Over the life cycle of large civil infrastructure, there is often significant degradation in capability and design performance due to extend usage and, in many cases, deferred maintenance. The ultimate limit states in this case can result in unexpected failure, with potentially large economic or life-safety consequences. Structural health monitoring (SHM) is a framework for monitoring the structure over its life cycle, and this field has expanded rapidly in the past two to three decades due to the urgency of infrastructure aging beyond its design life. One of the critical challenges in a monitoring process is obtaining the most valuable information from the structure responses in the field under a continuous monitoring paradigm. This dissertation will present a new optimal sensor design framework, using Machine Learning (ML) techniques, including Gaussian Process Regression (GPR), Bayesian inference, uncertainty quantification, and Bayesian optimization, that guides risk-informed SHM system design. In particular, this dissertation: (1) demonstrates a framework for optimal sensor design using Bayes risk as the objective function; (2) further explores the framework using risk-weighted f-divergence functions and implements it in a real miter gate problem as a case study; (3) investigates the effects of sensor reliability over a life cycle monitoring of the miter gate for informing optimal sensor arrangement.