|Ph.D Thesis||Department of Mechanical Engineering|
|Supervisor:||Prof. Palmor Zalman|
|Full Thesis text - in Hebrew|
Time delays (dead times) appear in many processes in industry, economics, biology, communication networks etc. The presence of dead times complicates the control design and exerts severe limitations on the performance of such systems. Hence control structures that can potentially improve the performance of those processes are of central importance.
Over twenty years ago Morari and co-workers arrived at a surprising result which holds to date and states that in some cases the dynamics performance of Multi Input Multi Output (MIMO) plants with multiple delays can be improved by artificially increasing dead times.
In this research, the loop shifting technique has been extended from the single delay case to MIMO stable/unstable plants with multiple input/output (I/O) delays. The latter technique led to three fundamental results. First, it enabled to convert the original control problem with I/O delays to an equivalent delay free one such that the internal stability is preserved, thus simplifying considerably the control design of such systems and particularly the optimal H2 design. Second, it revealed new special structures of dead time compensators (DTC) for MIMO plants with multiple I/O delays consisting of a modified Smith predictor and novel Inter-Channel Feed-Forward (ICFF) compensators operating on the controller input and output channels. Third, it was shown, via this technique, that FASP structures also are optimal in terms of robustness for unstructured uncertainties in the rational part or in the multiple input (output) delays of the process. The overall controller is termed Feed-Forward Action Smith Predictor (FASP). Through application of the novel optimal H2 FASP to simple cases it was shown that artificial increases of delays do not bring about improvement in performance. It was further demonstrated that the improved performance gained by the optimal FASPs is due to the novel ICFF components and that unlike the ad hoc procedures for constructing DTCs and the optimal methods for single delay cases, the structure of the optimal FASP does depend on the exogenous inputs and the control goals. Simple examples demonstrate the results.