Performance anomaly detection and prevention in complex industrial systems, involving many distributed embedded computing nodes, is an ever-present challenge. Understanding such complex systems using purely analytical, or experimental techniques is neither sufficient, nor cost efficient. Efficient high-level models that allow the study of current system behaviour, predict performance trends and assess possible optimisation measures, are in demand. This paper presents an approach to automatically infer such high-level system models by using system-level tracing and to represent the current state of the system using Discrete Event Simulation (DES) techniques. The current behaviour pattern of the system is followed by replaying observations, while behavioural potentials under alternative sets of circumstances are considered by exploring what-if questions. Our results are applicable to anomaly detection and anomaly prevention solutions for complex systems. The main approach in keeping our workflow as efficient and as compact as possible is the use of a communication-centric modelling approach for complex embedded systems. As our use-case, we discuss the main challenges of modelling software processes and resource utilisation in semiconductor photolithography machines produced by ASML and how such complex systems can be mimicked with high-level event-based simulation. An automatically inferred OMNEST simulation model is presented and the different steps that were taken to simulate the behaviour of this production grade system are described. Initial evaluations show the maximum difference in application process lifetimes between real and simulated executions is less than 1%.