OPTIMAL ESTIMATION ALGORITHMS FOR SYSTEMS WITH MULTIPLE TIME MODES AND DELAYED MEASUREMENTS

When controlling a process, the state of the system must be known before control action can be taken. Using variational calculus, state estimators were derived for systems whose parameters are distributed in space as well as time. These distributed-parameter state estimation algorithms were studied on systems which displayed multiple dynamic time modes in the states and time delays in the measurements. These traditionally overlooked phenomena can cause estimators to become computationally intensive and slow to respond to process dynamics.; Singular perturbation theory was used to simplify the set of complex stiff estimation equations which arises when multiple time modes are present in a dynamic system. The singular perturbation technique allows one large Kalman-type estimator to be broken down into a set of smaller estimators. These smaller estimators were able to predict the process states well in simulations. Only when very poor initial conditions were used was the difference between the complete estimator and the singular perturbation estimators noticed.; Two new estimation algorithms were developed for systems with measurement delays; one for systems with pure time-delays and the other for systems with dynamic delays. The first of these algorithms was able to reduce the measurement errors incurred by pure time-delayed measurements. When dynamic measurements were present, a transformation of the system helped to minimize the estimated error. These techniques were studied via simulations on a continuous stirred slurry tank reactor, a heated bar and a gas phase tubular reactor.; Finally, all the new techniques were combined and tested on a real-time styrene pilot plant system. This system exhibits multiple dynamic time modes in the reactor states (temperature, concentration and activity), along with dynamic and pure time delays in the measuring devices used. The process was monitored, controlled, and estimated using a distributed-computer network that included three separate computers. The combined estimators were able to estimate the measurable reactor states (temperature and concentration), along with the unmeasurable reactor catalyst activity. Using only a small number of measured temperatures (2) and the outlet reactor concentrations, the estimators correctly predicted these states to within 4% of their actual values. The catalyst activity was studied for various steam-to-oil ratios (SOR) and inlet reactor temperatures.