Performance analysis on implementation of certain multivariate statistical techniques and bayesian estimation method for fault diagnosis in shell and tube heat exchanger

Abstract

newline Early detection and diagnosis of abnormal events in industrial processes can represent economic, social and environmental profits. When the process have a great quantity of sensors or actuators, the Fault Detection and Isolation (FDI) task is very difficult. Advanced FDI methods can be classified into two major groups First is which do not assume any form of model information (process history-based methods) and second is which use accurate dynamic process models (model-based methods). For any fault diagnosis algorithm, there are some performance indices such as detection time, false alarm rate, missed alarm rate and multiple fault identifiability. In this work, Fault Detection and Diagnosis algorithm was tested in a pilot plant co-current shell and tube heat exchanger. In case of process history-based methods, a fault diagnosis system based on Dynamic Principal Component Analysis (DPCA), Fisher Discriminant Analysis (FDA) and Correspondence Analysis was tested for sensor fault, actuator fault and multiple faults. CA showed quicker detection for sensor and actuator fault with lower missed alarm rate. But DPCA showed significant detection delay and missed alarm rate for multiple fault identification. In case of model-based fault detection and diagnosis, an efficient ARX (Auto Regressive model with exogenous input) model and ANFIS (Adaptive Neuro-Fuzzy Inference System) model is constructed from the input-output measurements. These two models are used to analyse the symptoms for process fault (tube leak fault). ANFIS model shows the best fit with quicker detection and lower missed alarm rate when compared to ARX model