Year 2003
Tiina Komulainen

Online-Monitoring of a Dearomatization Process

Early detection of process disturbances and prediction of malfunctions of process equipment increase the uniform quality of products, improve the safety of the process and minimize the time and resources needed for maintenance. The objective of online-monitoring is to trace the state of the process and condition of process equipment in real time and detect faults as early as possible. The aim of this thesis was to study different online-monitoring methods applied to industrial processes and to apply one of these methods to a dearomatization process.

In the literature survey part of the study, different methods used for online-monitoring of industrial processes and the results are discussed. The survey also addresses the mathematical theories of the methods. Most of the online-methods were history-based statistical multivariate methods or methods based on neural networks.

In the experimental part of this study the monitoring needs of the dearomatization process were investigated. The flash point and distillation curve analyzers were selected as targets for monitoring. The status of analyzers and prediction of the future measurements were estimated by using the method of partial least squares, PLS.

First, all process variables were time-scaled. To capture the characteristic of the dearomatization process with a linear model, some simple computational variables were created.Next the combination of process and computational variables was selected for offline-testing of the monitoring system. To assure the validity of the model for all normal process states, the data for model development contained data blocks representing all types of solvents.

The data set for the offline test contained process data for about 455 hours with minute resolution. Results of the offline-test were encouraging, 96 – 99 % of normal states of analyzers and 67 – 97 % of the fault states were classified correctly A monitoring system was developed and tested online for a time period of 144 hours. During this time the type of solvent changed twice and also a disturbance hit the process. The monitoring system classified the type changes correctly as normal states and gave an alarm for the abnormal process state during the disturbance.

This info last modified 27 Apr 2024 by Jerri Kämpe-Hellenius