Drive-by Health Monitoring (DBHM) is a relatively new mobile health monitoring strategy that employs vehicle mounted sensors to monitor the health of bridge systems in an efficient and economical manner. Before DBHM can be realized as a viable health monitoring strategy, however, an approach for managing environmental and operational noise needs to be developed. In traditional health monitoring, machine learning techniques, such as neural networks, have been shown to reduce the effect environmental and operational noise has on damage detection accuracy; though, these methods typically require training on damage data, which can be difficult if not impossible to obtain for healthy structures. To resolve this issue, the authors proposed a methodology that utilizes a neural network architecture trained on realistic vehicle-bridge simulations to detect damage in physical highway bridges. For a simulation trained neural network to be able to detect physical bridge damage, numerical models must be able to accurately represent the behavior of a system when damaged. Therefore, the motivation of this work is to identify and validate physics-based techniques for modeling damage induced fluctuations in the dynamic response of highway bridge structures. This study focuses on one of the most common types of bridge damage, frozen support bearings. The authors introduce methods for modeling frozen bearing damage, and discuss the variety of variables that must be considered under certain environmental and operating conditions. The study concludes with demonstrating how to generally apply frozen bearing damage in healthy bridge models to represent possible future damage states.