Simultaneous data reconciliation and gross error detection using a robust function method for nylon 6 reactor

Abstract

On-line optimization is a powerful method for economic improvement and resource reduction in chemical industries, because it allows a process to be operated near its optimum operating condition.Thus, it is currently receiving increasing attention. In order to perform online optimization, sequence solving of three optimization problems which are data reconciliation, parameter estimation, and economic optimization are required. Normally, process measurements are used to determine the actual state of the process. However, these measurements usually contain random as well as gross errors. Both random and gross errors should be eliminated in data reconciliation step. Further, the reconciled data must satisfy process material and energy balances. In this work, we apply data reconciliation in VK column reactors for industrial nylon 6 production process at steady state condition. Three data reconciliation algorithms: Contaminated Normal, Lorentzian distribution function and Hampel’s redescending M-estimator, are investigated to compare the performances. We also compare these algorithms under the conditions where some process streams are unmeasured. The result shows that all algorithms perform well in this process. Because it can support with process measurements contain both gross and random errors. Moreover, these algorithms also work well in the case where unmeasurement process streams exist.