Control system monitoring and diagnosis are important to ensure that plant performance is close to optimal and that process variable data used in control algorithms is accurate. We are researching performance monitoring of feedback control algorithms for linear processes, and have developed methods that can handle single loop PID, model predictive control and multiloop control with or without constraints.We are also investigating the monitoring of process and sensor faults when variations in duration of batch steps occur. Both data-driven and physical models are being employed. Applications to semiconductor manufacturing and bioreactors are being studied.
Microelectronics manufacturing is an area where process modeling and control are receiving increased attention. We are carrying out a number of projects in cooperation with semiconductor companies. In semiconductor fabs run-to-run behavior can be influenced by reactor aging, first wafer effects, and other non-uniform processing conditions. In one project we are modeling the run-to-run behavior of these processes and utilize that information for improved control. To deal with multiple product/multiple tool control, sequential parameter estimation techniques are used to update the models and perform model-based control, with application to “high-mix” fabs with more than 20 products. Optimal sampling strategies are also being investigated.
With the recent increase in the demand for oil and the predicted decline in available supply, the ability to obtain oil efficiently and economically has become increasingly important. There is interest in automating decisions regarding secondary recovery techniques, particularly water injection schemes, to yield so-called “smart reservoirs.” Reservoir simulators have traditionally been too large and run times too long to allow for rigorous solution in an optimization algorithm. It has also proven very difficult to marry an optimizer with the large set of nonlinear differential equations required for reservoir simulation. We have used a recently developed inter-well connectivity model that concentrates on the relationship between injection and production wells. Then we can balance and optimize the effects of injections wells to maximize the net present value of oil production from a given reservoir.