Software sensors based on the grey-box modelling approach

Abstract

In recent years the grey-box modelling approach has been applied to wastewater transportation and treatment. Grey-box models are characterized by the combination of deterministic and stochastic terms to form a model where all the parameters are statistically identifiable from the on-line measurements. With respect to the development of software sensors, the grey-box models possess two important features. Firstly, the on-line measurements can be filtered according to the grey-box model in order to remove noise deriving from the measuring equipment and controlling devices. Secondly, the grey-box models may contain terms which can be estimated on-line by use of the models and measurements. In this paper, it is demonstrated that many storage basins in sewer systems can be used as an on-line flow measurement, provided that the basin is monitored on-line with a level transmitter and that a grey-box model for the specific dynamics is identified. Similarly, an on-line software sensor for detecting the occurrence of backwater phenomena can be developed by comparing the dynamics of a flow measurement with a nearby level measurement. For treatment plants it is found that grey-box models applied to on-line ammonia measurements from the aeration tank of an alternating plant provide information on the incoming ammonia load. It is also shown how measurements of the return sludge concentration from a secondary clarifier can be filtered to minimize the effect of the scraper. Thus, important information can be derived from on-line measurements if the appropriate grey-box model for the specific system is identified.

Evaluation of software sensors for on-line estimation of culture conditions in an Escherichia coli cultivation expressing a recombinant protein

“…Software sensors have been described for diverse applications, such as chemical reactors (Régnier et al, 1996), analytical devices (Matsumura et al, 1998), environmental processes (Carstensen et al, 1996), and bioprocesses (Kiviharju et al, 2008). In addition, software sensors support the ambition of process analytical technology (PAT) to establish on-line monitoring methods for high quality manufacture of pharmaceutical products (FDA, 2004;Gnoth et al, 2007).…”

“…-In contrast to these mechanistic modeling approaches are software sensors based on "black-box" models where inputs are correlated to outputs via a non-deterministic relationship (e.g. Carstensen et al, 1996). Methods of correlation include multivariate analysis (MVA), regression modeling, artificial neural network (ANN), and chemometric approaches (Clementschitsch et al, 2005;Ödman et al, 2009).…”

A software-based sensor for combined sewer overflows

“…Their application in wastewater technology is more common for wastewater treatment plants (WWTP) (e.g. Zyngier et al 2001;Dürrenmatt & Gujer 2012), but some examples from sewer systems have also been published, mainly in the 1990s (Bechmann et al 1999;Carstensen & Harremoes 1999;Carstensen et al 1996Carstensen et al ,1998. Measurement data are used as input to a mathematical model of the entire system or a subsystem.…”

“…Compared with complex deterministic models (mechanistic or conceptual models), grey-box models are characterized by a smaller number of parameters and therefore better statistical identiflability. Usually a simple conceptual model is used in combination with stochastic terms to describe the correlation of the residuals (Carstensen et al 1996). Black-box models (statistical relationships, artiñcial neural networks etc.)…”

Aplicación de métodos estadísticos multivariantes para la modelación y la monitorización de un reactor discontinuo secuencial para el tratamiento de aguas residuales

“…Önnerth et al (1996) emplean en una EDAR un sistema general de control (STAR, Superior Tuning and Reporting) que incorpora para el control del caudal de recirculación un soft sensor que permite estimar la capacidad de desnitrificación en el reactor anóxico. Carstensen et al (1996) ilustran la aplicación de dos soft sensors en una EDAR. Uno consiste en un modelo híbrido cuyo propósito es determinar la carga de amonio que entra en el reactor aerobio a partir de las medidas de amonio en dicho reactor, y el otro se basa en utilizar un filtro Kalman para corregir el ruido que presenta la señal de la medida registrada en tiempo real de la concentración de sólidos suspendidos que se recirculan desde el decantador secundario.…”