BN Engine Overview
The Bayesian Network (BN) Engine provides steady‑state availability analysis for repairable systems and mission reliability analysis for non‑repairable systems. It uses probabilistic inference to compute exact results given your model’s assumptions and CPDs.
What is the BN Engine?
The BN Engine applies Bayesian Network theory to your YAML model to calculate the probability that each component (and the overall system) is UP. Dependencies (do, and, or, n_of_k) are encoded as a directed acyclic graph (DAG), and per‑node CPDs are derived from component parameters.
Key advantage
Provides mathematically exact inference (within the chosen model/CPDs), ideal for steady‑state analysis and “what‑if” exploration.
When to use the BN Engine
Best for
- Steady‑state availability analysis (repairable systems)
- Mission reliability analysis (non‑repairable systems)
- Impact analysis (identifying critical contributors)
- Root‑cause analysis (posterior cause when the system is down)
- “What‑if” scenario exploration
- Quick, repeatable validation of architectures
Limitations
- No explicit timing/dynamics (use MC Engine for time series and RTO distributions)
- Can be resource‑intensive for very large models (safety limits apply)
- Relies on steady‑state / mission‑time assumptions encoded in CPDs
Modes & assumptions
-
Availability (repairable): Uses MTTF/MTTR (or
prob) to build steady‑state CPDs for each component. Default distribution is exponential unless otherwise specified. -
Reliability (non‑repairable/mission mode): Set
repair_enabled: falseand providesimulation_time(mission time) in settings. For exponential failure, the CPD uses survival:P(UP) = exp(−t / mttf). Other distributions (e.g.,norm,lognorm) require additional parameters such assigmaand are experimental. -
Dependencies: One dependency type per target (
do,and,or,n_of_k). Logical structure should be modeled on logical nodes; normal nodes usedo:as the target relation.
Core capabilities
Availability calculation
Computes exact marginal P(UP) for each component and the root, honoring all dependencies and per‑node CPDs.
Impact analysis (Top contributors)
Ranks components by their contribution to overall unavailability/reliability loss to guide mitigation priorities.
Root‑cause analysis (posterior cause)
Given evidence that the system is DOWN, computes posterior probabilities of component failures (useful for incident triage).
Sensitivity explorer
Interactive “what‑if” tool to set evidence on components and observe availability deltas at the root and elsewhere.
Technical approach
-
Dependency modeling: supports
do,and,or,n_of_kwith one type per target -
Distributions: default
exp;norm/lognorm(experimental) requiresigma -
Safety limits: resource guards and size limits protect the service on very large graphs
-
Cross‑validation: results can be compared with the MC Engine for consistency
Next steps
For new users
Advanced topics