Dynamic processing, batch design and scheduling with uncertainty

This thesis contributes towards an integrated approach to processing and planning for the batch chemical industry.; Transient processing policies for batches at certain dynamic stages can enhance production efficiency. These decisions are resolved simultaneously with broader planning decisions to demonstrate savings over operating with recipes or non-transient policies as well as sequential solution strategies. Dynamic process models are discretized using orthogonal collocation on finite elements, planning and scheduling is limited to a special class of operations allowing a convenient representation. Significant benefits are demonstrated for purely deterministic scenarios.; Under process uncertainty, recovery of this potential is equally critical. The proposed approach addresses uncertain scenarios through a multiperiod formulation with design aspects coupling the periods. A state feedback approach is employed and closed loop models are discretized with feedback parameters as design variables. This provides a practical implementation advantage with guaranteed performance very close to perfect information. It however requires solving multiperiod design problems, for which the computational resource requirement grows disproportionately with an increase in the number of periods.; Multiperiod problems have a block bordered diagonal structure that can be exploited to develop a linear decomposition algorithm in the number of periods. The MPD/SQP algorithm by Varvarezos and Biegler (1994), employed an active set strategy for the search direction quadratic programming (QP) problem at each iteration by projecting each period into the design space and using the design solution to coordinate their solution. In addition to desirable features of MPD/SQP such as range and null space decomposition in each period and a periodic decomposition, the MPD/rISQP algorithm proposed here employs an interior point strategy for addressing bounds and inequalities. The advantage is solving the QP in a fixed number of steps each time and dealing with a fixed structure at each iteration independent of the active set. Four examples are solved with up to 500 periods including examples from dynamic processing and planning under uncertainty.; Finally, for future direction, steps are outlined to extend the algorithm to handle multiperiod problems with pass on variables where the periods appear successively in time and are linked to previous period in the sequence.