Smart healthy habitats

People-Centric Habitats:

A Smart Approach to Healthy and Comfortable Living

Abstract of the research line:

People spend more than 80% of their time indoors (Superior Health Council, 2017). Their living and working environments play a key role in their health, lifestyle, and well-being. In current building practice the focus is on energy performance. However, health, comfort and user acceptance should receive equal attention.

The surge of smart devices and IoT creates opportunities to learn about the condition of the building, and act upon it to create an ‘optimal’ condition. The large amount of available sensors and actuators prohibits their direct use by people, and demands for an automated approach.

In this thesis we will aim at modelling a building from a ‘system’ perspective, i.e. by considering the inputs (controllable and uncontrollable) and the outputs:

Inputs Outputs uncontrollable inputs
weather conditions (outside temperature, wind, sun, …)
preferences of the habitants

Simulation tools of buildings and their envelopes are available, such as to allow for virtual prototyping of realistic buildings, and for obtaining years worth of usage data. The latter will be used to construct simplifying meta models, which are required in MPC or reinforcement learning types of control strategies.

The main goal is to create people-centric habitats which support their users in creating a healthy, comfortable and energy efficient indoor environment. This will be achieved by providing its users with contextualized feedback. Well-being and empowered user-control are the main drivers to reach this goal. At the same time, to assure that climate goals can be met, we should quantify the impact of human-centric buildings on the energy consumption.

In addition, one-size-fits-all solutions no longer meet today’s needs: people want a habitat that they can adapt to their individual requirements. The problem is that the control actions of users are not always effective. The transition from energy-hungry to nearly-zero energy buildings requires a shift from intuitive control (e.g. opening of windows) to informed use (Ehrhardt-Martinez, 2010).

As comfort control and energy efficient behavior are far less intuitive in nearly-zero energy buildings than in traditional homes, efficient control actions are far from straightforward. We will create control algorithms based on artificial intelligence technology to create efficient and robust feedback to the users (engineering aspect). To speed up the learning curve of the algorithm, it is fed with simulated data from physical modelling (including building geometry, heat transfer, impact of the sun and wind, ...). Feedback actions can be automated (through motors f.i.) or manual (human action). The user interface and user-feedback mechanisms should be designed for high user acceptance (behavioral aspect). Nudging will be used to optimize user acceptance.

The rise of connected devices (IoT) e.g. light and temperature sensors, of artificial intelligence (AI) and machine learning enables this. It allows

  1. better mapping of people’s behavior,

  2. a better understanding of their requirements, and

  3. adapting comfort systems accordingly.

We will explore these aspects in depth to formulate a novel procedure to develop future-proof, integrated comfort control systems with user feedback. We strive for a symbiosis in which users, indoor environment and comfort systems reinforce each other to find an optimal balance between health, comfort and energy efficiency. Throughout the whole process, we treat users as unique human beings. We explicitly put them in the center of our projects by continuously considering and thoroughly investigating the effects of our (technological) developments on their wellbeing and (mental) health.

Topic 1: Surrogate models of buildings

This topic handles the improvement of modelling the dynamics of buildings, based on state-of-the-art building simulators (EnergyPlus). In these simulations the following aspects need to be taken into account:

  1. The physical quantities that are considered in the models include the airflow, temperature distribution and concentration of particles (related to air quality) in buildings. 

  2. The boundary conditions (which can also be interpreted as design parameters) are: the construction of the building (building layers, insulation etc.) and its technical installations (heaters, cooling, fans, ...).

  3. In addition, presence/absence of users (habitants, dwellers) have a great impact on the evolution of above physical quantities. These can be modelled as noise sources, of which the statistical properties are known to some extent. Namely, different user profiles have been measured that model how people move through a building, in which rooms they reside, how much energy they use and how these impact for instance the airflow and temperature distribution in the building.

In this thesis proposal, we would like you to build on a framework to derive thermal models of buildings based on data. These models should be of sufficiently low complexity, such that their simulations are very fast (much faster than by using CFD solutions), and such that they faithfully describe the observed data (i.e. use data-driven modelling).

The approach would be to obtain a simplified model by using system identification, applied to simulation data (coming from EnergyPlus). To that end, model reduction and machine learning techniques can be used.

An RC-equivalent (resistor capacitor) of the thermal behaviour of the building can be constructed to provide insight into the thermal energy transfers in the building.

This model should reflect the effect of, amongst others,

  1. opening windows

  2. turning on and off the heater, ventilation and cooling system (HVAC)

  3. people that move throughout the building

  4. electrical appliances that are being used and produce heat

The goal of the model is to be used, at a later stage, for optimizing control strategies, in near real-time. Also, the same identification tools would be used afterwards to update the model as real data is collected (see topic 2), such as to reflect the dynamics of buildings in real situations, in varying conditions.

Topic 2: Instrumentation for healthy buildings

This topic is complementary to the first topic. The goal is to design and deploy sensor configurations in buildings, such as to achieve the required level of accuracy to provide reliable information to the controller. This may be achieved via sensor fusion, with (multiple) low cost sensor(s), via calibration.

Smart homes are equipped with sensors that measure the physical parameters in a room and its environment (in- and outdoor temperature and humidity, wind direction, solar irradiance, pollutants ….). These parameters influence the visual and thermal comfort in a room and have a possible impact on the physical and mental health of the inhabitants. The HVAC actions (Heating Ventilation and Air Conditioning) act on the indoor conditions, e.g. when the indoor temperature gets too high, the airconditioning is activated, or when the solar irradiance exceeds a certain threshold, the solar blinds are deployed to reduce visual hindrance. It is important that the information provided to these systems are of a sufficient quality, such that the right decisions can be taken.

A recent trend is to fully automate building control. However, experience has learned that users dislike this, as they prefer to have a certain degree of control (e.g. some people prefer to open a window when the indoor temperature gets too high, as opposed to turning on the ventilation system, which consumes energy). On the other hand, in ‘Nearly Zero Energy Buildings’ (NZEB), the most optimal decision is often not the most intuitive one, such that users need informed decision making. For instance, depending on the outdoor temperature (e.g. summer or winter), it may be more energy efficient to either open a window or to trigger the ventilation system to supply fresh air.

The challenge in this thesis topic is to match a sensor configuration with the required precision and accuracy to reach the desired level of health and comfort, while not intruding too much in the user’s personal space.

Topic 3-5 Improved control strategies for solar shading devices

Solar shading devices as e.g. venetian blinds or roller blinds control the amount of sunlight to enter through daylight openings ( e.g. windows and skylights) of a building. (see Figure 1)

Figure 1: different solar shading devices

They prevent overheating of the building in the cooling season and can be used to reduce daylight glare. Glare is a feeling of visual discomfort that arises when too much light enters a daylight opening or when too high luminances in the field of view of an observer cause visual hindrance. Currently, the most widespread parameter to assess if a user experiences visual hindrance is the daylight glare probability (DGP) which expresses the percentage of people that would experience visual discomfort.

Depending on the geographical location and meteorological conditions, different control strategies can be developed to deploy the solar shading devices. A widespread strategy is the so-called “cut-off strategy” which orients the slats of the venetian blinds to block any direct solar radiation (see Figure 2.)

Figure 2: cut-off strategy

The approach to the research on control strategies is two-fold: on the one hand, numerical simulations of office environments allow to compare the effect of different control strategies on visual comfort and energy use. A standard office (which is internationally defined) is modeled and by coupling the software packages EnergyPlus and Radiance the visual comfort and energy consumption can be simulated. On the other hand, experimental studies can be used, either in a laboratory environment (mimicking the standard office) or in a living laboratory, e.g. a real office environment. These experimental studies involve real users that behave individually and depending on different conditions. When these user-data are coupled back to the simulation software, simulated predictions become more realistic.

The aforementioned techniques have been developed in the framework of the PhD thesis of Charlotte Goovaerts (2018). To assess the visual discomfort, she compared the use of an expensive high-resolution camera to a low-cost camera connected to a raspberry-pi. Both yielded similar results when placed behind the user with the camera pointed at the window, as in Figure 3.

Figure 3: camera view

Future research tracks

Future research in this topic will focus on three research tracks that correspond to different research profiles:

Engineer or Engineering:

In real office environments, the location of the camera should be placed against the ceiling (possibly integrated to the electrical lighting.) This would require at least one camera per office. Also, the effect of having multiple users in one room should be looked into, as well as the effect of their viewing direction (visual discomfort will depend on whether people are looking straight at the window or if the window is to their side.) Secondly, one could research whether it is possible to use a single outdoor camera and use a one-time calibration in each room.

Machine learning / Artificial Intelligence

Machine learning techniques could be applied to optimize the control strategies to deploy the solar shading devices. By coupling EnergyPlus and Radiance and feeding them synthetic data, the system will self-learn to yield better control algorithms. The control strategies should be robust, meaning that they should adapt to the many uncertainties associated to buildings, as e.g. the number of people present or different building topologies. More specifically, a self-learning algorithm can be developed, which makes suggestions to the user, and which can be overridden by this user. The algorithm learns based on the acceptance and overrides by the user to improve its decision making. If multiple/different users are at stake, the algorithm can cluster the users, resulting in personalized suggestions/decisions.

Social sciences

To increase the acceptance of automated control strategies, feedback will be provided to the users to inform them about the decision tree of the control strategy. Also, in case the user wants to override the automated system, feedback about the override action on visual comfort or energy performance can be provided to the user. User interfaces are the access points where users interact with designs. User interface design is a craft that involves building an essential part of the user experience; users are very swift to judge designs on usability and likeability. We will aim at building interfaces that users will find highly usable and efficient.



John Lataire
ETEC, TONA (Valéry Ann Jacobs), ARCH (Filip Descamps)
Back to top