TEA-bot

摘要

We propose TEA-bot (Thermography Enabled Autonomous Robot) to address the problem of labor-intensive manual inspections of thermal leaks in Heating Ventilation and Air Conditioning (HVAC) systems while avoiding installation of sensor networks, which can be challenging and expensive for existing buildings. TEA-bot is an Unmanned Ground Vehicle (UGV) designed to navigate in ceilings using visual-based Simultaneously Localization And Mapping (SLAM) while detecting thermal leaks from HVAC systems using Convolutional Neural Networks (CNN). TEA-bot uses inexpensive 3D printed parts as the bones, a single-board computer (SBC) as the brain, an RGB-D camera as the eyes, and a thermal camera as the leak detector. We build an in-lab ceiling environment as a true-to-size testing site with five types of common leaks in HVAC systems (e.g., improper connection) to test the feasibility and effectiveness of TEA-bot. Results show that TEA-bot can 1) generate high-resolution ceiling maps in the format of point clouds for existing buildings without Building Information Models (BIMs), with an average error of 1.56 inches in duct length measurements; 2) identify leak types with a precision of 86.60% using the thermal camera images; and 3) register leaks onto the point cloud maps, providing a holistic and intuitive view of leak locations in an unknown ceiling environment.