Due to the observability limitations imposed by large-scale systems, it becomes vital to introduce methods to understand the state of the system and its components to aid in their maintainability. This study presents a Bayesian Network based approach to mapping out the system and component states of complex systems. Specifically, the Holdup Tank case study is considered to represent the case in which there is a dynamic dependence between system components. Moreover, in complex systems, learning the parameters and structure of the model can be extremely costly. This study incorporates the use of constraint graphs to reduce the variables in the model and improve the learning accuracy of the parameters as well as structure of the Bayesian Network Model. Accounting for the constraint graph allows decision makers to make more accurate inferences on the system and component states with less data. The results show an improvement to the learned model given a relatively small dataset.