Principal Investigator George Stephanopoulos
Models that describe process behavior at several spatial and temporal scales are essential for a number of engineering tasks ranging from process analysis to process design to operational monitoring, control and optimization. There are three fundamental reasons for this need of “multiscale” models:
(a) Physical, chemical, and biological phenomena occur at different spatial and temporal scales.
(b) The operational tasks of closed-loop feedback control, adaptive control, fault diagnosis, and scheduling and planning of operating procedures are deployed with process models of progressively increasing dominant time scales.
(c) Sensors may provide measurements of process behavior at different sampling rates, inducing control actions at correspondingly different rates. The optimal fusion of measurement information at various time-scales with control actions requires the availability of process models at time scales, which are commensurable with the sampling rates of various measurements and application rates of control actions.
Multiscale models of dynamic systems offer a representation that links states at different time scales over distinct time periods. As such they allow a natural bridging of scales (ranges of frequencies) and time, and are more attractive than models defined in the time- or frequency-domain, alone. Earlier research in our group as introduced a wavelet-based modeling framework for the representation of process dynamics over a broad range of temporal scales. This is in essence a “multiresolution” modeling framework emanating from a single underlying model, and has been used in the analysis and design of multi-scale model-predictive control systems.
Current research work attempts to expand previous developments in order to address a number of outstanding problems related to process operations, such as: nonlinear model-predictive control, analysis and diagnosis of operational trends, planning and scheduling of process operations.