| 초록 |
Multivariate statistical projection methods were applied to an industrial chemical process, in order to reduce the variations in the product quality in a terephthalic acid manufacturing. Process and quality measurement data were modeled and analyzed using partial least squares methods to find out the sources of quality variability and to simulate the effects of these sources on the product quality. The results show that the major sources that cause the final product quality to vary highly are the variability in catalyst concentration and the changes in process throughput, and reveal that the standard deviation of the product quality after eliminating both two sources can be reduced by about 37 percent compared to the present value. To eliminate the major sources found from the analyses, it was proposed that the catalyst concentration in the acid solvent flowing into the oxidation reactors should be always maintained at a certain level over a long period of operating time by employing an online measurement and control system. In addition, a new operating condition was obtained from partial least squares modeling to cope with the frequent changes in process throughput so that the quality variations can be minimized. |
