Physics-guided Information Acquisition and Learning For Urban Infrastructure Systems With Constrained Sensing Capabilities

Urban infrastructure monitoring is essential to ensure the safe and efficient functioning of urban services. Recently, a lot of advanced sensing systems have been developed
to improve the management and maintenance of urban infrastructure systems. Timely and accurate information acquisition and learning for urban infrastructures, such as structural health, traffic conditions, surrounding air quality, etc., is the main goal of these urban sensing systems.
To learn the infrastructure information from these large set of sensing data, many conventional data-driven approaches have been introduced using statistical models or
machine learning techniques. However, urban infrastructure monitoring systems often have constrained sensing capabilities due to improper deployment conditions, budget
limits, or the complexity of physical infrastructure systems. The constrained sensing capabilities include 1) noisy data from complex physical systems, 2) lack of labeled
data limiting the accuracy of data-driven models, 3) inefficient sensor deployment with low sensing coverage, 4) a lack of proper sensors to be deployed in the target infrastructures to collect the target information from the infrastructures. These constrained sensing capabilities reduce the quality of information acquisition, which may significantly degrade the performance of information learning using conventional data-driven methods. To address these challenges, my research objective is to utilize physical knowledge to acquire and learn high-fidelity urban infrastructure information. The physical interactions
between different physical components of infrastructure systems, between the physical components of infrastructures and sensors, and between the physical components of infrastructures and the ambient environment, are always governed by the physical properties of the urban infrastructure systems and thus subjected to consistent physical principles. Such underlying physical knowledge provides an important aspect to better understand the infrastructure systems and thus improve information
acquisition and learning under sensing constraints. Specifically, my research focuses on the following four topics:
a) To address the challenge of noisy data in complex physical systems, I introduce an information-theoretic approach to extract the changes of wave propagation
patterns between different physical components of infrastructures.
b) To address the challenge of lack of labeled data, I propose a new knowledge transfer scheme across different infrastructures based on the physical understanding
of how sensing data changes with different infrastructures’ physical properties.
c) To address the challenge of inefficient deployment of sensors, I introduce an incentivizing algorithm to optimize the sensing distribution considering vehicle
mobility and human mobility.
d) To address the challenge of lack of proper sensors, I introduce an indirect sensing method to monitor the target infrastructures using ambient infrastructure sensing
systems.
The physical knowledge contains the understanding of the physical patterns that urban infrastructure systems subject to. These physical patterns are governed by universal
and classical laws of physics. The physical patterns are validated to reflect the underlying physical processes happened inside urban infrastructure systems in previous
studies, but difficult to be learned directly from the less informative data collected under constrained sensing capabilities. Under sensing capabilities constraints,
the combination of physical knowledge helps to effectively acquire high-fidelity sensing data and learn information from the collected sensing data for urban infrastructure systems.