Data Collection and Communication

1
Data Collection and Communication

• Stephen Potter, University of Edinburgh

2
Overview

• At its simplest the FireGrid vision involves
• Collecting data from sensors in the built
  environment
• Interpreting this data to assist end-users
• Presenting this interpretation to the users.
• (Here we focus on fire-fighters as the
  end-users.)
• This presentation describes the manner in which
  this data is converted into fire-fighting
  intelligence.

3
Data collection

• Data are collected from a variety of sources
• Principal sources are conventional sensors
  (temperature, heat flows, gas concentrations,
  etc).
• But more generally these sources might usefully
  include air conditioning units, CCTV, sprinkler
  systems, computer systems, mobile phones, etc.
• Data generally correspond to sets of point
  facts about the environment at specific times.
• These dynamic data are transmitted over the
  internet to a system database (hosted on a remote
  computer).
• Database also contains static data which describe
  each type of sensor, its location, measurement
  unit, etc.
• However, the quality of each data point must
  first be assessed
• Sensors can fail (sometimes temporarily) and
  will fail, during serious incidents!
• This assessment can be difficult since an
  unexpected reading could indicate a failed sensor
  or a significant event
• Developed approach is to use known operating
  ranges, along with internal consistency checking
  to grade the plausibility of data points.

4
Data and meaning

• Database contains a large number of data points
  describing different things at different times.
• At this point, the data as a whole will not mean
  much to a fire-fighter.
• Data must be interpreted to provide useful
  information.
• Two phases of interpretation in FireGrid
• Model (objective) interpretation Computational
  models of fire and related phenomena interpret
  the data in terms of current and projected
  conditions.
• User (subjective) interpretation Artificial
  Intelligence techniques interpret (and aggregate)
  the various model interpretations in terms of the
  needs of the end user(s).

5
FireGrid user requirements

• Tactical decision-making involves choosing the
  appropriate coordinated approach that will
  achieve objectives.
• Generally this is the lowest level of
  decision-making at which there is scope for
  incorporating computer-based decision-support
  tools.
• Effective tactical decision-making during fire
  emergencies requires accurate and relevant
  information.
• For many incidents the available information is
  partial and approximate, at best.
• FireGrid aims to improve this situation.
• In the UK the primary tactical decision-maker at
  fire incidents is the designated Fire Incident
  Commander (FIC), who is
• The most senior officer among the first
  responders, responsible for command and control,
  deployment of resources and tactical planning.
• (Since the FIC has to manage multiple lines of
  communication, an officer supporting the FIC is a
  more likely target user.)

6
FireGrid user requirements

• Having arrived at a fire, a major decision that
  the FIC has to make is whether to send
  fire-fighters into the building
• Fire-fighters enter the building if the chance of
  saving trapped people outweighs the risk to the
  fire-fighters (this is called adopting offensive
  tactics).
• Otherwise, the fire will be tackled from outside
  until conditions change or the fire is
  extinguished (defensive tactics, the default).
• This weighing of risk against benefit is called
  the dynamic risk assessment process
• Process must be performed rapidly under intense
  pressure.
• Relies on the experience and expertise of the
  FIC
• but the appropriateness of the decision reached
  will ultimately depend on the range and quality
  of information that is currently available to the
  FIC.
• FireGrid aims to support this process by
  providing more and better information about
  current and potential hazards.
• However the process has implications for
  information presentation
• Must provide an immediate and accurate overview
  of the incident, while also providing detailed
  rationale to support decisions.

7
Fire-fighting hazards

• Based on a suggestion from a serving fire
  officer, a traffic light display for indicating
  the level of a hazard has been adopted
• green the system is unaware of any hazard to
  fire-fighters
• amber additional control measures may be
  needed
• red dangerous for fire-fighters
• Hazards arise due to the conditions that hold and
  events that occur at particular locations and
  times
• The task of Model Interpretation is to infer
  these conditions/events using the available
  sensor data.
• The task of User Interpretation is to decide on
  the nature and level of hazard which these
  conditions/events imply for the fire-fighter in
  the context of the current incident.

8
The use of fire models

• Existing simulation code is generally based on
  models that have been developed by academics.
• Their goals are to build accurate models of fire
  and related phenomena
• but not necessarily models that can be used in
  the context of fire-fighting operations.
• One difficulty is that the models will usually
  not have been developed to operate with dynamic
  data
• Addressing this requires innovative approaches
  for producing and revising interpretations in
  real time.
• Another difficulty is that these models can be
  computationally expensive (i.e., slow).
• A problem addressed by the use of High
  Performance Computing resources accessed over the
  Grid

9
HPC and Grid Computing?

• High Performance Computing is
• The use of supercomputers or (more usually)
  computer clusters to provide greater
  computational power for running complex code.
• Since these are valuable/expensive resources,
  they tend to hosted at dedicated sites, with
  (remote) user access to the resource strictly
  controlled.
• FireGrid introduces specific requirements we
  need urgent access to resources.
• Grid Computing is
• Conceptually, an approach to computation in which
  computational resources and data are distributed
  and their use is dynamic.
• In practical terms, a uniform computational
  infrastructure (the Grid) that facilitates
  invocation of remote resources and transfer of
  data, along with security, authorization of
  users, etc.
• In FireGrid, the Grid allows us to run code on
  HPC resources, to provide input data and to
  retrieve output data.

10
User interpretation

• Deployed models interpret sensor data to generate
  (possibly a very large amount of) information
  about conditions and events, places and times
• and potentially, once the model is started, it
  will continue to generate information
  periodically during an incident.
• Role of User Interpretation is to interpret the
  nature and level of hazards to the fire-fighter
  implied by this information.
• This involves continuously applying a two-stage
  cycle of reasoning
• Belief revision as new information is received,
  the set of beliefs (that is, information that
  is considered to be true) is revised to maintain
  a internally consistent and coherent description
  of the incident.
• Hazard identification fire-fighting knowledge is
  applied to the current set of beliefs to identify
  hazards.

11
User interpretation belief revision

• Information from models consists of facts about
  conditions or events at certain times and
  locations.
• However, these facts may be uncertain or else
  inconsistent with the output of other models or
  with what is already believed about the incident.
• The purpose of belief revision is to maintain the
  consistency of the description of the physical
  conditions
• This requires some knowledge of what is meant by
  the consistency of the description of physical
  conditions
• as well as some knowledge of the fire models and
  the nature of the information they produce.
• This becomes especially difficult when dealing
  with temporal information
• Involves adopting and modifying beliefs about the
  durations for which specific conditions hold and
  times when events occur.

12
User interpretation hazard identification

• With a consistent set of beliefs, the second
  stage is to identify hazards.
• This is done using a set of hazard rules
• IF maximum-temperature 100C in location X from
  time TTHEN hazard-levelamber in location X
  from time T
• Rules applied by searching current set of beliefs
  for any that match the rule conditions.
• The matching beliefs are said to justify the
  conclusion that the hazard exists
• but will need to revise the set of identified
  hazards whenever the set of beliefs changes.
• Rules would be supplied by fire-fighting experts
  and could be specific to particular
  buildings/incidents
• e.g., if it is known that fuel is stored in a
  particular location

13
User interpretation hazard display

• Traffic light for each location
• Lower current overall hazard level
• Upper future overall hazard level
• Pop-up window gives specific hazards for a
  location displayed along a time-line.
• Text provides justifications for these hazards
• Also advice for handling uncommon hazards.
• Interface developed with assistance and feedback
  from serving fire officers.