Due to the aging of civil infrastructure and the associated economic impact, there is an increasing need to continuously monitor their structural and non-structural components for system life-cycle management, including maintenance prioritization. Many times, this monitoring process involves different types of data sources collected at different time scales and resolutions, such as abstracted rating data from human inspections, historical failure record data, uncertain cost data, high-fidelity physics-based simulation data, and online high-resolution structural health monitoring (SHM) data. The heterogeneity of the data sources poses challenges to the diagnostic/prognostic implementation of decision-making for maintenance. Using components of a miter gate as the exemplary case study, this dissertation presents a condition-based maintenance framework to demonstrate how to integrate various data sources using Bayesian and Machine Learning (ML) methods for effective SHM, and Prognostics and Health Management (PHM). In particular, this thesis focuses on the different pieces of the proposed framework, which are (1) surrogate based model updating for damage diagnosis; (2) integration of abstracted data and SHM for damage prognosis; (3) diagnostics of multiple forms of damage in miter gates; and (4) validation of physics-based diagnostic models using hybrid structural testing.