Refinery operations involve hundreds of equipment and complex material flows through various processing units. Thousands of sensors monitor the process for any deviations from the set points leading to a flood of alarms that overload the operator in an emerging situation. Refinery processes are highly controlled where the deviations in controlled variables of a process are rare, but drifts are observed in manipulated variables as the control system automatically addresses the drifts. Digital twins of the processes offer additional insights, compared to the sensor data, into the processes in greater detail and can be used as predictive models and soft sensors to estimate impending deviations. However, to mitigate the deviations, a refinery operator or a process engineer uses knowledge from various sources in addition to the trends observed in monitored variables. Process knowledge scattered across diverse sources such as incident reports, root cause analysis documents, and standard operating procedures need to be reconciled to take a decision. In this chapter, we show how a formal representation of refinery process knowledge can be combined with digital twins and sensor data to derive more value from digital twins for process monitoring applications. We illustrate this in the case of fluid catalytic cracking and fractionation units of a refinery. We introduce a semantics-based platform called TCS PREMAP, where process ontology is used to express failure knowledge, orchestrate digital twins in the changing context of the refinery, and connect to real-time data to carry out root cause analysis.