Multi-modal inference: How to add value to infrastructure monitoring data
Dr. Rodrigo Sarlo, Virginia Tech
Hosted by: Soto Structures Lab (www.sotostructures.com)
The growing popularity of concepts like Smart Cities, Internet-of-things, and Digital Twins means we are collecting more sensor data on infrastructure than ever before. However, raw data alone has little value, what is more important is the information it provides and the decisions that it enables. Certain types of sensing modalities (e.g. accelerometers, cameras, and LIDAR) collect high resolution data that can yield a variety of rich, and sometimes unexpected, information. Through targeted signal processing and analysis, it is possible to infer high-level information for a variety of civil infrastructure applications. “Inference” is the key word, since much of the information contained in physical data is not readily obvious from observation alone. Our lab is multi-modal inference techniques which produce high-level, actionable data, including: 1) Using LIDAR point clouds to infer a building’s structural design, 2) Using traffic cameras to infer structural damage in traffic light masts, 3) Using vibrations to infer building occupancy patterns for operation management. Each of these examples demonstrate one of three main advantages of multi-modal inference techniques. Firstly, they value to certain data acquisition techniques by making them more versatile. Secondly, they are useful for “recycling” existing infrastructure data for a new purpose. Finally, they can be used to design infrastructure hardware which adapts to future needs.