Multivariate on-line monitoring: challenges and solutions for modern wastewater treatment operation C. Rosen*, J. Röttorp** and U. Jeppsson* * Industrial Electrical Engineering and Automation, Lund University, Box 118, SE-221 00 Lund, Sweden (E-mail: christian.rosen@iea.lth.se; ulf.jeppsson@iea.lth.se) ** IVL, Swedish Environmental Research Institute, SE-100 31 Stockholm, Sweden (E-mail: jonas.rottorp@ivl.se) Abstract In this paper, a number of challenges, which need to be overcome if multivariate monitoring of wastewater treatment operation is to be successful, are presented. For each challenge, one or several solutions are discussed. The methodologies are illustrated using an example from full-scale wastewater treatment operation. Some guidelines regarding choices of methods and implementation aspects are given. Keywords Adaptive monitoring; detection; principal component analysis (PCA); wastewater treatment operation Introduction On-line monitoring of industrial processes is carried out to ensure that process outputs meet requirements on product quality, process safety and efficient use of resources. Modern wastewater treatment (WWT) plants collect large numbers of on-line measurements as more and more process variables can be measured. The on-line measurements have become an important source of information in the effort to achieve efficient operation and manage- ment of WWT plants. The large number of measured variables together with the nature of WWT processes put high demands on the techniques to extract on-line information from the data. In most process industries, monitoring of the process and its outputs is an important part of the operation. Monitoring can be said to consist of three phases: 1. Detection – recognising that there is a deviating event or that the process is not operating at its normal operational point; 2. Isolation – finding the deviating measurement variables that have triggered the detection; 3. Interpretation – finding the physical causes of the deviation and assessing its impact on the process. The first phase is a task well suited for computers, as it is monotonous and quantitative. The second phase is normally integrated with the first task and is, thus, also suitable for computers. However, the third phase requires process knowledge and is mainly a task for the operator relying on his experience (and possibly intuition), although attempts have been made using knowledge based systems, such as expert systems, for the interpretation. In order to facilitate the third phase, the methods used for the first two tasks must extract and organise the information appropriately and present the information in an easily inter- pretable way. During the last two decades, techniques based on multivariate statistics have become increasingly popular within many industrial fields. Principal component analysis (PCA) and developments based on PCA, such as principal component regression (PCR) and pro- jection to latent structures (PLS), have been applied successfully to various industrial Water Science and Technology Vol 47 No 2 pp 171–179 © IWA Publishing 2003 171 Downloaded from https://iwaponline.com/wst/article-pdf/47/2/171/423906/171.pdf by guest on 08 November 2018