Bayesian Models for Infrastructure Failure Risk

ISEF Category: Environmental Engineering

Subcategory: Other  ·  Difficulty: Advanced  ·  Setup: Home Setup  ·  Time: Full Year

The Hook

When a storm knocks out power, water pumps can fail next, and roads can flood soon after. That chain reaction matters because one broken system can pull down the others. You can model those dominoes with real city data. A good project here is not about predicting every outage, it is about finding which links in the chain matter most.

What Is It?

A Bayesian network is a map of connected events with probabilities attached. Think of it like a decision tree that can loop through many possible paths at once. In this project, the events are things like heavy rain, power outage, pump failure, road closure, and service restoration. The model estimates how likely one problem is to trigger the next.

This topic sits at the intersection of infrastructure and weather risk. Extreme storms do not hit one system at a time. Power, water, and transport depend on each other, so a failure in one can spread. Your job is to use open datasets to estimate those links and test how the network changes under different storm conditions.

Why This Is a Good Topic

This makes a strong science fair topic because you can test clear cause-and-effect questions with real data, even if you do not build physical hardware. You can study a real-world problem that affects public safety, emergency response, and city planning. You also get to learn data cleaning, probability, model validation, and systems thinking. That mix can grow into a serious research project if you compare multiple cities, storm types, or dependency structures.

Research Questions

• How does storm severity change the predicted probability of cascading failures across water, power, and transport systems?
• What is the effect of including infrastructure dependency links on model accuracy compared with a model that treats each system separately?
• Does a Bayesian network trained on one city generalize to another city with a different infrastructure layout?
• To what extent do flood-prone roads increase the chance of delayed power restoration after extreme weather?
• Which infrastructure node, power, water, or transport, has the largest impact on downstream service failures in open city resilience data?
• How does adding historical outage data change the uncertainty of cascade predictions?
• What is the effect of different weather thresholds on the number of predicted cascading failure paths?

Basic Materials

• Laptop or desktop computer with internet access.
• Spreadsheet software for data cleaning and basic charts.
• Open city-resilience datasets from city open data portals.
• Weather and storm history data from NOAA or NASA.
• Python installed with pandas and networkx.
• Jupyter Notebook for analysis and notes.
• GIS viewer or map tool such as QGIS for inspecting location-based patterns.
• External storage or cloud folder for versioned project files.

Advanced Materials

• Python with pgmpy or bnlearn for Bayesian network modeling.
• Jupyter Notebook for reproducible analysis.
• QGIS for spatial joins and map-based analysis.
• PostGIS or SQLite for handling larger linked datasets.
• ImageJ or another image tool if you also analyze flood or damage photos.
• Public infrastructure, outage, and transportation datasets from city portals, NOAA, USGS, FEMA, or utility reports.
• Statistical package such as R for cross-validation and uncertainty testing.
• High-resolution GIS layers for roads, flood zones, and critical facilities.

Software & Tools

• Python: Cleans datasets, builds the Bayesian network, and runs probability calculations.
• Jupyter Notebook: Keeps your analysis, code, and notes in one place.
• pandas: Organizes outage, weather, and infrastructure tables.
• pgmpy: Helps you build and test Bayesian network models in Python.
• QGIS: Lets you compare spatial patterns in flood zones, roads, and critical infrastructure.

Experiment Steps

1. Define the exact cascade you want to study, such as weather to power to water to transport, and decide which city or cities you will compare.
2. Identify open datasets that contain event timing, outage records, storm data, and location data, then check that the variables can be linked.
3. Choose the nodes in your Bayesian network and decide which links represent direct dependence instead of simple correlation.
4. Plan how you will turn raw records into probabilities, categories, or time windows so the model can learn from the data.
5. Design a validation plan that compares your network predictions with held-out events, alternate cities, or a simpler baseline model.
6. Decide which sensitivity test will show the strongest failure pathway and which variable will matter most when conditions worsen.

Common Pitfalls

• Mixing datasets with different time stamps, which makes the model connect outages to the wrong storm event.
• Treating correlation as causation, which can create fake dependency links between systems.
• Using inconsistent location names or boundary files, which prevents water, power, and transport records from matching.
• Building too many nodes at once, which makes the Bayesian network sparse and hard to train from public data.
• Ignoring missing outage records, which can make the cascade look weaker or stronger than it really was.

What Makes This Competitive

A stronger version of this project goes past a simple network diagram. You can compare several cities, test different storm types, and show whether the same dependency structure holds up across places. You can also add uncertainty analysis, then explain which links remain stable and which ones are data weak points. That kind of careful validation makes the work feel like real systems research, not just a class model.

Project Variations

• Use a coastal city dataset and test how storm surge changes cascade risk across drainage, power, and road systems.
• Focus on one infrastructure pair, such as power and water, and build a tighter Bayesian network with stronger validation.
• Add equity analysis by comparing cascade risk in neighborhoods with different flood exposure or service access.

Learn More

• NOAA National Centers for Environmental Information: Search for storm, flood, and extreme weather datasets that you can pair with infrastructure records.
• NASA Earthdata: Find satellite and climate data that help you describe storm conditions and flood exposure.
• USGS Water Data: Look up stream, flood, and watershed data for connecting weather to local hazard conditions.
• FEMA OpenFEMA: Search for disaster, mitigation, and infrastructure-related datasets and reports.
• PubMed: Search for review articles on infrastructure interdependence, resilience modeling, and cascading failures.
• MIT OpenCourseWare: Find free materials on probability, networks, and systems modeling that support Bayesian analysis.