Project Title: Smart Sensor Deployment in Buildings: Evacuation Planning and Energy Management

Principal Investigators: Prof. Krithi Ramamritham (Professor, Dept. of CSE, IITB), Prof. Mark S. Fox (Professor, University of Toronto)

Co-Principal Investigators: Prof. Ronita Bardhan (Asst. Prof, CUSE, IITB), Dr. Arindam Pal (Innovation Lab, Tata Consultancy Services)

Funding Agency: IC-IMPACTS (India-Canada Centre for Innovative Multidisciplinary Partnerships to Accelerate Community Transformation and Sustainability) and Department of Science and Technology, Govt. of India



This project focuses on the design of smarter and greener environments, particularly buildings. Using sensor data combined with rich representations (i.e., ontologies) and smart algorithms, we will address major problems that arise in two kinds of applications:

  1. Smart Building Energy Management, and
  2. Smart Building Evacuation Planning and Dynamic Response System

Smart Building Energy Management:

Sensor-driven building management involves many challenging tasks including reducing and optimizing power consumption, monitoring the health of appliances, maintaining quality of atmosphere and tracking occupants in various parts of the building (useful for purposes such as building safety and emergency evacuation) to name a few. This demands observation of various influencing factors on a continuous/regular basis. We call the factors that affect energy consumption or affect in general building management as “facets of observability”. One simple approach for observing these facets is to place sensors at all locations which need to be monitored. However, installing numerous sensors in different parts of the building can a) be tedious and expensive b) cause inconvenience to the users c) increase the Return on Investment period and, d) affect the aesthetics of the building.

These issues prompted us to ask the question, “What types of sensors are useful for building management and where do we deploy them in a building?”

We exploit the fact that a sensor, suitable for observing a particular facet, may in turn help to infer other facets. This insight can be exploited to reduce the number of physical sensors deployed in a building. However, the locations for sensor deployment have to be selected strategically. We have applied our approach to monitoring various facets in different areas, including a smart classroom complex, in IIT Bombay’s CSE department building. We have shown that impressive reductions are possible in the number of hard sensors needed without compromising on observability.

1) Use “hard” or “physical” sensors, e.g., smart meters to track power consumption, voltage fluctuations, current levels; “entry sensors” equipped with smart card readers; CO2 measuring sensors, etc.

2) Use “soft” sensors or “logical sensors” which depend on inference engines which track phenomenon P1, but can infer P2, albeit with some inaccuracy.

3) Assigning hard sensors to every unit of the building to track the relevant phenomena of interest to that unit will lead to cost and maintenance overheads that will result in longer RoI.

Through this project, we will also study possible approaches to building sensors. Particularly, we explore the types of readily available soft sensors which can be used to infer phenomena that would normally require hard sensors? Here are some soft sensors that have proven effective:

  1. Neural Networks as a model for predicting power.
  2. Soft sensing of current without actually measuring it.
  3. Soft Monitoring of Occupancy.
  4. Soft Monitoring of the number of ON Air Conditioners.

We also examine how a set of hard and soft sensors can be composed to produce a sensor that has a set of specific capabilities. We also determine the functionality and accuracy of the resulting sensor. We explore an incremental approach to hard sensor installation wherein the building manager is not burdened with huge initial investments, and physical sensors can be used to incrementally replace soft sensors as per accuracy demands.

Building a smart evacuation planning system as an exemplar of sensor-based building management:

This part of the project includes all aspects such as design and implementation of new algorithms, deploying sensors to obtain various parameters from the environment and give appropriate messages and alerts through delivery channels.

The following action items will be implemented:

  • Design and implementation of new routing algorithms for pathfinding.
  • Deploy different sensors in the building to collect useful information.
  • Design of a probabilistic behavior model of people in an emergency.
  • Sending appropriate messages and alerts through various delivery channels.

Many studies approach the problem of evacuation planning as a combinatorial optimization problem and some from a system development perspective. There are very few that completely address the evacuation planning problem from both the algorithm and systems perspective. A complete system with efficient and scalable path-finding (routing) algorithms, sensor-based data collection and a probabilistic behavior model to describe people’s behavior is the need of the hour. This project will model the problem mathematically as follows:

  • We represent the building floor plans as graphs.
  • We recognize that people don’t follow the suggested routes during an emergency.
  • A probabilistic behavior model can describe the behavior of people.

Note that the graph representation will be extracted from the ontology-based representation of the building. We can extend such a building evacuation planning system to evacuate entire regions and cities by employing highly efficient and scalable algorithms, along with low-cost and precise sensor technologies.


Developing a Building Ontology Standard that Supports Energy Management and Evacuation:

The objective is to develop a building ontology standard for not only supporting energy management and evacuation planning but to enabling interoperability across the building applications. To achieve this we will:

  • Define the ontology’s requirements in the form of Competency Questions (Gruninger & Fox, 1995).
  • Review existing ontologies and data standards to ascertain which portions satisfy subsets of the competency questions.
  • Integrate existing, relevant ontologies.
  • Extend the existing classes and properties and design new ones to satisfy the Competency Questions.
  • Evaluate how well the new ontology satisfies the two applications’ needs.

Research Outcomes:

The approaches detailed above will be showcased in campus buildings belonging to the investigators. In particular, we will instrument several homes and office buildings to drive this project. The facility which has been developed already as part of preliminary work at IITB will also be available for this project; We have instrumented a complete building (CSE department’s current building). We have deployed sensors to gather electricity usage data at various spatial granularities. We will similarly instrument two office buildings as well as several homes. This will allow us to gather detailed usage data from both homes and office environments to drive our research on peak usage capping, demand response, and emergency evacuation.

The end product of this project will be a family of algorithms, ontologies, and prototypes that demonstrate the key benefits of our methods. Research papers presenting our results will be published at premier conferences and journals. Sensor datasets from our deployments and source code from our prototypes will be released to researchers for further experimentation.

Unique and Innovative Aspects of the Project:

We have several new ideas for the project. Some of them are:

  • Design and implementation of efficient, scalable and (near-) optimal algorithms.
  • Design of a probabilistic behavior model to simulate people’s behavior.
  • Use of sensors to discover the threats (e.g. fire) and the presence of people in dangerous areas and to give messages and alerts.
  • Design and implementation of a system to plan and execute all aspects of evacuation from buildings and cities.
  • New ontologies for building, energy and sensor integration.