Chapter 1 · Lesson 1.3
Overview of dynamic models
Four main families of models deal with time, order and dependencies. They complement each other more than they exclude each other.
Dynamic fault trees
They keep the tree reading and add gates that express dynamics (Dugan and his colleagues proposed them in the early 1990s):
- PAND (priority AND): true only if the inputs fail in a given order;
- SPARE: a spare part (cold, warm or hot) takes over from the main part;
- FDEP (functional dependency): the failure of a trigger element causes the failure of other elements.
They are readable for an engineer used to trees, but their exact calculation generally goes through an underlying Markov chain.
Markov chains
The system is described by a set of states (pumps running, failed, under repair) and transition rates between states. If the rates are constant (exponential durations), the calculation is exact: availability, reliability, mean time to failure. Limitation: the number of states grows very quickly with the number of components. This is chapter 2.
Petri nets
They describe the system with places (conditions), transitions (events) and tokens (resources, states). They naturally represent concurrency (the shared repairer), queues and sequences. Stochastic Petri nets, whose transitions have random durations, reduce to a Markov chain when the durations are exponential. This is chapter 3.
Monte Carlo simulation
The model is run thousands of times by drawing durations at random, then the quantities sought are estimated by averages. There is no restriction on the distributions or the size, but the result is an estimate with an uncertainty, and very rare events are costly to estimate. It is the method of last resort, and often of checking.
A few mentions
Richer formalisms exist: BDMPs (Boolean logic Driven Markov Processes) combine the tree reading and Markov dynamics, and the AltaRica language describes systems made of components and flows. They rely on specialised tools whose names, licences and versions change: check them before any choice for a project.
| Model | What it represents well | Result | Main limitation |
|---|---|---|---|
| Dynamic tree | Order, standby, dependency, while keeping the shape of a tree | Exact or approximate | Calculation often via Markov |
| Markov chain | States, repair, standby, tests | Exact | State explosion; constant rates |
| Petri net | Shared resources, sequences, queues | Exact (reachability) or simulated | Readability of a large net |
| Monte Carlo | Everything, including arbitrary distributions | Estimated with uncertainty | Rare events, computing time |
Dynamic trees, Markov chains, Petri nets and simulation respond to the limits of static trees. The first keep the reading of a tree, the second give an exact result when the rates are constant, the third describe shared resources well, and simulation handles all cases at the price of an uncertainty.