The vastly increasing availability of data types and quantities, emerging sensing technologies, artificial intelligence, and machine learning models are enabling a digitally enabled view of infrastructure systems. In addition, the use of interconnected sensors and data mining techniques form pivotal blocks for effectuating the establishment of smart buildings and smart cities. However, the major limitation of using “black-box” or purely data-driven models is the strong dependence of sensor data. The quality of sensor data, representativeness of sampled data, choice of data for training are factors that highly influence the effectiveness of learned models. Moreover, the learned models feature little physical understanding of the underlying system or process. In this talk, the speaker will discuss how to leverage the domain knowledge to enhance the effectiveness of machine learning models, hence to convert sensor data into explainable information, in the form of physically interpretable and generalizable models, in the service of infrastructure integrity assessment, maintenance, and further decision-making. The talk will cover: (1) a newly proposed parametric modeling technique based on sparse regularization constrained by underlying physics, termed sparse structural system identification; (2) linear/nonlinear structural damage characterization using sparse structural system identification; (3) full-field structural monitoring and vibration analysis using dynamic vision sensors (event-based cameras) and physics-guided data science. Laboratory experiments and real-world case studies are provided for demonstrating the proposed frameworks.