Campus Seismometer Array With MEMS Sensors

ISEF Category: Embedded Systems

Subcategory: Sensors  ·  Difficulty: Advanced  ·  Setup: School Lab  ·  Time: Full Year

The Hook

Your campus already shakes all day. Most of those vibrations come from traffic, footsteps, construction, or distant seismic activity. A sensor array can tell them apart if you place the sensors well and process the signal smartly. That gives you a real engineering problem with real-world stakes.

What Is It?

This project uses small motion sensors called MEMS accelerometers. MEMS stands for microelectromechanical systems. Think of them as tiny chips that can feel acceleration, tilt, and vibration. If you mount several of them in fixed locations, they can act like a simple seismometer network.

The PVC pipe part matters because it gives each sensor a repeatable housing and helps protect the electronics outdoors. Your job is to turn raw motion readings into useful labels. That means you will collect signals, clean them up, and teach a model, like a CNN, to tell one kind of vibration from another. A CNN, or convolutional neural network, is a pattern-finding model that can learn shapes in data, not just images.

The deeper challenge is localization. If a vibration reaches one sensor before another, you can estimate where the source started. That turns your project from a simple detector into a small campus mapping system.

Why This Is a Good Topic

This is a strong science fair topic because you can test real signals, compare classes of events, and measure whether your system gets better with more sensors or smarter placement. It connects to earthquake sensing, infrastructure monitoring, and smart cities. You can learn hardware design, signal processing, and machine learning without needing a university lab. The project also gives you lots of room to make original choices in sensor layout, event labeling, and model design.

Research Questions

• How does sensor spacing affect how accurately a campus array localizes vibration sources?
• What is the effect of mounting depth or housing material on vibration signal quality?
• Does adding a second or third sensor axis improve classification of construction, traffic, and seismic-like events?
• To what extent does a CNN outperform simpler models for event classification from MEMS accelerometer data?
• Which frequency bands carry the most useful information for separating foot traffic from heavy equipment vibrations?
• How does the number of labeled training events affect classification accuracy across different campus locations?

Basic Materials

• MEMS accelerometer breakout boards with a common interface, such as I2C or SPI.
• Microcontroller boards for data collection, such as Arduino or Raspberry Pi.
• PVC pipes and caps for sensor housings.
• Jumper wires, breadboard, and connectors.
• USB cables and a laptop for data logging.
• Tape measure or survey wheel for mapping sensor positions.
• Mounting hardware, such as brackets, zip ties, or foam pads.
• Notebook or spreadsheet for event logs and ground truth labels.

Advanced Materials

• Industrial or research-grade MEMS accelerometers with known noise characteristics.
• Multi-channel data acquisition hardware with synchronized sampling.
• GPS or network time sync hardware for time alignment across stations.
• Weatherproof enclosures and cable glands for outdoor deployment.
• Calibration shaker or vibration table for sensor response checks.
• Reference geophone or seismometer for comparison measurements.
• Laser rangefinder or survey equipment for precise station coordinates.
• Soldering tools and PCB prototyping supplies for cleaner sensor nodes.

Software & Tools

• Python: Cleans accelerometer data, extracts features, and runs model training.
• Jupyter Notebook: Helps you test processing ideas and document your analysis in one place.
• ImageJ: Can help you inspect signal plots or exported visual summaries, if you create image-based inputs.
• QGIS: Maps sensor locations and event source estimates across campus.
• TensorFlow or PyTorch: Trains a CNN to classify vibration patterns.

Experiment Steps

1. Define the event classes you want to separate, such as traffic, construction, footsteps, and seismic-like tremors.
2. Plan the sensor layout so you can compare how position and spacing change source localization.
3. Design a data logging format that keeps timestamps, sensor location, and ground truth labels together.
4. Build a calibration plan so each sensor reports signals on a comparable scale.
5. Choose a signal representation, such as raw waveforms, spectrograms, or summary features, before you train any model.
6. Set up evaluation metrics that check both classification accuracy and location error.

Common Pitfalls

• Letting each sensor sample at a slightly different clock rate, which breaks source timing comparisons.
• Mounting sensors loosely inside PVC pipes, which adds extra vibration and distorts the signal.
• Mixing events from different campus spots without clean labels, which teaches the model the wrong patterns.
• Ignoring background noise from HVAC units, doors, and foot traffic, which hides the vibration classes you care about.
• Training and testing on the same day or same location, which makes the model look better than it really is.

What Makes This Competitive

A stronger version of this project does more than classify a few vibrations. It checks whether the system still works across different weather, surfaces, and campus zones. You can raise the level by comparing several feature types, testing multiple models, and reporting confusion patterns, not just accuracy. If you also estimate source location with error bounds, your project starts to look like a real sensing system, not a classroom demo.

Project Variations

• Focus only on classifying daytime campus vibrations versus nighttime background noise, then compare performance across locations.
• Replace the CNN with a simpler model, such as random forest or SVM, and test whether deep learning really helps on this dataset.
• Add a mapping layer that estimates the source point of each event and compares the result with known construction sites or road edges.

Learn More

• NASA Earthquake and Seismic Resources: Search NASA for background on vibration sensing, wave propagation, and Earth observation links to seismic events.
• USGS Earthquake Hazards Program: Use USGS pages and data tools to learn event types, waveform basics, and how seismologists describe signals.
• NIH PubMed: Search review articles on MEMS accelerometers, vibration sensing, and event classification for peer-reviewed background.
• MIT OpenCourseWare: Look for signal processing and machine learning course materials that cover filtering, feature extraction, and classification.
• NOAA National Centers for Environmental Information: Use NOAA data tools to understand environmental noise sources and regional context for field measurements.