NLPCA as a diagnostic tool for control valve stiction

A significant number of control loops in process plants perform poorly due to control valve stiction. Developing a method to detect valve stiction in the early phase is imperative to avoid major disruptions to the plant operations. Nonlinear principal component analysis (NLPCA), widely known for its capability in unravelling nonlinear correlations in process data, is extended in this paper to diagnose control valve stiction problems. The present work is based on distinguishing the difference between the shapes of the signals caused by stiction and other sources, and utilizes the operating data of controlled variable-controller output (pv–op). The structure of pv–op data used in this work is of sufficiently low dimension such that the NLPCA’s output allows the usage of simple mathematical tests in quantifying the nonlinear behavior of the loop. It is shown that if the underlying structure of pv–op data is linear, the NLPCA output generally approximates to a straight line with a regression coefficient (R²) greater than 0.8, otherwise there is a possibility of presence of nonlinearity or non-Gaussianity. The presence of stiction is then detected via a new and simple NLPCA curvature index, I_NC. Results from simulated and real industrial case studies show that NLPCA is a very promising tool for detecting valve stiction.     

Heat exchanger fouling model and preventive maintenance scheduling tool

The crude preheat train (CPT) in a petroleum refinery consists of a set of large heat exchangers which recovers the waste heat from product streams to preheat the crude oil. In these exchangers the overall heat transfer coefficient reduces significantly during operation due to fouling. The rate of fouling is highly dependent on the properties of the crude blends being processed as well as the operating temperature and flow conditions. The objective of this paper is to develop a predictive model using statistical methods which can a priori predict the rate of the fouling and the decrease in heat transfer efficiency in a heat exchanger. A neural network based fouling model has been developed using historical plant operating data. Root mean square error (RMSE) of the predictions in tube- and shell-side outlet temperatures of 1.83% and 0.93%, respectively, with a correlation coefficient, R², of 0.98 and correct directional change (CDC) values of more than 92% show that the model is adequately accurate. A case study illustrates the methodology by which the predictive model can be used to develop a preventive maintenance scheduling tool.     

Prediction of industrial debutanizer column compositions using data-driven ANFIS- and ANN-based approaches

Neural Computing and Applications - The work in this paper is based on an industrial debutanizer column in a petroleum refinery located in Malaysia, which produces LPG (liquefied petroleum gas) as...     

Integrated OBF-NN models with enhanced extrapolation capability for nonlinear systems

This paper proposes a nonlinear system identification using parallel linear-plus-neural network models that provide more accurate predictions on the process behavior even on extrapolated regions. For this purpose, a residuals-based identification algorithm using parallel integration of linear orthonormal basis filters (OBF) and neural networks model is developed and analyzed under range extrapolations. Results on the van de Vusse reactor case study show enhanced extrapolation capability when compared to the conventional neural network (NN) and the series Wiener-NN models.     

Neural network applications in fault diagnosis and detection: an overview of implementations in engineering-related systems

The use of artificial neural networks (ANN) in fault detection analysis is widespread. This paper aims to provide an overview on its application in the field of fault identification and diagnosis (FID), as well as the guiding elements behind their successful implementations in engineering-related applications. In most of the reviewed studies, the ANN architecture of choice for FID problem-solving is the multilayer perceptron (MLP). This is likely due to its simplicity, flexibility, and established usage. Its use managed to find footing in a variety of fields in engineering very early on, even before the technology was as polished as it is today. Recurrent neural networks, while having overall stronger potential for solving dynamic problems, are only suggested for use after a simpler implementation in MLP was attempted. Across various ANN applications in FID, it is observed that preprocessing of the inputs is extremely important in obtaining the proper features for use in training the network, particularly when signal analysis is involved. Normalization is practically a standard for ANN use, and likely many other decision-based learning methods due to its ease of use and high impact on speed of convergence. A simple demonstration of ANN’s ease of use in solving a unique FID problem was also shown.

     

A hybrid formulation and design of model predictive control for systems under actuator saturation and backlash

In this paper, we develop a hybrid design framework of model predictive controller (MPC) for multivariable systems that simultaneously and explicitly addresses the actuator saturation and backlash. The discrete characteristics of the actuator backlash allows us to mathematically express it as a set of mixed-integer linear inequalities constraint in the inputs. As a result, the constrained MPC design is formulated as solving a mixed-integer quadratic programming (MIQP) problem. Furthermore, the proposed MIQP-based design is applied only in the proximity of steady state operating points after locating the active backlash and providing the estimate of the backlash size. Simulation studies are presented to demonstrate how the hybrid MPC performs when applied to an industrial case study of fluid catalytic cracking unit. It is shown that in the presence of actuator saturation and backlash the closed-loop performance can be improved substantially when applying the hybrid method as compared to the traditional design approaches.     

Improved process monitoring using the CUSUM and EWMA-based multiscale PCA fault detection framework

Process monitoring techniques are of paramount importance in the chemical industry to improve both the product quality and plant safety. Small or incipient irregularities may lead to severe degradation in complex chemical processes, and the conventional process monitoring techniques cannot detect these irregularities. In this study to improve the performance of monitoring, an online multiscale fault detection approach is proposed by integrating multiscale principal component analysis (MSPCA) with cumulative sum (CUSUM) and exponentially weighted moving average (EWMA) control charts. The new Hotelling's T² and square prediction error (SPE) based fault detection indices are proposed to detect the incipient irregularities in the process data. The performance of the proposed fault detection methods was tested for simulated data obtained from the CSTR system and compared to that of conventional PCA and MSPCA based methods. The results demonstrate that the proposed EWMA based MSPCA fault detection method was successful in detecting the faults. Moreover, a comparative study shows that the SPE-EWMA monitoring index exhibits a better performance with lower values of missed detections ranging from 0% to 0.80% and false alarms ranging from 0% to 21.20%.