A hybrid method for online monitoring of internals performance in distillation columns

Abstract

Distillation columns are widely used for separation in industry. To ensure separation stability, it is essential to online monitor the internals performance of a distillation column. The separation efficiency can be evaluated by estimation of Murphree Efficiency of the column. However, as the Murphree Efficiency is affected by both the internals and the tower operating states, it cannot be directly used to represent the internals performance until the influence of state variation influence is excluded. To address this problem, a hybrid method with both the mechanism-based and data-driven models is proposed in this work. Initially, steady-state segment is extracted through a wavelet transform. Then, a mechanism-based model is used to derive the Real-time Murphree Efficiency through parameter estimation and data reconciliation for the extracted steady-state segment. Next, an online and offline two-stage strategy is presented for internals performance detection. In the offline stage, a data-driven Bayesian regression model is developed to correlate the tower states and Murphree Efficiency by assuming stable performance of the internals. While in the online stage, an internal performance index is computed by comparing the Expected Murphree Efficiency, predicted by the Bayesian regression model, and the Real-time Murphree Efficiency developed by the mechanism-based model. Lastly, the proposed method is applied to a phenylenediamine distillation system with three columns, for which, degradation of the packing is effectively monitored.