Optimal design and control of silane off-gas recovery process with dividing wall column under periodic disturbances from parallel batch reactors

Abstract

This thesis has presented a rigorous design and a plant-wide control structure scheme for the effective separation of off-gas from a polysilicon plant with periodic disturbances in the feed. Solar power generation is eco-friendly energy resource, polysilicon is used as a raw material to manufacturing solar cells. In order to obtain the high efficiency of solar cells, it requires ultra-high purity of polysilicon which essentially involves the advanced recovery processes. The target process studied in this thesis produces 10,000 MT polysilicon per year with 28 chemical vapor deposition reactors. The reaction kinetic models were constructed, and their parameters were identified using a genetic algorithm, and that were utilized to design the separation process which separates into hydrogen, hydrogen chloride, and silanes. Though proportional-integral controllers were installed to remove the periodic disturbances, they were not enough to meet the active constraints. A centralized supervisory layer based on model predictive control to minimize the fluctuation of the recovered concentration, was proposed. Proportional-integral controllers regulates the inventory levels, thereby operating the process stable, and a model based controller showed a good performance rejecting the periodic disturbances. Using a rigorous plant-wide dynamic model, four cases were studied with different control schemes with and without buffer tanks. Separated silane off-gas is consist of dichrloro silane, trichloro silane, and silane tetrachloride. Among them, dichloro silane and trichloro silane go into the reactors again. Dividing wall column was designed to achieve the lower energy consumption for the ternary mixture, and control structures were studied. From the novel approach, the energy consumption was decreased up to 37 % compared with classical sequential columns, and the settling time also reduced by proposed research results.