Multiple Fault Diagnosis in Distillation Column Using Multikernel Support Vector Machine

Abstract

Fault detection and diagnosis (FDD) in process industries is an important task for efficient process monitoring and plant safety. It is also essential for improving product quality and reducing production cost by reducing process downtime. Real-time multiscale classification of faults plays a vital role in process monitoring. However, some major issues such as high correlation, complexity, and nonlinearity of data are yet to be addressed. In this paper, a fault diagnosis approach based on multikernel support vector machines is proposed to classify the internal and external faults such as reflux failure, change in reboiler duty, column tray upsets, and change in feed composition, flow, and temperature in a distillation column. The data set is generated using Aspen plus dynamics simulation at normal and faulty states. The classification has been done by various methods such as decision tree, K-nearest neighbors, linear discriminant analysis, artificial neural network, subspace discriminant, and multikernel support vector machine. It is observed that the SVM based diagnostic system provides more accurate root cause isolation. The proposed MK-SVM method was evaluated by using the confusion matrix as the performance evaluator. The result showed that the proposed model has a high FDR which is 99.77% and a very low FAR, i.e., 0.23%.

Narx Network Based Data-Driven Algorithm for Detection of Tray Faults in Nonlinear Dynamic Distillation Column

“…flow, top and bottom compositions, column pressure, and top and bottom level controllers. The detailed description of the development of steady-state and dynamic simulation can be obtained from our previous publications [4,5,7,25,38]. The Aspen Plus dynamic simulation is shown in Figure 1.…”

“…Distillation column is a multivariable system and holds a substantial part of the refinery and other chemical process industries [5,6]. The change in reboiler steam pressure, tray efficiency, feed composition, or overall pressure and temperature of distillation column may significantly affect the overall performance of the column [7,8]. It is reported that significant financial losses are incurred when a highthroughput plant is shut down or partially operated.…”

“…Fault detection and diagnosis is an essential area for accident prevention in process industries to achieve most of the desired and challenging goals [10]. For a few decades, process industries have used fault detection and diagnosis extensively for efficient process monitoring where timely detection of the faults is vital for safety and profitability [7]. Fault detection is to evaluate whether the process is operating in the acceptable range, while fault diagnosis identifies the cause and characteristic of the fault [11].…”

“…Data-driven approaches consider fault detection and diagnosis as classification tasks [21]. This classification can be done by supervised or unsupervised learning methods, including Partial Least Square (PLS) [22], Principle Component Analysis (PCA) [23,24], Artificial Neural Network (ANN) [25][26][27], Fisher Discriminant Analysis (FDA) [28], Bayesian Networks [29] and Support Vector Machine (SVM) [7].…”

A Fault Identification Method in Distillation Process Based on Dynamic Mechanism Analysis and Signed Directed Graph

“…A robust fault detection method for the distillation column was effectively verified by setting and analyzing the fault types in dynamic simulation [11]. Dynamic simulation provided a data set, at normal and faulty states, for the fault classification of multikernel support vector machines [12]. A quantitative HAZOP method with dynamic simulation as a deviation reasoning tool was developed to reduce uncertainty in manual HAZOP analysis [13].…”