SHORT COMMUNICATION Feed-forward-based software sensor for outlet turbidity of coagulation process considering plug flow condition W. Liu 1 • H. Ratnaweera 1 Received: 12 August 2016 / Revised: 1 January 2017 / Accepted: 15 February 2017 Ó Islamic Azad University (IAU) 2017 Abstract Physical online sensors are widely used in wastewater treatment plants. The high costs of acquisition and maintenance as well as the delayed response due to long hydraulic retention times, applications of some physical online sensors are limited. Consequently, studies on developing software sensors have been drawn much attentions these years. Aiming to predict treatment results after water is coagulated, and this paper focuses on developing a software sensor of outlet turbidity for coag- ulation process. Solution to address the potential non-plug flow conditions in sedimentation tanks is discussed. Model validation results show that the proposed software sensor has 86% of accuracy within the main working range. This paper expresses a novel concept that what the proposed software sensor predicted has been not formed in the coagulation process. Since the software sensor enables to know the treatment results without waiting for long hydraulic retention time of sedimentation tank, it not only shortens the response time of manual dosing control but also serves as a feedback parameter to define optimal dosage for coagulant dosing control systems. Keywords Coagulant dosing Á Model Á Non-plug flow Á Sedimentation tank Introduction Online sensors are becoming more and more common in modern wastewater treatment plants (WWTP), and mea- surement signals from these online sensors provide a basis for process monitoring, modelling and control (Luttmann et al. 2012). Although cleaning and calibration of online sensors are part of the routine works at WWTP, they are still unable to provide accurate measurements continu- ously throughout the day because of fouling, drifting, ageing problems, operational errors etc. Moreover, time lag, resulting from long hydraulic retention time (HRT) of treatment processes, hinders some online sensors at outlet to feed timely information back to monitoring and con- trol. In addition to the above two disadvantages, high costs and harsh working conditions are sometimes also challenges for the application of online sensors in WWTP. A software sensor is a proposed solution to substitute the traditional physical online sensor. It has many advantages compared with physical online sensors, as it does not require any additional investments or maintenance, but utilizes existing online sensors measuring other parameters. Based on the process knowledge and other online sensors, software sensors as conceptual devices are already reported to use for estimating some of the key measurements of a treatment process (Kadlec et al. 2009). Increasing number of researches and applications of this field have been seen in wastewater treatment (WWT). Significant efforts are made towards developing procedures (data acquisition, data pre-processing, model design and model maintenance) and selection of modelling methods such as artificial neural network (ANN), partial least square (PLS) as well as principle component analysis (PCA) (Kadlec et al. 2011; Francisco et al. 2016). Editorial responsibility: T.Y. Wu. & W. Liu weiliu_2015@126.com 1 Norwegian University of Life Sciences, PO Box 5003-IMT, 1432 Aas, Norway 123 Int. J. Environ. Sci. Technol. DOI 10.1007/s13762-017-1284-4