As part of my final engineering project, I have been working on a digital twin of an automated production line using Siemens NX Mechatronics Concept Designer (MCD) integrated with PLCSim Advanced and TIA Portal.

While the model itself is functional and the PLC logic performs as intended, I encountered significant issues during real-time simulation. Specifically, after 2–3 minutes of running the simulation, the system begins to slow down, behave inconsistently, or produce incorrect mechanical behaviour, such as components misaligning or moving at unintended speeds.

Root Cause: Limitations of NX MCD's Real-Time Physics Engine

After analysing system performance, CPU usage, and simulation behaviour, it became clear that hardware limitations do not cause the problem. 

Despite this, Siemens NX MCD heavily relies on single-threaded performance during simulation. The physics engine, motion constraints, and PLC communication all run mostly on one CPU core, which quickly becomes saturated in complex models.

Additionally, my simulation involves:

• A large number of components and mechanical joints

• Real-time collision detection

• High-frequency data exchange with the virtual PLC

This combination overwhelms MCD’s real-time engine, which is not designed for large-scale systems with full physics and high update rates.

What I’ve Tried (and Why It Didn’t Fully Help)

I experimented with:

• Lowering visual quality and disabling shadows

• Simplifying geometry

• Increasing simulation step size

• Reducing collision checks

Although these optimizations improved performance slightly, they did not prevent the simulation from gradually desynchronizing or slowing down. This confirmed that the core issue lies in the software’s architecture, not my hardware or model complexity alone.

Conclusion & Lessons Learned

This experience highlighted a real-world constraint when working with digital twin technologies. While NX MCD is excellent for visualising and testing concepts, it has scalability limitations for high-detail, real-time simulations integrated with PLC systems.

To overcome this in future iterations, I may:

• Break the simulation into smaller modules

• Use event-based logic instead of complete physics

• Consider hybrid simulation platforms (e.g. SIMIT, Unity with OPC UA)

It’s a valuable reminder that even powerful tools have boundaries, and part of engineering is learning how to work around them effectively.