Virtual Sensor Development Based on Reduced Order Models of CFD Data

Abstract

Manufacturing industry is facing the challenge to cope simultaneously with the high demanding trends of the market (i.e. personalization) and the high quality standards. In this context, it is of utmost importance the implementation of zero-defect strategies relying on sensing and control systems to improve automation and minimize failures. Non-invasive physical sensing is not always feasible. In those cases, virtual sensors may stand for a great tool to support the development of enhanced control systems. In this work, a new methodology for building virtual sensors is presented. The novelty of the methodology relies on the coupling of two modeling techniques: detailed computer-aided engineering (CAE) simulations and model order reduction (MOR) techniques. The result is a realtime simulation model that can provide the variables required by the control system, using other available data (physically sensed) as input. Hence, the coupling between virtual and physical sensors can provide a richer description of the process to improve the process control and performance. To demonstrate the methodology, a case study is presented, where the virtual sensor is aimed at defining the flow pattern (fluid interface position) within a flow-focusing microfluidic device. The resulting virtual sensor can predict in real-time the shape and location of a multiphase fluid interface inside the microchip, based on the volumetric flow rates measured at the chip inlets. The interface position monitoring provided by the virtual sensor is of great relevance to improve the control and performance of such a microfluidic system.