Process fault detection and diagnosis using Boolean representation on fatty acid fractionation column

Abstract

Nowadays, detecting and diagnosing process fault is an important issue because it can improve system availability and protect chemical plant from accidents. There are many method introduced to conduct process fault detection and diagnosis (PFD&D), but this paper will focus on the use of artificial neural network (ANN) in detecting and diagnosing faults. ANN has the capability of recognizing multivariable pattern very well. This advantage is useful in systematically detect failures in process plant. Therefore, an algorithm for the development of process fault detection system in dynamic processes using artificial neural network (ANN) is presented. The algorithm utilizes process simulator to develop plant model in order to conduct sensitivity analysis and provide dynamic data on selected fault. Various process conditions are specified and simulated using commercial process simulator. Sensitivity analysis is conducted to identify whether or not the specified process condition effect the operation of the plant. If it does, each of the faults identified is represented by a specific Boolean representation. In other words, each fault has its own pattern indicated by a Boolean representation. Input for the ANN model will be the faulty data for all of the identified fault and the output will be the specified Boolean representation for each fault. The topology of the ANN model was founded on multilayer feed forward network architecture and the training scheme conducted using back propagation algorithm. The effectiveness of the proposed fault detection system on a simulated fatty acid fractionation column is presented. Through the proposed algorithm, various faults could be simulated and detected using the system. Results show that the system was successful in recognizing and detecting selected fault introduced within the process plant model.