1. Management Of The Built Environment To Reduce Exposure Risk
Team Members: Pramod Varshney, Can Isik, Chilukuri Mohan, H. Ezzat Khalifa,
Onur Ozdemir, Ramesh Rajagopalan, Priyadip Ray,
James Smith, Jensen Zhang

2. Outline SAC 2005 - Main Concerns Problem
Definition
Motivation and Objectives
Research Needs
Integrated Components of this Task
Optimization and Control
Indoor Sensor Networks
Testbeds
Summary and Future Work

3. SAC Main Concerns
“…a lot of the work was premature pending definition of a plausible problem scenario…”
“…an approach based on temperature or CO2 control might be feasible but should only be considered if the state of the art in indoor environmental controls will be advanced…”
“…necessary to refocus this effort…”
“…consult with practicing HVAC engineers, and make inquiries from professionals in the industry…”

4. The Problem Definition
How can we improve the indoor air quality (IAQ) around each individual in a built environmental system (BES) while keeping the cost at a reasonable level?
Treat built environment as a collection of multiple controllable zones
Shift from one-size-fits all (OSFA) paradigm and move towards have-it-your-way (HIYW) paradigm.

5. There is no Free Lunch!
Improved health, productivity and comfort, at the expense of increased system complexity:
Additional infrastructure
Increase in cost, computation, communication and actuation
Additional burden of coupling effect between zones

6. Motivation and Objectives
Existing “one size fits all” solutions leave many occupants dissatisfied with their environments, limiting productivity and affecting health
“New research shows that higher IAQ improves health, learning and productivity” - Dr. Ole Fanger
Empowering each occupant with the ability to control one’s own environment improves satisfaction and productivity; expected technological changes will help realize this vision. This will require a paradigm shift in HVAC technology
“I predict a dramatic change in HVAC technology in the future” - Dr. Ole Fanger

7. Research Needs
Controlling individual environments while keeping the cost at a reasonable level is an optimization problem, whose solution requires:
Measuring environmental parameters at an individual level with a complex network of sensors
Providing control actuators at an individual scale but with coordination
Reacting to changes in the system variables such as occupancy and weather conditions
In order to customize the IAQ, a network of sensors, controllers and actuators is needed. A wireless network allows low-cost retrofitting of existing buildings.

8. Goal
Implement cost effective personal control of the microenvironment, which enhances individual health, satisfaction and productivity, by integrating sensing, intelligent information processing and distributed control.

9. Integrated Components of this Task
Improve indoor air quality by:
Optimization and control – new methodologies for real-time control
Indoor sensor network – design, placement, data processing, and spatio-temporal profiling
Test-beds – design of new test-beds and implementation of developed methodologies on these test-beds

10. Micro-Level Demand-Controlled Ventilation (DCV)
Main Objective: To improve IAQ around each individual in an office building
Approach: Develop optimization algorithms that will improve IAQ in every single office for each individual*
Constraints: Energy consumption, costs and individual comfort
* Ari, I.A. Cosden, H.E. Khalifa, J. F. Dannenhoffer, P. Wilcoxen, C. Isik, ”Constrained Fuzzy Logic Approximation for Indoor Comfort and Energy Optimization”, 24th International Conference of the North American Fuzzy Information Processing Society (NAFIPS 2005) Proceedings, Ann Arbor, Michigan, June 2005

11. DCV At The Micro Level Emissions in a room*:
CO2 - surrogate for emissions by people or through human activity
TVOC – surrogate for emissions from room contents and furniture
Our Focus: CO2 levels as the IAQ criterion for optimization
Goal: Optimization of ventilation rates (CO2 levels) and energy costs in a multi-zone BES
Approach: DCV at the micro-level with one controllable diffuser in each room (e.g., Variable Air Volume – VAV)
* ASHRAE Std allows for ventilation rates based on both occupancy and floor area. A DCV system based only on CO2 will address the people component but not the passive component.

12. CO2 Levels and Energy For a single zone
CO2 concentration at steady state is known
The energy model is known*
Main energy consumed by the HVAC is proportionally related to the cooling/heating coil load
* S.Atthajariyakul, T. Leephakpreeda,”Real-time determination of optimal indoor-air condition for thermal comfort, air quality and efficient energy usage”, Energy and Buildings, vol.36, pp , 2004

13. Illustrative Example An office building with 25 rooms:
Intra-room uniformity: One temperature and RH value assumed within each room(well-mixed conditions)
Inter-room uniformity: All rooms have identical DBT and RH values
Occupancy: 82 people in the building, with 1-5 people in each room
Goal: Find optimum outside airflow rates (Fo) for each room, with:
Minimal energy consumption
CO2 levels Ci < 800 ppm threshold in each room

14. Preliminary Results
For the same total airflow rate, we compare two alternatives:
OSFA solution: Same airflow rate (Fo) in each office
HIYW solution: Fo adjusted using occupancy information (room-level DCV)

15. # of Occupants for whom Ci > 800 ppm
Preliminary Results
Improved IAQ at the same energy cost
Future Work: Optimization based on airflow dynamics between rooms (Task3.2), variable indoor air temperatures, occupancy variations and variable metabolic rates
DCV at the Micro Level
(HIYW)
OSFA Solution
Power
Consumption (kW)
56.6
# of Occupants for whom Ci > 800 ppm
56 of 82
# of Rooms
where Ci>800 ppm
13 of 25

16. Paradigm Shift from OSFA to HIYW
Requires
Distributed Sensing
Distributed Communication
Distributed Actuation
OSFA
HIYW
Increased system
complexity

17. Integrated Components of this Task
Improve indoor air quality by
Optimization and control – improved IAQ via new methodologies for real-time control
Indoor sensor network – design, placement, data-processing and spatio-temporal profiling
Test-beds – design of new test-beds and implementation of developed methodologies on these test-beds

18. Why Distributed Large-scale Sensor Networks?
Multiple sensors needed to acquire state information in each micro-environment, for implementation of HIYW paradigm
Higher resolution and fidelity data available in a sensor-rich environment will improve distributed monitoring
Sensors need to be networked for system-wide optimization and real-time control of i-BES
Wireless networks facilitate lower-cost retrofitting of existing buildings

19. Previous Work
N. Lin, C. Federspiel and D. Auslander, “Multi-sensor Single-Actuator Control of HVAC Systems”, Int. Conf. For Enhanced Building Operations, Richardson, TX, 2002
Simulation results demonstrating the advantage of using at least one sensor for each room compared to one sensor for many rooms
Assumes a single actuator
Wang, D. E., Arens, T. Webster, and M. Shi. "How the Number and Placement of Sensors Controlling Room Air Distribution Systems Affect Energy Use and Comfort." International Conference for Enhanced Building Operations, Richardson, TX, October, 2002
Simulations results showing the benefits of using more than one temperature sensor to control conditions in the occupied zone of a room

20. Previous Work (cont’d)
H.Zhang, B.Krogh, J.F. Moura and W.Zhang,”Estimation in virtual sensor-actuator arrays using reduced-order physical models”, 43rd IEEE Conference on Decision and Control, December 14-17,Atlantis, 2004
Application of sensor networks for real-time estimates of the values of a distributed field at points where there are no sensors
Assumes linear system models
Clifford C. Federspiel, “Estimating the Inputs of Gas Transport Processes in Buildings”, IEEE Trans. on Control Systems Technology, 1997
Estimation of the strength of a gas source in an enclosure
Applies Kalman filtering for state estimation

21. Multi-sensor Detection and Sensor Placement
Application of multiple sensors for improved detection of indoor pollutants
Improved detection enables reduced exposure of occupants to pollutants
Development of cost-effective sensor placement strategies for improved indoor air quality*
* S.L.Padula and K.R.Kincaid, “Optimization Strategies for Sensor and Actuator Placement”,NASA LaRC Technical Library Digital Repository,

22. Multi-sensor Detection – Example
Simulated concentration profile of a gaseous pollutant released in still air at a point
in a room of 9m x 10m x 6m dimensions with source located at (5,5,0)

23. Numerical Example
An example demonstrating the utility of multiple sensors for fast detection of a pollutant. The simulations are for a room with source located at the center of the room
Detection Probability: Probability of detecting whether pollutant concentration exceeds a predefined threshold

24. Sensor Placement Problem
Goal: To determine optimal locations of sensors for detection of gaseous pollutants in a room or a large hall
Evaluation measure: Improved probability of detection of gaseous indoor air pollutants
Constraint: Number of sensors to be deployed
We assume a spatial probability distribution for the location of the source.
Approaches:
Uniform placement where sensors are equi-spaced from each other
Closest point placement where sensors are placed at locations closest to the mean of the spatial source distribution
Intelligent placement strategies for quick pollutant detection and reduction of exposure time

25. Sensor placement results with 3 sensors placed in a 9x10x6m room
Simulation Results
Sensor placement results with 3 sensors placed in a 9x10x6m room
Intelligent placement strategy outperforms intuitive strategies such as uniform placement in terms of the detection probability
Approach
Detection probability
Uniform placement
0.35
Closest point placement
0.68
Intelligent placement strategy (Evolutionary algorithm with local search)
0.97

26. Spatio-temporal Profiling of Environ. Parameters
In environmental applications, sensor networks monitor physical variables governed by continuous distributed dynamics
Results in correlated sensor observations
Difficulty: Spatial and temporal irregularities in sampling
Problem: Produce real-time estimates of the values of a distributed field at points where there are no sensors*
* H.Zhang, B.Krogh, J.F. Moura and W.Zhang,”Estimation in virtual sensor-actuator arrays using reduced-order physical models”, 43rd IEEE Conference on Decision and Control, December 14-17,Atlantis, 2004.

27. Our Approach
We propose a supervised local function learning approach to estimate the observation of a sensor from a subset of its neighbors
The goal is to estimate an unknown continuous-valued function in the relationship
y = g(X) + n
where, the random error/noise (n) is zero-mean, X is a d-dimensional vector (e.g., Position coordinates of sensors) and y is a scalar output (e.g., concentration of CO2)

28. Our Approach (cont’d)
A generalized regression neural network (GRNN) has been used, due to its superior interpolation abilities and fast convergence
GRNN learns the spatial concentration function from the observations provided by sensors in the neighborhood (circular region) of a desired location c
: Location of sensors
Illustrative Example
The objective is the real time estimation
of concentration value at point C from the
neighboring sensors

29. Simulation Results for Spatial Profiling of CO2 - 1/2
Estimation of the CO2 levels at each microenvironment from sparsely distributed sensors
One realization of the spatial profile of CO2 and location of sensors

30. Simulation Results for Spatial Profiling of CO2 – 2/2
How many sensors are adequate ?
Summary : About 70 sensors are adequate for MSE of 0.06;
additional sensors do not significantly improve MSE

31. Outcomes
The “right number” of sensors can be determined to reduce cost
Development of multi-criteria, system-wide optimization methodologies for HIYW systems
Multiple, interdependent, individually customized microenvironments controlled by distributed environmental control systems (e.g., PVDs) will be aided by the fine-grained characterization of IAQ parameters
Future Work: Reduced order models for building inter-zonal
transport (Task 3.2) will provide more realistic and computationally
faster models to test our algorithms

32. Experimental CO2 Data - Location of Sensors
In collaboration with Reline Technology, India;
University of Technology, Sydney, Australia

33. Experimental* CO2 Data – Concentration Levels
*More details about this experimental test-bed are provided in subsequent slides

34. Integrated Components of this Task
Improve indoor air quality by
Optimization and control – improved IAQ via new methodologies for real-time control
Indoor sensor network – design, placement,data-processing and spatio-temporal profiling
Test-beds – design of new test-beds and implementation of developed methodologies on these test-beds

35. Sensor Network Testbed in Link Hall
Goal : Data collection and evaluation platform for various control and data-processing algorithms being developed
Wireless sensor network test-bed is being set-up on 3rd floor of Link Hall
Four closed spaces/rooms on the third floor of Link Hall will be monitored by the WSN
Each closed space will have 5 ABLE ARH-T-2-I-W temperature & RH sensors, 5 TI 4GS CO2 sensors,1 pressure sensor and a information processing unit called “Sensor Network Access Point (SNAP)”
In future this test-bed will be incorporated in the building control system for evaluation of the complete system

36. Sensor Network Test-bed in Link Hall (cont’d)
Possible Locations of SNAPs

37. ICUBE Lab (2007)
Goal: Allow people to adjust their own indoor air parameter settings within an office or a cubicle while maximizing overall energy usage
Giving people (users of the lab) what they want for control settings without increasing baseline costs for the larger space
The lab will consist of one test lab (with uniform settings) and one control lab (with individual settings)
The lab will feature intelligent controls, sensor networks
The lab will employ raised floor diffusers with actuated damper controls
The lab will behave like a real building, with people able to adjust their own thermostat settings
Multiple configurations will be possible: classroom to office to lab

38. ICUBE Lab

39. Testbed Ready for Evaluation of New Algorithms
Main Goal: Develop and test multiple-input multiple-output (MIMO) control algorithms for intelligent HVAC control
DAQ system and the actuators have been installed on the HVAC demonstrator (Hampden H-ACD-2-CDL) in Link-0031
System Characterization Experiments (full capability)
Single-Input Single-Output (SISO) Control Experiments (full capability)
MIMO Control Experiments with combination of CO2, TVOC and temperature control (technically ready, awaiting CO2 and TVOC sensors)

40. System Characterization
SISO Temperature Open Loop Response
Illustration of DAQ on the HVAC demonstrator
Enables system characterization so that control algorithms can be developed*
* M.L. Anderson, M.R. Buehner, P.M. Young, D.C. Hittle, C. Anderson, J. Tu, and D. Hodgson, “An Experimental System for Advanced Heating, Ventilating, and Air Conditioning (HVAC) Control”, to appear in Energy and Buildings, 2006

41. SISO Control SISO Temperature Control
Illustration of developed SISO controller on the HVAC demonstrator
Future Work: Develop and test MIMO controllers that will enable us to overcome coupling effects existing in today’s state-of-the-art HVAC systems resulting in improved performance*
* M.L. Anderson, M.R. Buehner, P.M. Young, D.C. Hittle, C. Anderson, J. Tu, and D. Hodgson, “MIMO Robust Control for Heating, Ventilating, and Air Conditioning (HVAC) Systems”, submitted to IEEE Transactions on Control Systems Technology, 2005
Tref1 = 79 ºF
Tref2 = 77 ºF

42. Summary and Future Work
Optimization and control
Algorithms that enable us to move towards HIYW paradigm without compromising comfort and energy costs are being developed
More complex and realistic algorithms will be developed and tested on real test-beds
Indoor sensor network
Algorithms are being developed for improved multi-sensor detection, sensor placement and spatio-temporal profiling
Algorithms will be tested on more realistic models (obtained from Task 3.2) and actual test-beds (Task 5)
Test-beds
Work is in progress on the set-up of sensor network test-beds
Work is in progress on the set-up of HVAC test-beds

43. Conclusion
“…a lot of the work was premature pending definition of a plausible problem scenario…”
We have presented a specific problem scenario involving the shift from OSFA to HIYW paradigm
“…an approach based on temperature or CO2 control might be feasible but should only be considered if the state of the art in indoor environmental controls will be advanced…”
We are focusing on temperature and CO2 control at the micro-environment level (personal level)
“…necessary to refocus this effort…”
Significant refocusing of the effort as detailed in the presentation has been done
“…consult with practicing HVAC engineers, and make inquiries from professionals in the industry…”
We have initiated collaboration with United Technologies Research Center