Skip to content Skip to navigation

Robust Design and Operation of Natural Ventilation Systems

Project Team

Catherine Gorlé , Martin Fischer, Chen Chen
 

Overview

Natural ventilation has the potential to save 10% to 30% of building energy consumption. The primary challenge is that natural ventilation is strongly influenced by uncertainties in the building's design and operating parameters. Our project goal is to enable robust design and operation of natural ventilation system by developing an efficient multi-fidelity modeling framework that can predict system performance with quantified confidence intervals. Fast, robust models will support initial design choices; more expensive, detailed simulations will support fine-tuning the design. Once operational, robust performance will be achieved by learning improved model settings and establishing robust control systems based on measurement data.
 
To evaluate the predictive capabilities in an operational building, we focus on the natural ventilation system in the Yang and Yamazaki Environment and Energy Building (Y2E2) building. We will perform experiments and establish novel multi-fidelity modeling and inference methods to support robust natural ventilation system design and operation.

Project Background

Research Motivation

40% of the total US energy consumption is in residential and commercial buildings. Efficient natural ventilation strategies could save 10% to 30% of that energy consumption, but design procedures for optimal and robust natural ventilation systems are not well established. The primary challenge is that natural ventilation flow and heat transfer phenomena are strongly influenced by the building’s highly variable and uncertain operating conditions. Ignoring or underestimating the effect of this variability in the design process translates into a risk of failing to meet thermal comfort and air quality criteria. 

Industry Example

The Y2E2 building was designed to use 56 percent less energy than a comparable structure built in a traditional fashion. It has a variety of sustainable features to achieve this goal, including the use of a passive natural ventilation system to keep the building cool and ventilated in hotter months. The hallways, open areas, and lounges connected to 4 atria are cooled using a buoyancy-driven night flush. Cool air enters through mechanically operated windows on the 1st through 3rd floors and drives out warmer air through mechanically operated louvers at the top of each atrium. The resulting overnight cooling of the building’s thermal mass balances out the subsequent heating during the day.
 
The selection of the Y2E2 building as out test case is motivated by its extensive measurement system, which provides the necessary data for validation of out models. Validation with an operational building is essential to evaluate the true predictive capabilities of the framework.

Research Objectives

Based on the results of previous research results, we have formulated two research objectives that will advance our goal of developing a multi-fidelity modeling framework with UQ to support the design and operation of robust natural ventilation systems:
  1. Using Bayesian inference to quantify uncertainty in the internal load
  2. Measuring spatial variation in the temperature in Atrium D and surrounding always
  3. Based on the CFD temperature field, selecting locations to place thermocouples in zones with different cooling rates, and to ensure that the average of all sensor measurements is representative of the volume-averaged temperature.
  4. Deploying sensors on each floor, and collecting temperature data.

The measurements will support the formulation of a functional relationship between the building sensor data and the volume-averaged temperature. This will support validation of the multi-fidelity modeling framework, but will also provide more accurate data for the inference procedure. 

Research Update:

CFD simulation set-up

The CFD simulations are performed using ANSYS Fluent v16, solving the Reynolds-averaged equations for conservation of mass, momentum and energy. The effects of buoyancy are modeled using the Boussinesq approximation. The Reynolds stresses are modeled using the RNG k-epsilon turbulence model. The equations are discretized using second order schemes in space and a second order implicit scheme in time with a 0.5s time step. The PISO algorithm is used for pressurevelocity coupling.

The computational domain is comprised of the common areas around atrium D of the building and the surrounding outdoor area. The far field boundary is at least 20m (around 1 building height) away from the building. The computational grid for Atrium D consists of 3.5 million hexahedral cells. Gradual grid refinement is used to achieve a resolution of 0.03 m to 0.06m around the windows. A constant uniform pressure condition is imposed on the far field boundary, together with the measured outdoor temperature as a function of time. The initial condition for the indoor air temperature on each floor is specified following the available measurements at the start of the nightflush. An isothermal boundary condition imposes a temperature that is 2K below the initial air temperature on the thermal mass surfaces, while an adiabatic boundary condition is used on the walls. On the window openings, a porous jump condition is added to represent the pressure drop caused by the opening angle of the window.

Design of experiments

The locations of the thermocouples were selected based on CFD simulation results for the temperaturefield. First, we divided each floor into five zones: two zones for each hallway, the atrium, and two zones near the windows. Secondly, we considered the temperature field predicted by the CFD simulations for each zone and calculated the error between the local point-wise temperature T and the volume-average temperature T_{volavg} as follows:

Error Function

with n_{t} the number of time steps in the simulation. The lower the error, the more representative the point-wise local temperature is for the volume-averaged air temperature of the relevant zone. A contour plot of this error is shown in Figure 1; temperature sensors were located at positions where the error is small, also considering practical factors for placing sensors in an operational building. In addition, some sensors were placed in areas where local validation of CFD results is of interest. The final sensor locations are illustrated in Figure 2.

Initial experiments were conducted during the nightflush, recording temperatures at 29 locations within Atrium D using thermocouples. Sensors measuring the indoor air temperature were positioned 0.5m away from the nearest wall; each floor also has 2 to 3 thermocouples attached to the floor surface to measure the thermal mass temperature. In addition, flow rates near windows on the 2nd and 3rd floor were recorded using hotwires. Temperatures and flow rates were logged at a sampling rate of 10s.

Temperature error contour plot of 2nd floor                                      Thermocouple and hotwire locations (1

Figure 1. Temperature error contour plot of 2nd floor                      Figure 2. Thermocouple and hotwire locations

Results and future work

The presentation of the results focusses on a 4-hour period, from 20:00 to 24:00 on September 6th 2018. The air temperature measurements, shown in Figure 3, indicate that zone 1, which is a hallway between offices, has the highest temperature during the nightflush compared to other zones, and the cooling rate is relatively low. The building built-in sensor is located in this zone, which is not directly exposed to the flow of outdoor air entering through the windows. The buoyancy-driven flow effect is much more pronounced in the atrium (zone 4), where the warm air exits through the top louvers. As a result, the temperature in zone 4 is relatively low during the nightflush, and the cooling rate is high. Thus, the measurements confirm that spatial variability in the temperature is an important reason for discrepancies observed in the previous study. 

To further support this point, we used the integral model with UQ to predict the temperature during one of the experimental nights. In Figure 4, the results are compared to the point-wise building sensor measurements and to an approximated volume-averaged temperature calculated from the thermocouple measurements. The discrepancy between the model predictions and the measurements is significantly smaller when using the approximated volume-averaged temperature.

Future work will focus on deriving a transfer function that converts building measurements to an approximated volume-averaged temperature, such that historical building data can be used for model validation. In addition, we will compare the temperature field predicted by the CFD model to the experiments.

Sensor locations and experiment measurements results                        ,            Validation predictive computational models

Figure 3. Sensor locations and experiment measurements results      Figure 4. Validation predictive computational models

Original Research Proposal

 

Final Project Report

Funding Year: 
2019
Stakeholder Categories: 
Operators/Facility Managers
Designers