Summary for CIFE Seed Proposals for Academic Year 2017-18 Proposal number: Proposal title: Principal investigator(s) 1 and department(s): Research staff: Robust Design of Natural Ventilation Systems Catherine Gorlé, Martin Fischer, Civil and Environmental Engineering Department Chen Chen Total funds requested: $80,644 Project URL for continuation proposals Project objectives addressed by proposal 2 Expected time horizon Type of innovation Abstract (up to 150 words) http:// Sustainable 2 to 5 years Incremental The problem: Natural ventilation could save 10% to 30% of a building s energy consumption, but design procedures for natural ventilation systems are not well established. The primary challenge is that natural ventilation is strongly influenced by the building s uncertain operating conditions, which translates into a higher risk of failing to meet design criteria. The proposed solution: We will develop an efficient multifidelity modeling framework to predict natural ventilation system performance with quantified confidence intervals. Fast, robust models will support initial design choices; more expensive, detailed simulations will support fine-tuning the design. Uncertainty quantification will enable accounting for variability in the operating conditions to mitigate the risk associated with naturally ventilated buildings. The proposed research approach: Our modeling efforts will focus on the Y2E2 Building to evaluate the model s predictive capabilities in an operational building. We will implement additional measurements, integral modeling strategies, and CFD model capabilities to establish the multi-fidelity framework and support robust natural ventilation system design. 1 The PI(s) must be academic council member(s) at Stanford. 2 For this and the next points, delete the answers that don t apply to your proposal.
Engineering Problem Robust Design of Natural Ventilation Systems 40% of the total US energy consumption is in residential and commercial buildings [1]. Efficient natural ventilation strategies could save 10% to 30% of that energy consumption [2], 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 these uncertainties in the design process translates into a risk of failing to meet thermal comfort and air quality criteria. Our project goal is to develop an efficient multi-fidelity modeling framework, as shown in Fig. 1, to predict the performance of natural ventilation systems with quantified confidence intervals. The development of a multi-fidelity framework is essential to support important decisions in early design stages, such as the location and size of atria for buoyancy driven ventilation. Simple but robust models with very fast turnaround times will support these initial design choices, while more detailed and expensive simulations can verify the simple model s assumptions and fine-tune the design in the detailed design stages. The incorporation of uncertainty quantification is essential to account for the inherent variability in the building s operating conditions at all design stages. This will effectively mitigate the risk associated with naturally ventilated buildings and result in smart systems that can compensate for variability in operating conditions over the building s lifespan. Hence, the proposed framework will support robust design and optimal operation of natural ventilation systems, and promote their more widespread implementation. The modeling efforts will focus on the natural ventilation system in The Yang and Yamazaki Environment and Energy Building (Y2E2) building, shown in Fig. 2. 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 1 st through 3 rd 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 building is equipped with an extensive measurement system. 5+ years of energy consumption and building Figure 1: Multi-Fidelity Computational Framework with Uncertainty Quantification. C. Gorlé, M. Fischer Robust Design of Natural Ventilation Systems 2
conditions data for 2,400 sensors at a 1-minute interval are available. Of particular interest are the air temperature measurements on each floor in all four atria, and the outdoor temperature, wind speed, and wind direction recordings, since these will be used for analysis and validation of the model. Validation with an operational building is essential to evaluate the true predictive capabilities of the framework, and it also enables identifying the dominant uncertainties to prioritize future research. Figure 2: Y2E2 building Theoretical and Practical Points of Departure Motivation for Model Component Selection Fig. 1 indicates the two types of models incorporated in the proposed framework: an integral model that solves two equations for the evolution of the volume-averaged indoor air and thermal mass temperatures, and a detailed three-dimensional computational fluid dynamics (CFD) model. The selection of these two models, which represent very different levels of fidelity, is motivated by considering the hierarchy of models available to evaluate the performance of a natural ventilation design, as shown in Fig. 3. It reflects increasing levels of complexity in the representation of the physical processes. When moving through this hierarchy of models one will obtain a more accurate prediction of the flow and heat transfer phenomena, thereby reducing the uncertainties related to reduced order modeling. At the same time, the computational cost increases several orders of magnitude, and a higher level of detail in the input is required, resulting in a higher number of uncertain input parameters. This significantly increases the cost of quantifying the uncertainties related to the design or operating parameters. As a result, a combination of the extremes in this model hierarchy has the potential to maximize the benefits of using a multi-fidelity simulation strategy: the integral model supports quantifying uncertainties in multiple input parameters, and the CFD model can reduce uncertainties related to simplified physics models in the integral model. Motivation for Multi-Fidelity Framework with Uncertainty Quantification (UQ) A vast amount of literature on the fluid mechanics of natural ventilation, experimental measurements, and modeling has been published in the past decades. The potential of CFD to provide detailed flow and temperature fields is recognized, and a variety of investigations on the use of CFD to model natural ventilation flows has been performed. Unfortunately, this has not resulted in the routine integration of CFD in the design process. Two important review papers indicate principal reasons for this. Figure 3: Hierarchy of models for predicting the performance of natural ventilation systems C. Gorlé, M. Fischer Robust Design of Natural Ventilation Systems 3
Linden [3] provides a comprehensive review of the fluid mechanics of natural ventilation. The potential of CFD is recognized, but the conclusions state: Even if CFD calculations can provide accurate answers to ventilation flows, the question still remains of how these will be used in building design. The designer requires an intuition of the likely effects of changes in the design or the operation of a building. Even specific answers from each design option will not provide that. This motivates our choice for a multi-fidelity simulation strategy, where simple robust models can provide this intuition in early design stages. In the detailed design stages, CFD simulations can then be used to support the simple model s assumptions and fine-tune the design. A recent review paper by Etheridge [4] highlights the need for uncertainty quantification. The author specifically notes that the specification of the boundary conditions is a major challenge, and that sensitivity of the results to the boundary conditions requires more understanding. The proposed research will not only increase our understanding of the sensitivity to the boundary conditions, but will provide a tool to quantify the uncertainty introduced by the inherent variability in these boundary conditions. The resulting predictions with quantified confidence intervals will effectively support robust design decisions. Research Methods and Work Plan Preliminary study A preliminary study was performed to investigate the predictive capability of the proposed multifidelity model [5]. The analysis focused on the night-flush ventilation during two nights of 2013, one in spring (Tuesday April 30th) and one in summer (Wednesday July 3rd). The evolution of the volume-averaged indoor air temperature and the one-dimensional profile of the thermal mass temperature (T a and T tm ) is given by: V a, ρ " and C p,a are the volume, density and specific heat of air, Q conv,n is the convective heat flux from the n th thermal mass, Q nv is the natural ventilation heat flux, Q i the internal heat flux, and k n the thermal conductivity of the n th thermal mass. The equations are coupled through the convective heat flux, which defines the boundary conditions of the equation for the thermal mass temperature. Five uncertain model parameters were identified: the heat transfer coefficients on the floors and on the walls, which determine Q conv,n ; the discharge coefficient, which determines Q nv ; the internal heat flux Q i ; and the initial value for the thermal mass temperature T tm,n. Two uncertainty quantification studies were performed, using different input distributions for these parameters: 1. Initial UQ study: all uncertain parameter distributions are estimated using information from previous studies and literature [6]. 2. Updated UQ study: the initial distributions for the heat transfer and discharge coefficients are updated with results obtained from detailed three-dimensional CFD simulations. The uncertainties were propagated using a polynomial chaos approach [5]. Figure 4 plots the results for both distributions. Despite its simplicity, the initial UQ analysis predicts a 95% confidence interval that encompasses the measured air temperature 75% of the time. However, the width of confidence intervals is relatively large at 1.5-2K. When using updated distributions for the heat transfer coefficients and the discharge coefficients, obtained from the CFD simulations, C. Gorlé, M. Fischer Robust Design of Natural Ventilation Systems 4
Figure 4: Indoor (red) and outdoor (blue) temperature measurements, compared to mean and 95% confidence interval for the air temperature from the updated UQ study (black) and the initial UQ study (grey). the maximum standard deviation of the computed indoor temperature drops by 20 to 30%. However, the measurements are now outside the predicted confidence interval in 45% of the points, and the overall too high cooling rate indicates a bias in the predictions. The preliminary study indicates the promising capabilities of the proposed framework to provide fast predictions with quantified confidence intervals in initial design stages, and enable decreasing uncertainty in the predictions using more expensive models later. However, the bias observed after introducing more accurate information from the CFD model warrants a more detailed analysis of the model behavior and results, which will be the focus of the proposed research. Work Plan The proposed research will implement additional measurements, CFD model capabilities, and integral modeling strategies, to address four fundamental challenges identified from the preliminary results. 1. Characterization of the uncertainty in the initial thermal mass and office wall surface temperatures. A sensitivity analysis performed for both the integral and CFD model has indicated the importance of the initial thermal mass temperature, and of the office wall temperatures, which also exchange heat with the common spaces. Additional measurements will be aimed at better characterizing these uncertainties, such that the model can be improved to correctly account for these effects. The measurement campaign will focus on measuring the thermal mass and office wall temperatures throughout Atrium D shortly before the start of the night flush. In addition, time series of surface temperatures throughout the night flush will be recorded in select locations to enable validation of the model results for the thermal mass temperatures. 2. Characterization of the uncertainty in the mass flow rates calculated from the CFD and integral models. In the preliminary study, the mass flow rates calculated from the CFD were used to extract updated distributions for the discharge coefficients. However, the CFD results can also be subject to uncertainties, resulting from model assumptions and simplifications. Additional measurements of the flow rates through the different windows during the night flush will be performed to better characterize these uncertainties, and identify opportunities for improving the CFD predictions. The capability to account for uncertainties in the CFD results when using them in the integral model will result in a more robust multi-fidelity model. In addition, CFD simulations are likely to be used for evaluating internal flow patterns in the more detailed design stage. The flow rate measurements will provide essential information to validate the predicted velocity fields. C. Gorlé, M. Fischer Robust Design of Natural Ventilation Systems 5
3. Implement a dynamic thermal simulation CFD model. At present the CFD model assumes a constant thermal mass temperature. The influence of this assumption on the calculated values for the heat and discharge coefficients should be determined. In addition, this assumption is likely to affect the flow rates through the building, since it influences the air temperature and the buoyancy driven flow effects. To investigate the resulting inaccuracies in the prediction, we will implement a dynamic thermal model that solves for the evolution of the thermal mass temperature in the CFD simulation. The results of this analysis will provide guidelines regarding the definition of boundary conditions at thermal mass surfaces, and the uncertainty that can be expected in CFD results. 4. Investigate inference methods to characterize uncertainty in the internal load and infiltration. The preliminary results showed that the internal heat load and infiltration in the building have an important effect on the model results, and are both highly uncertain. This significantly complicates the validation of the results, and it is difficult to conclude on the model s predictive capabilities without a more accurate characterization of these parameters. To address this challenge, we will use Bayesian inference techniques to extract probability distributions for these uncertainties. The analysis will first consider nights where the ventilation system was not active to eliminate the effect of the natural ventilation heat flux. The distributions inferred using that data will then be applied in the forward analysis for nights where the ventilation system was operational. Expected Results: Findings, Contributions, and Impact on Practice The primary result of the proposed research will be a multi-fidelity computational framework for designing natural ventilation systems in buildings. To achieve this result, we will obtain experimental data, and develop advanced integral and CFD models with uncertainty quantification. The analysis of the measurements and model results will improve our understanding of the dominant physical processes in naturally ventilated buildings and thereby advance the predictive capabilities of the models and uncertainty quantification algorithms. The research is intended to support the design of robust natural ventilation systems. Incorporating the results of the proposed project in design practice will be relatively straightforward, and will be promoted by dissemination of the results at conferences and workshops. Incremental implementations, such as adopting guidelines for the definition of boundary conditions in CFD simulations, or the use of an uncertainty quantification algorithm in combination with an integral model are envisioned in the short term. The final multi-fidelity modeling framework is intended to provide engineers with the capability to obtain performance predictions with quantified confidence intervals at all design stages. We anticipate that such modeling frameworks will effectively mitigate the risk associated with naturally ventilated buildings and promote their widespread implementation, thereby significantly reducing building energy consumption. Industry Involvement Since we have access to the building, its energy-related performance data for many years, and are in control of the measurements and modeling tools needed, the success of this proposal does not depend on access to data from CIFE members. CIFE members with design software could help the research team connect the multi-fidelity modeling framework to software strategies of combining computational building representations with a range of prediction tools. Building owners and design-build teams interested in high-fidelity predictions of building performance can help the research team connect the multi-fidelity modeling framework to current practice to speed up its C. Gorlé, M. Fischer Robust Design of Natural Ventilation Systems 6
widespread adoption. Should a CIFE member have an appropriate dataset it could potentially be helpful to validate the findings of this project. Research Milestones and Risks The following milestones will be used to measure the progress of the proposed effort: 1. November 2017: Completed additional measurements for thermal mass temperatures and flow rates. 2. April 2018: Completed investigation on the influence of using a dynamic thermal mass model, with comparison to the measurements. Submission of paper. 3. June 2018: Completed inverse analysis to obtain distributions for the internal load and infiltration in the Y2E2 building. 4. September 2018: Submitted paper on validation of the multi-fidelity modeling framework with Y2E2 measurements. The main risk in the proposed project is related to the complexity of studying an operational building, for which the models inherently have a very large number of uncertain parameters. The challenge is to correctly isolate the most important physical processes, and ensure each individual one is modeled as accurately as possible. If this is not achieved, incorrect conclusions regarding the physical modeling of separate processes, such as the convective heat transfer or natural ventilation heat flux, could be drawn. The proposed approach of performing additional measurements and incorporating uncertainty quantification in the models is specifically intended to mitigate this risk. Next Steps The proposed research will provide an excellent starting point for additional research on robust design and operation of natural ventilation systems. The multi-fidelity modeling strategy should be extended to consider wind-driven ventilation in addition to the Y2E2 s buoyancy driven system. This entails a multi-scale modeling challenge, involving a more detailed analysis of interaction between the building and the surrounding environment. The availability of sensor measurements in operational buildings also provides exciting opportunities for the implementation of smart natural ventilation systems, where measurement data is used in combination with inverse methods and machine learning to optimize the operation of the natural ventilation system over the building s lifetime. Proposals addressing these topics will be submitted to NSF and/or DOE. References [1] U.S. Department of Energy Building Technologies Office. Multi-Year Program Plan, 2016. [2] A. Walker, Natural Ventilation, retrieved from http://www.wbdg.org/resources/ naturalventilation.php (2014). [3] P.F. Linden, The fluid mechanics of natural ventilation, Annual Review of Fluid Mechanics, 31, 201 38 (1999). [4] D. Etheridge, A perspective on fifty years of natural ventilation research, Building and Environment, 91, 51-60 (2015). [5] G. Lamberti, C. Gorlé, Uncertainty quantification of an integral model and a CFD model to predict natural ventilation in Stanford s Y2E2 building, Proceedings of the 7 th European and African Conference on Wind Engineering, Liege, Belgium, July 3-6, 2017 (accepted). [6] E. Hult, G. Iaccarino, and M. Fischer, Simulation of night purge ventilation using CFD and airflow network models, Stanford Internal Report (2011). C. Gorlé, M. Fischer Robust Design of Natural Ventilation Systems 7
Budget Sponsor: Stanford CIFE seed 2017 Submission Type: New Budget Preparation Date: 4/13/2017 Budget Start Date: 10/1/2017 Project Name: - Department: Civil Engineering Principal Investigator: Catherine Gorle Administrator: Tanvi Samdani Period 1 All Periods From 10/1/2017 10/1/2017 To 9/30/2018 9/30/2018 Personnel Salaries Faculty Gorle, Catherine Summer 15.0% 6,451 6,451 Total Faculty Salaries 6,451 6,451 Graduate Students TBD Academic 50.0% 29,757 29,757 Summer 50.0% 9,919 9,919 Total Graduate Student Salaries 39,676 39,676 Total Salaries 46,127 46,127 Benefits Faculty 2,093 2,093 Graduate 2,143 2,143 Total Benefits 4,236 4,236 Total Salaries and Benefits 50,363 50,363 Other Direct Costs Tuition TBD Academic 50.0% 20,461 20,461 Summer 50.0% 6,820 6,820 Total Tuition 27,281 27,281 Other Direct Costs Measurement Equipment 3,000 3,000 Total Other Direct Costs 30,281 30,281 Total Direct Costs 80,644 80,644 Less Tuition (27,281) (27,281) Modified Total Direct Costs 53,363 53,363 Total Amount Requested 80,644 80,644 Rates Used in Budget Calculations Benefit Rates Faculty: FY 1 32.45%; FY 2 32.45%; FY 3+ 32.45%; Graduate: FY 1 05.40%; FY 2 05.40%; FY 3+ 05.40%; Indirect Cost Rates C. Gorlé, M. Fischer Robust Design of Natural Ventilation Systems 8