Granularity determination of activated sludge through on-line profiles by means of case-based reasoning Xavier Berjaga, Marta Coma, Joaquim Meléndez, Sebastià Puig, Jesús Colprim and Joan Colomer ABSTRACT Aerobic granulation from floccular sludge is difficult to detect in first stages with the naked eye. This work proposes a combination of multi-way principal components and case-based reasoning to predict the granulation state of a sequencing batch reactor, based solely on the on-line registered profiles of common sensors (i.e. pH, dissolved oxygen and oxidation-reduction potential). The methodology is able to discriminate between two active sludge granularities ( floccular and granular). Two different scenarios are presented: one in which both granularities are present, and another scenario for which the granular state is not initially available. Analysis reported pH as the key variable in the transition between both states according to its variation, and that, in general, the granularity of the process can be correctly predicted at the end of the anaerobic phase. This methodology improves process monitoring capabilities during granulation and is an on-line alternative to a microscope analysis before the batch release. Xavier Berjaga (corresponding author) Joaquim Meléndez Joan Colomer Laboratory of Control Engineering and Intelligent Systems (eXiT), University of Girona, Girona, Spain E-mail: xavier.berjaga@udg.edu Marta Coma Sebastià Puig Jesús Colprim LEQUIA, Institute of the Environmental, University of Girona, Girona, Spain Marta Coma Laboratory of Microbial Ecology and Technology (LabMET), Ghent University, Gent, Belgium Key words | aerobic granulation, case-based reasoning, on-line sensors, principal component analysis (PCA) INTRODUCTION Aerobic granulation sludge is an alternative technology for wastewater treatment that can handle high loading rates or large volumes of waste in small facilities (Liu & Tay ) thanks to high biomass retention and fast settling properties. The major challenge of this technology is the start-up of the process, as granulation conditions are not yet fully under- stood. The major selection pressures for granulation have been identified as short settling times and high volume exchange ratios (VERs) with the purpose of selecting biopar- ticles according to their settling velocity (Liu et al. ). However, a high washout of biomass is usually obtained during the process (McSwain et al. ). Furthermore, gran- ulation treating low-strength wastewater with lower organic loading rates (OLRs) than 2 kg COD m À3 d À1 takes longer to reach the steady state than at high OLRs (Tay et al. ) and the transition state, where a system becomes fully granu- lated, is sometimes difficult to detect. Some off-line analyses such as particle distribution, may help in the turning point detection, but they are expensive and time consuming. Activated sludge systems are usually equipped with on- line sensors such as pH, dissolved oxygen (DO) and oxi- dation-reduction potential (ORP) sensors, which have been proven to give ample information regarding biological nutri- ent removal processes (Puig et al. ; Yuan et al. ). Despite their utility for controlling and enhancing these reactions, on-line parameters have neither been analysed nor related to granulation. Contrary to nutrient removal reactions, granulation is a physical transformation that does not directly affect the bio-chemistry of the process. However, diffusion of nutrients, ions and oxygen may be affected and reflected in the on-line profiles. However, the interpolation of the on- and off-line data must be assessed to determine the granulation state. The amount of data released by sensors is difficult to handle on-line for continu- ous operation. However, the combination of dimension reduction techniques as principal component analysis (PCA) and rule/instance-based models such as case-based reasoning (CBR) adds learning capabilities to monitoring systems to enhance automatic detection and classification of their operating conditions. PCA and its extensions are commonly used in the analy- sis of wastewater records. Aguado et al.() used unfold- PCA for steady-state identification in a laboratory scale sequencing batch reactor (SBR) operated for biological 760 © IWA Publishing 2014 Water Science & Technology | 69.4 | 2014 doi: 10.2166/wst.2013.776