Virtual reactor faults revealed hot spots wall sensors can miss.
A virtual reactor exposes the internal hot spots that wall sensors can miss, creating training examples without putting a plant through a failure.
Create a virtual stirred reactor, vary cooling, mixing and feed disturbances, and record both internal conditions and realistic wall-sensor readings for future warning and control models.
Edited demonstration briefThe result
Each fault run paired wall readings with the reactor’s internal state.
Fault scenarios pair internal temperature and concentration with realistic wall-sensor readings. These paired examples provide a foundation for future sensing and control models.
A virtual-reactor dataset, not a demonstrated plant-control or early-warning system.
Recorded simulation
Internal temperatures and wall readings show different parts of the fault.
Watch the internal temperature change as the reactor develops a fault.
Recorded outputs, replayed in your browser. No new simulation is running.
Wall sensors cannot measure every hot spot inside a reactor.
The world we created
OpenFOAM models temperature and concentration inside a stirred vessel. Cooling, mixing and feed disturbances run as parallel scenarios, each recording the hidden field and sensor response.
What this enables
The intended model would infer internal conditions from wall sensors, giving a controller a richer picture. The current result is the paired dataset; learned sensing and control remain to be developed and tested.
Fault scenarios varied cooling, mixing and feed conditions.
Change cooling, mixing and feed conditions. Save the internal fields alongside sensor histories.
Paired fields and sensor readings can train virtual sensing.
Use the paired examples to train virtual sensing. This stage remains in development.
Learned warnings and control still need evaluation.
Future models need testing on unseen faults and control decisions replayed in simulation.
Physics has produced the experience for a learning system: limited observations paired with a richer internal state. Training and control evaluation are the next stages.
Original research artifacts
The original files retain the recorded measurements and model details.