1
   Lighting and Medical Personalization
Optimizing Efficiency and Customer Satisfaction

• Alice M. Agogino
• Roscoe and Elizabeth Hughes Professor of
  Mechanical Engineering

2
Researchers

• Professor Alice Agogino, Faculty Advisor
• Marisela Avalos, MS/PhD student
• Matt Dubberley, MS student, Fall 2003
• Jessica Granderson, PhD student
• Mary Haile, undergraduate student
• Johnnie Kim, undergraduate student
• Catherine Newman, MS/PhD student
• Jaspal Sandhu, PhD student
• Yao-Jung Wen, MS/PhD student
• Rebekah Yozell-Epstein, MS student, Spring 2003

3
Motivation

• Indoor Environmental Quality Lighting
• Increased productivity
• Increased quality of experience
• Energy Efficiency
• Increased importance world-wide
• Impact on pollution, global warming, expense

4
Motivation Commercial Lighting

• Electrical Consumption and Savings Potential
• 2/3 of electricity generated in US is for
  buildings
• Lighting consumes 40 of the electricity used
  in buildings
• Advanced Commercial Control Technologies
• Up to 45 energy savings possible with occupant
  and light sensors
• Limited adoption in commercial building sector

5
Background Commercial Lighting

• Problems With Advanced Control Technologies
• Simple control algorithms dim the lights in
  direct proportional response to the sensor
  signal uncertainty is not considered --gt sensor
  signals, estimation/maintenance of desktop
  illuminance
• Time of day/week is not considered, lost savings
  through demand reduction
• All occupants are treated the same in spite of
  the vast differences in perception and preference
  that exist between individuals

6
Background Commercial Lighting

• Problems With Advanced Control Technologies
• Systems are hard-wired into the line electricity
  of the building making retrofitting expensive and
  prohibitive
• Required calibration and installation expertise
  make successful commissioning difficult
• Algorithms can result in annoyance to the users
  through inappropriate switching or speed of
  switching

7
Research Goals

• Increased energy savings through the
  implementation of demand-responsive
  decision-making
• Increased user satisfaction with the system
• Personalized, improved decision-making balancing
  conflicting preferences/ perceptions among
  individuals sharing a common light source/switch
• Improved maintenance of target illuminance at the
  worksurface
• Increased user satisfaction with the system

8
Benchmark Occupancy
9
Potential Energy Savings for Office Building
10
User Studies

• Survey Feedback
• 54 want a dimly lit view of rest of room
• 59 require slightly different light levels
  throughout the day (desk lamp)
• 32 want automatic overhead lights with override
  and manual task lamps
• 77 like same lighting throughout the day
• 73 want to rely on default settings at first and
  then enter preferences later

11
Life Cycle Assessment of the Intelligent Lighting
System using the Distributed Mote Network

• MS Project, Matt Dubberly
• Goals
• To evaluate the environmental impacts associated
  with implementing the proposed Intelligent
  Lighting
• Compare the electricity saving benefits of the
  Intelligent Lighting System to the environmental
  burdens associated with implementing the system.
• Provide insight for design choices, such as what
  type of battery should be used or which materials
  and components should be minimized

12

• The negative environmental impacts of the
  proposed Intelligent Lighting System range from
  17 to 344 times smaller than that of conventional
  lighting systems for the different environmental
  impact categories.
• The components that contribute the most to the
  system impact are
• Mote printed circuit board
• Mote integrated circuit
• Lithium battery
• Ballast housing paint
• The silicon steel and copper in the ballast
  transformer and inductor

13
Intelligent Decision-Making and Smart Dust Motes
Granderson

• An intelligent decision algorithm allows
• Validation fusion of sensor signals
• Differences in user preferences and perceptions
• Peak load reduction/demand responsiveness
• Influence diagrams allow
• Real-time decision-making and control
• Uncertainty in knowledge (sensor values and
  non-deterministic relationships)
• Ability to represent complex interdependencies
• Rules for combining evidence, based on rigorous
  probability theory or fuzzy logic

14
Intelligent Decision-Making and Smart Dust Motes

• Smart dust motes potentially offer
• wireless sensing at the work surface, increased
  sensing density, simpler retro-fitting and
  commissioning, wireless actuation, and an
  increased number of control points

15
Intelligent Framework Modeling the Decision Space

• Initially models demand-responsive and
  personalization aspects of the problem.
• Variables Included
• Day, time, electricity price, workstation
  occupancy, sensed workstation occupancy,
  actuation decision, task type, resulting
  illuminance (following actuation), resulting
  perception of the occupant
• Constants Included
• Preferred ideal illuminance, min/maximum
  actuation, ideal reward, vacancy penalty/reward

16
Regional Decision Space with Local/Individual
Factors
           
17
(No Transcript)
18
Intelligent Framework Modeling the Decision Space

• After developing the personalized,
  demand-responsive decision model, daylighting
  factors were incorporated
• Variables added
• Month, weather (cloudy), latitude, solar azimuth
  and altitude, room geometry, sensed and true
  solar contribution to to the region, solar
  contribution to the ith worksurface

19
(No Transcript)
20
Empirical Preference Testing

• Purpose to identify the illuminance ranges over
  which occupants find the lighting to be ideal,
  too dark, and too bright at their personal
  workstations
• This gives us
• conditional probabilities required for the
  decision model
• information to use in the value function

21
Empirical Preference Testing

• Results
• Probabilistic conditional preference data of the
  form P(IlluminancePerception), P(Perception),
  that can be used in the personalized,
  preference-balancing control model

22
Preference Testing - Results

• Paper-based tasks required significantly more
  light than computer-based tasks

23
Preference Testing - Results

• No illuminance range proved to be ideal for all
  four occupants, even though all share the same
  switch (computer histogram)

24
Preference Testing - Results

• No illuminance range proved to be ideal for all
  four occupants, even though all share the same
  switch (paper histogram)

25
Preference-balancing Value Function

• Goal is to create a function that
• heavily favors meeting the ideal illuminances of
  those present
• heavily favors turning the lights off/min in the
  absence of occupants
• heavily penalizes turning the lights on/max in
  the absence of occupants
• assigns a value of difference-from-ideal for each
  occupant present, and each possible actuation
  decision

26
Future Research Evaluation of Research Goals

• Evaluation of preference-balancing value function
  computer simulation
• Evaluation of target illuminance maintenance
  hardware simulation
• Evaluation of energy savings achieved with
  demand-responsiveness computer simulation
• Evaluation of user satisfaction w/ the system -
  implementation in a daylighted test space,
  complimented with user surveys

27
Validation of Motes and Network

• Construct test architectures for mote sensor
  networks in target office spaces
• Characterize the motes signals and failure
  patterns
• Develop appropriate validation and fusion
  algorithms
• Calibration on mote sensors
• Evaluate fuzzy probabilistic fusion algorithm
  on sensor networks

28
Illuminance Calibration

• Hardware/Experimental Set-up
• Light sources
• Fluorescent room light
• Incandescent desk lamp (75W bulb) .
• Halogen floor lamp.
• Minolta T-10 illuminance meter

29
Illuminance Calibration
30
Illuminance Calibration
31
Illuminance Calibration
32
Probability Distribution of the Mapping Curve

• Illustration of probability distribution when
  mapped readings are around 500 lux

33
Temperature Calibration
34
Temperature Calibration
Thermometer on basic sensor board
Thermometer on MICA sensor board
35
Fuzzy Validation and Fusion on BESTnet v1.0

• Real-time Fuzzy Sensor Validation And Fusion
  (FUSVAF) algorithm

36
(No Transcript)
37
(No Transcript)
38
Feasibility of Using Accelerometer as Occupancy
Sensor

• Hardware setup

39
Motes as Decentralized Autonomous Agents Sandhu

• Agents with collective intelligence may be more
  efficient than centralized control.
• Model the motes as a collection of intelligent
  agents that share the same global utility
  function.
• Agents communicate on wireless network to
  maximize their local and gobal utilities.

40
Agents with Collective Intelligence have Been
Successful in other Domains
MINI-ROBOT RESEARCH Sandia National
Laboratories (Photo by Randy Montoya)
Entertainment computing
Large groups of small vehicles
41
Lighting Mote Collectives

• z worldline - action/state vector of agents and
  environment (sensors actuators)
• ? agent, ? other agents
• z? , z?
• The key is finding good utility functions
• G(z) global utility that balances energy and
  performance multiobjective function.
• g?(z) private utility that might take on the
  preferences on different room occupants.

42
Medical Home SecurityMarisela Avalos

• High density wireless motes could detect changes
  in patient patterns in a manner that is less
  intrusive than other devices such as cameras or
  pressure sensors on toilets.
• Such networks could be useful for other security
  concerns
• Intruders
• Fire or extreme temperatures
• Extend network for self-reporting of injuries

43
Personalization in Medical Care - Avalos, Newman,
Ng, Rahmani Sandhu
A sense of community beyond that contained
within the walls of a long-term care residence is
important to improving the quality of life of the
confined elder. Without a community presence
relieving the isolation, the culture of illness
and debilitation overtakes a culture of
living. Susan E. Mazer, President of Healing
HealthCare Systems
Personal mote on keychain
THE CONCEPT CommuniCast is an electronic
broadcast display, or bulletin board, that
dynamically posts events, activities, and other
pertinent information in the presence of a
wireless device and based on the users display
preferences. The goal is to improve the level of
communication and social interaction among the
senior citizen community.
44
MS Project Proposal -Newman

• To construct a complete product prototype
  integrating
• Develop system for assigning preferences to
  announcements that takes privacy
• considerations into account.
• HCI Considerations
• Display Readability and Comprehension
• Unobtrusive Wearable Motes
• Customer System Preferences
• Ethnography Study
• Manufacturing Issues
• Expect prototype to be designed for
  manufacturability

45
System Architecture
Administrative Office
Upon finding a user in range, display device uses
user identification to request appropriate
information to display for that user. This
information is then displayed to the end-user.
Display device (WLAN- mote- enabled)
Web interface
Web interface connects to server application in
order to view and update information in the
database.
Display device requests are forwarded to server
application.
server application
WLAN
database
Server application queries and modifies database
based on incoming requests from local Web
interface or remote display devices.
Mote-based keychain
Mote-enabled display scans for mote-based
keychains containing user identification over RF.
Slide care of Sir Jaspal Sandhu
46
Evaluation Research Goal 1

• Computer simulation (McGrath) of personalized,
  preference-balancing decision-making
• Model a room w/o windows in order to control the
  simulation, restrict attention to preference
  only. The room should contain multiple users
  sharing one switch, or bank of lights.
• Specify various occupancy patterns in the space,
  and backtrack from preference testing data to
  determine how many ideal perceptions, too dark
  and too bright perceptions are registered under
  the two competing control algorithms.
• Preference data was obtained though an
  experimental hardware simulation in this case we
  are computer-simulating these preferences for the
  group of occupying the space

47
Evaluation Research Goal 2

• Experimental simulation (McGrath) of improved
  target illuminance levels
• The goal is to quantify how the intelligent
  controller compares to a commercial controller in
  maintaining target illuminance at the worksurface
• In order to do so, test the commercial and
  intelligent systems under the same external
  (room) conditions, perturbing/varying the
  worksurface illuminance.
• Each system will have a target illuminance that
  it is trying to maintain. Therefore, by recording
  the deviation from target for each perturbation,
  a comparison of the two systems is possible.

48
Evaluation Research Goal 3

• Computer simulation (McGrath) of increased E
  savings through demand-responsiveness
• Model a space without natural light, in order to
  provide the most conservative estimate, and to
  control the simulation.
• Condition the space throughout one 24hr. weekday,
  assuming a typical 8hr workday. To provide an
  upper bound on savings, simulate a second case,
  in which all bodies are present throughout the
  entire 24hr. day.
• Commercial algorithms will set one target
  illuminance for that whole period, while the
  intelligent algorithm will set varying targets
  depending upon the price schedule. The
  illuminance will determine the luminance (output)
  of the lights, from which we can calculate energy
  consumption, and cost.

49
Evaluation Research Goal 4

• Laboratory experiment, sample survey (McGrath) to
  evaluate user satisfaction
• Select a test space with significant amounts of
  natural light throughout the day. Install a
  commercial daylighting system, run it for a week
  under variable cloud conditions and issue a
  survey.
• Install the intelligent daylighting system, run
  it for a week, under the same conditions, and
  issue a second survey. The surveys are to be
  complimented with bottom-line data number of
  manual overrides, electricity consumption and
  expense, and worksurface illuminance patterns.
• Candidate test spaces include the BID office
  space (full-scale test), and LBNLs
  electrochromic windows test-bed (controlled
  prototype test).