Saitej Velaga
Research Mentor(s): Lars Junghans
Mentor Department: Architecture
Authors: Lars Junghans, Saitej Velaga
Session: Session 2 (10:00am – 10:50am)
Presentation Type: Poster 107
Abstract
Heating and cooling costs burden the environment and the running costs of a building. This has caused a great deal of innovation in this space to reduce these issues. The approach we are proposing is the use of advanced materials for regulating humidity, along with multiple sensors and motors to dynamically and predictively control ventilation, avoiding the use of traditional HVAC methods. Utilizing a self-hosted network of sensors, sensors will be able to talk to each other wirelessly, allowing for the transmission of both sensor information, such as temperature, and commands, such as opening a window. This will allow for multiple sensors in each room of a building, enabling greater information sharing and fine-tuning of optimal control modes at a particular time, as well as plug-and-play operation. In order to control ventilation once the data is gathered, we use a neural net to predict the conditions of rooms, such as its construction and how much sunlight they get. This is done through large-scale data generation using simulations and training the Convolutional Neural Net to pick up on patterns in data generated by the sensors. This will allow the predictive control algorithm to optimally open vents to heat and cool the rooms.
