Risk assessment of Complex Engineering Systems (CES) is traditionally performed offline using methods like Failure Mode and Effect Analysis (FMEA), event trees, Fault Tree Analysis (FTA) and bow ties, among others. Conventional risk assessment methods are proficient at taking into account the logic of a system. However, there is a need to take advantage of the available multivariate streaming data sources and change the focus from static risk assessment to dynamic risk monitoring to enable timely response and decision making. This emerging field of study has game changing potential, which has only been partially exploited. In this study, we first review the related literature to dynamic risk assessment that have used logic-based models and discuss the used methods and implications. Then, we propose a novel method that integrates logic-based and deep learning models. We explore the capabilities of our method by implementing it on a real-world mining rock crusher system, where deep learning is used along with a Fault Tree to combine modern data based techniques with systems logic models. In the proposed model, the top event probability is updated by continuously updating the basic events probabilities using deep learning models to process the online streaming data collected by the installed sensors throughout the system. The obtained results suggest that the proposed methodology is capable of becoming an effective tool for the application of risk monitoring.