Parzen window distribution as new membership function for ANFIS algorithm-Application to a distillation column faults prediction

Abstract

In the domain of chemical engineering, the distillation column is one of the most important reactors in the operations unit. These chemical reactors can represent high maintenance costs and can disrupt production for long periods of time in addition to having risks of disastrous impacts. Unfortunately, preventive maintenance is both expensive and insufficient. Thus, the optimal solution is predictive maintenance; modeling a pre-crash control system that enables a greater comprehension of the future path of reactors. This research paper will present the Adaptive Neuro Fuzzy Inference System (ANFIS) as a superior technique for forecasting the future path of the distillation column system, then will propose Parzen windows distribution as a new membership function to improve ANFIS performance. Improvements can result from reducing consumption time and making the process more reflective of real-time application or by minimizing the Root Means Square Error (RMSE) between real and predictive data. The methodology was tested on real data obtained from a distillation column with the aim of forecasting potential failures in the automated continuous distillation process. A comparative study was necessary for the selection of the best membership function to be used for the ANFIS algorithm when applied to the distillation column data. Results reflected the effectiveness of the proposed technique and the Parzen windows was the smallest RMSE for several signals.