METHODS AND TOOLS FOR MODEL-BASED VIRTUAL SENSORS APPLIED TO CONDITION MONITORING Mikel GONZALEZ 1,2 , Ekaitz ESTEBAN 1 , Oscar SALGADO 1 , Jan CROES 2 , Bert PLUYMERS 2 , Wim DESMET 2 1 Control and Monitoring Area, IK4-Ikerlan, J. M. Arizmendiarrieta 2, 20500 Mondragon, Spain 2 KU Leuven, Department of Mechanical Engineering – member of Flanders Make, Celestijnenlaan 300 B, B-3001, Heverlee, Belgium Key words: Modeling, Virtual sensing, FMI, state estimation Abstract Model-based virtual sensing techniques are a valuable approach to estimate system variables which are difficult to measure. Instead of measuring these variables directly, physics-based models and estimation algorithms are used to compute them. As the system grows in complexity the use of dedicated modeling tools is required to reduce modeling effort and errors. However the integration of these tools with the estimation algorithms is not always straightforward. In this paper an overview of the main virtual sensor algorithms and the way to connect them with modeling tools is presented. The Functional Mock-up Interface (FMI) is discussed as the most suitable way to accomplish this. The advantages of using a symbolic modeling language such as Modelica in the implementation of virtual sensors are also discussed. These advantages are highlighted by means of an application example. 1 INTRODUCTION Maintenance services account for a significant part of the operating cost of high-value systems, especially when possible failures result in large down times [1]. Condition based maintenance strategies provide a cost efficient way to cope with the safety and reliability requirements of these systems [2]. Condition monitoring strategies rely on the measurement of specific variables of the system. In many applications, however, the direct measurement of these variables is either too costly or not possible. Virtual sensing is an attractive option to overcome these difficulties. Virtual sensing is a technique to estimate variables and parameters of interest using available measurements and physics-based models instead of direct physical sensors. A review of fault and damage detection methods can be found in [3], [4] and [5]. In contrast to other fault and damage detection methods, model-based techniques incorporate physical knowledge of the system, allowing a deeper understanding of the process behavior [6]. Therefore these techniques provide not only crucial information on unobservable quantities but also physical insight in why the system’s performance is degrading. The efficacy of these techniques highly depends on the capability of the model to accurately represent the physics of the system and identify uncertain system parameters [1]. This can be a major drawback of these techniques, as the high complexity of modern electromechanical systems results in models which are either too costly or not accurate enough. In addition, cyber-physical systems are composed of several domains such as mechanical, electrical or electronic. The required level of detail of each of these disciplines is dependent on which physical phenomena are most dominant. Often a combination of these 8th European Workshop On Structural Health Monitoring (EWSHM 2016), 5-8 July 2016, Spain, Bilbao www.ndt.net/app.EWSHM2016 More info about this article:http://www.ndt.net/?id=19994