Chapter 1 · Lesson 1.4
Choosing a model
There is no universal model. The right choice depends on what you are looking for, the size of the system and the person who will have to read the result.
Three questions for choosing
- How big is the system? Up to a few components, the Markov chain can be built by hand and checked. Beyond about ten interacting components, the number of states becomes unmanageable.
- Do you want an exact result or an estimate? For a regulatory decision or an inspection, an exact result is valuable. To explore a large model, an estimate with a confidence interval is enough.
- Who has to read the model? A fault tree can be read without training. A Markov graph of forty states cannot. A well-structured Petri net reads like a diagram of how the system works.
Practical rules
- Start with the static model: if it is enough, stop there. It also serves as a plausibility check.
- Move to Markov for the subsystems where dynamics matter (pumps, periodic tests), then feed the result back into the tree as a basic event.
- Use a Petri net when there are shared resources, sequences or cascades.
- Reserve simulation for models that are too large or for non-exponential distributions, and keep it to check an exact model on a subsystem.
- Note the assumptions of each model next to the result: without them, a figure cannot be compared with anything.
The loading bay
| Question asked | Model chosen | Reason |
|---|---|---|
| Availability of the two pumps | Three-state Markov | Small system, exact result, figure checkable by hand |
| Periodically tested shutdown function | Markov (or formula validated by Markov) | No repair between tests: the calculation can be done by hand |
| Loading sequence and hidden failures | Petri net | Order of events, “detector failed undetected” state |
| Cascade to the neighbouring bay | Event tree then Petri | Barriers that act or not before propagation |
| Complete model | Global Petri, simulation, sub-parts reduced to Markov | Too many states for a single chain |
Key takeaway
The choice of model is made on the size of the system, the need for an exact or estimated result and readability for the team. Start with the simplest, add complexity where the dynamics matter, and keep a simple model to check the richer one.