Detecting change in complex process systems with phase space methods
Year: 2006
Type: dissertation
Abstract: Model predictive control has become a standard for most control strategies in modern \nprocess plants. It relies heavily on process models, which might not always be \nfundamentally available, but can be obtained from time series analysis. The first step \nin any control strategy is to identify or detect changes in the system, if present. The \ndetection of such changes, known as dynamic changes, is the main objective of this \nstudy. In the literature a wide range of change detection methods has been developed \nand documented. Most of these methods assume some prior knowledge of the system, \nwhich is not the case in this study. Furthermore a large number of change detection \nmethods based on process history data assume a linear relationship between process \nvariables with some stochastic influence from the environment. These methods are \nwell developed, but fail when applied to nonlinear dynamic systems, which is focused \non in this study. \nA large number of the methods designed for nonlinear systems make use of statistics \ndefined in phase space, which led to the method proposed in this study. The \ncorrelation dimension is an invariant measure defined in phase space that is sensitive \nto dynamic change in the system. The proposed method uses the correlation \ndimension as test statistic with and moving window approach to detect dynamic \nchanges in nonlinear systems. \nThe proposed method together with two dynamic change detection methods with \ndifferent approaches was applied to simulated time series data. The first method \nconsidered was a change-point algorithm that is based on singular spectrum analysis. \nThe second method applied to the data was mutual cross prediction, which utilises the \nprediction error from a multilayer perceptron network. After the proposed method was \napplied to the data the three methods’ performance were evaluated. \nTime series data were obtained from simulating three systems with mathematical \nequations and observing one real process, the electrochemical noise produced by a \ncorroding system. The three simulated systems considered in this study are the \nBelousov-Zhabotinsky reaction, an autocatalytic process and a predatory-prey model. \nThe time series obtained from observing a single variable was considered as the only \ninformation available from the systems. Before the change detection methods were \napplied to the time series data the phase spaces of the systems were reconstructed with \ntime delay embedding. \nAll three the methods were able to do identify the change in dynamics of the time \nseries data. The change-point detection algorithm did however produce a haphazard behaviour of its detection statistic, which led to multiple false alarms being \nencountered. This behaviour was probably due to the distribution of the time series \ndata not being normal. The haphazard behaviour reduces the ability of the method to \ndetect changes, which is aggravated by the presence of chaos and instrumental or \nmeasurement noise. Mutual cross prediction is a very successful method of detecting \ndynamic changes and is quite robust against measurement noise. It did however \nrequire the training of a multilayer perceptron network and additional calculations that \nwere time consuming. The proposed algorithm using the correlation dimension as test \nstatistic with a moving window approach is very useful in detecting dynamic changes. \nIt produced the best results on the systems considered in this study with quick and \nreliable detection of dynamic changes, even in then presence of some instrumental \nnoise. \nThe proposed method with the correlation dimension as test statistic was the only \nmethod applied to the real time series data. Here the method was successful in \ndistinguishing between two different corrosion phenomena. The proposed method \nwith the correlation dimension as test statistic appears to be a promising approach to \nthe detection of dynamic change in nonlinear systems.
Author: Paul Jacobus Botha
Institution: Stellenbosch University
Language: English
Subfield: Control and Systems Engineering
Field: Engineering
Domain: Physical Sciences
Keywords: phase-space methods, correlation dimension, nonlinear dynamical systems, singular spectrum analysis, change point detection, electrochemical noise, corrosion, Belousov-Zhabotinsky reaction, predator-prey model, time-delay embedding, model predictive control, process monitoring, false alarm rate, multilayer perceptron, time series
Open Access status: green
OpenAlex ID: W817986181
