Computational Fluid Dynamics and Data–Based Mechanistic Modelling of a Forced Ventilation Chamber ⋆ O. Tate * E. D. Wilson * D. Cheneler * C. J. Taylor * * Engineering Department, Lancaster University, UK (e-mail: o.tate@lancaster.ac.uk, e.d.wilson1@lancaster.ac.uk, d.cheneler@lancaster.ac.uk, c.taylor@lancaster.ac.uk) Abstract: The research behind this article ultimately concerns control system robustness and optimisation for the regulation of temperatures in multiple buildings that are linked to a controllable external heating supply network. Lancaster University campus is being used as a case study, for which the building management system provides data. Nonetheless, situations arise when it is difficult or expensive to obtain suitable data for specific rooms or buildings and, in such cases, computational fluid dynamics (CFD) models are utilised to investigate relevant heat transfer phenomena. Such models can be limited by their complexity and they are inappropriate for model–based control design. Hence, the present article investigates a hybrid approach based on both CFD and data–based mechanistic (DBM) models. DBM models are obtained initially from statistical analysis of observational time–series but are only considered credible if they can be interpreted in physically meaningful terms. A laboratory forced ventilation chamber is used to investigate the modelling issues arising and to make recommendations relating to the wider project. The chamber is first discretised into finite volumes and the associated Navier–Stokes equations are solved to determine the physical properties of each zone. The model responses are compared with experimental data and analysed using the DBM approach. Keywords: Computational fluid dynamics (CFD); data–based mechanistic (DBM); heating, ventilation and air conditioning (HVAC); micro–climate. 1. INTRODUCTION The research behind this article concerns control system robustness and overall system optimisation, for the regula- tion of temperatures in buildings that are linked to a con- trollable external heating supply network. This is the case, for example, with the Lancaster University campus, for which a central energy centre supplies the hot water used to heat around 50% of the buildings (Ioannou, 2016). The authors are developing demand–side control concepts (e.g. Kim, 2013) to address multiple buildings on this network, i.e. the control actions for one building are accounted for when choosing actions for the other buildings, potentially increasing energy efficiency and improving thermal condi- tions for the building occupants. Heating, Ventilation and Air Conditioning (HVAC) sys- tems have high energy requirements, hence there is con- siderable interest in the development of improved op- timisation tools, micro–climate control algorithms and energy management systems. Although the literature is vast, selective examples of such research include Yang and Wang (2013), Kim (2013), Goyal et al. (2013), Kossak and Stadler (2015) and Mayer et al. (2017), while Mirinejad et al. (2012) and Lazos et al. (2014) provide useful reviews ⋆ This work is supported by Engineering and Physical Sciences Research Council (EPSRC): EP/M015637/1. The DBM modelling algorithms are available within the CAPTAIN toolbox, which may be downloaded from: http://www.lancaster.ac.uk/staff/taylorcj/tdc. focusing on intelligent control and energy management, respectively. Numerous approaches for modelling heat transfer phenom- ena and energy use have been developed over the past few decades. The models obtained are commonly categorised into either physically–based models or models that are statistically identified from data (Foucquier et al., 2013). Whilst the former include various zonal and multi–zone approaches, CFD models are probably the most widely used in practice. CFD models consist of deterministic equations based on the classical laws of physics (e.g. Hong et al., 2017). However, such models are limited by their complexity and they are generally inappropriate for model–based control design. Data–based models, by contrast, are usually much sim- pler and are identified using techniques from the machine learning or system identification literature. Examples in the context of building micro–climate include the con- sideration of well–mixed zones (Janssens et al., 2004), genetic algorithms (Ryozo and Kazuhiko, 2009), change point models (Paulus et al., 2015) and Hammerstein model forms (Tsitsimpelis and Taylor, 2014), among many oth- ers. Various hybrid approaches have also been proposed. For example, Agbi et al. (2012) apply data–based meth- ods to multi–zone thermal systems, while Price et al. (1999) develop data–based mechanistic (DBM) models for agricultural buildings. DBM models are obtained initially