Probabilistic graphical model · BEGINNER LESSON
Hierarchical Bayesian network
A hierarchical Bayesian network is a Bayesian network arranged in levels, so variables at different levels represent different kinds of information and can be linked by directed conditional probabilities. It is used to combine uncertainty from multiple sources and update beliefs when new data arrive.
What it is
A hierarchical Bayesian network is a probabilistic graphical model made of nodes and directed links, organized into levels. Each node represents a variable, and the links describe conditional dependence, meaning the probability of one node depends on others above or beside it. In the sources, these levels are used to combine discrete inspection ratings, continuous monitoring measurements, or fault layers in a system. The goal is to produce a unified probabilistic picture from different kinds of evidence rather than treating each data source separately.
[1] Hierarchical bayesian fusion of inspection and monitoring ... [2] Improved bayesian network with graph attention and prior algorithm for aircraft engine fault root cause analysis [3] Machine learning and bayesian network based on fuzzy AHP ...
How it works
The mechanism is Bayesian updating. First, prior information is chosen, which is an initial probability statement before new evidence is used. Then source-specific likelihoods are built from the observed data. Bayes’ theorem combines prior and likelihood to produce a posterior, which is the updated probability after evidence. In the bridge example, the model has layers for dynamic crack behavior, deterioration risk from inspection, and fusion of both. In the aircraft engine example, layers run from microscopic damage to component, subsystem, and system consequences, with directed links constrained to follow the expected propagation path.
[1] Hierarchical bayesian fusion of inspection and monitoring ... [2] Improved bayesian network with graph attention and prior algorithm for aircraft engine fault root cause analysis
Why hierarchy matters
Hierarchy lets the model separate different information levels. One source describes long-term risk from inspections, while another describes time-varying behavior from continuous monitoring. This helps with adaptive updating when new data become available. In the engine example, hierarchy also encodes engineering knowledge about how fault effects move upward through the system. A Bayesian network is also useful when objective data are limited, because it can incorporate expert judgment and other prior knowledge into conditional probability tables, which are tables that list probabilities for each node state under different parent states.
[1] Hierarchical bayesian fusion of inspection and monitoring ... [2] Improved bayesian network with graph attention and prior algorithm for aircraft engine fault root cause analysis [3] Machine learning and bayesian network based on fuzzy AHP ...
Limitations
These models still depend on data quality and modeling assumptions. The bridge study notes weaker performance when there are abrupt damage events, rapid localized deterioration, missing sensing, or long gaps in monitoring. The engine study notes that a directed acyclic graph, meaning a graph with arrows but no cycles, cannot directly represent feedback loops or bidirectional coupling. Another limitation is sparse or incomplete data, which can distort probability estimates or require expert-derived inputs.
[1] Hierarchical bayesian fusion of inspection and monitoring ... [2] Improved bayesian network with graph attention and prior algorithm for aircraft engine fault root cause analysis [3] Machine learning and bayesian network based on fuzzy AHP ...
SEE THE IDEA
How a hierarchical Bayesian network combines evidence
Inspection records, monitoring measurements, or expert knowledge provide different evidence types.
Variables are grouped by level, such as damage, risk, component, subsystem, or system.
Arrows encode how one node depends on parent nodes through probabilities.
Prior information and new likelihoods are combined to produce posterior probabilities.
Evidence flows into layered variables, and the model updates probabilities as new information is added.
A hierarchical structure organizes different evidence sources so Bayes’ theorem can update connected probabilities.
TRY IT YOURSELF
Test your understanding
Choose an answer, then check the explanation.
PUT IT TOGETHER
Match each concept to its meaning
Drag a label onto its description, or select a label and then select a description. This also works with a keyboard.
4 pairs to match
Sources and further reading
This general lesson was generated from public source material. It explains a concept; it does not interpret a private paper or provide medical advice.