements (41) and manipulated vahables (12), a large number of control structures are possible (Banerjee and Arkun, 1995). Thus, a number of control strategies ranging from linear control to nonlinear control to the trendy neural network have been tested on the TE process simulation over the past six years. Testing the robustness of control strategies was only part of the challenge established by Downs and Vogel. In the TE process.

Increased interest in on-line plant operation optimization has been supported by increasing global competition, increasing feed costs, and increasing product quality requirements. However, the field of on-line optimization is relatively new and applications to the TE process have been sparse. This research mainly focuses on the application of steady state, model based, on-line optimization to the TE process.