Estimation of a mass transfer coefficient for nylon manufacture using multiple neural networks

✍ Scribed by Sarah S.Y. Lam; Alice E. Smith; Badie I. Morsi

• Publisher: Society of Manufacturing Engineers
• Year: 2001
• Tongue: English
• Weight: 939 KB
• Volume: 20
• Category: Article
• ISSN: 0278-6125
• DOI: 10.1016/s0278-6125(01)80053-4

✦ Synopsis

The manufacture of nylon involves the use of cyclohexane, whose successful production is strongly dependent on the kinetics and gas-liquid mass transfer of its oxidation. To properly design the cyclohexane oxidation process, both equipment and process variables must be considered; however, previous studies have been quite limited and specific only to a certain equipment setup. In this paper, a more generalized predictive model of cyclohexane oxidation is proposed that can be used to design the best production method by accurately estimating the mass transfer coefficient. Using experimental data, two neural network approaches are devised and tested. The first is a single network approach whereas the second contains multiple neural networks in a hierarchy, termed a "mixture of experts" approach. These models are assessed using cross validation, which leverages the available data for both the construction and validation of the predictive models. The capabilities of the neural network approach are compared with that of statistical nonlinear regression on a published data set.