Low-Cost Masonry Crack Monitor

ISEF Category: Engineering Technology: Statics and Dynamics

Subcategory: Civil Engineering  ·  Difficulty: Advanced  ·  Setup: School Lab  ·  Time: Full Year

The Hook

A wall can look fine from across the room and still be failing at the cracks. That is why engineers watch masonry for tiny changes before the damage spreads. You can build a cheap camera system that spots those changes on mock walls. Then you can test how close your model gets to a laser-based baseline.

What Is It?

This project asks a simple question. Can a tiny computer and a cheap camera detect wall damage before a person would notice it?

Think of the system like a guard dog with a camera for eyes. The camera watches a masonry surface. The model, YOLOv8-nano, scans each image for signs of cracks or tilt. YOLO means "You Only Look Once," a fast object-detection method that can label parts of an image in one pass. Nano means the small version, which fits on low-power hardware like a Raspberry Pi Zero 2 W.

The key idea is calibration. You first create known test surfaces, such as plaster-of-Paris brick mockups. Then you induce controlled cracks or shifts, record what the camera sees, and compare the model's output with a laser-pointer baseline that gives a more stable reference. The result is a machine-vision system that tries to turn visual damage into measurable data.

Why This Is a Good Topic

This is a strong science fair topic because you can test it with clear variables, real measurements, and a real engineering problem. Buildings, retaining walls, and older masonry structures all need early damage detection. You can study image quality, crack size, camera angle, lighting, and model accuracy, then see which factors matter most. That gives you a project with both technical depth and a practical use.

Research Questions

• How does crack width affect the accuracy of YOLOv8-nano detection on plaster-of-Paris masonry mockups?
• What is the effect of camera angle on tilt detection accuracy for a Raspberry Pi Zero 2 W monitoring system?
• Does changing the lighting condition alter false positive rates in crack detection?
• To what extent does image resolution improve the model's ability to detect hairline cracks?
• Which baseline method, laser-pointer alignment or manual image marking, gives more consistent ground truth labels?
• What is the effect of mockup surface texture on crack detection confidence scores?

Basic Materials

• Raspberry Pi Zero 2 W.
• $5 USB camera.
• microSD card.
• Power supply for Raspberry Pi.
• Computer for setup and data transfer.
• Plaster of Paris.
• Brick mold or simple rectangular molds.
• Cardboard, foam board, or wood base for mockups.
• Ruler or caliper.
• Laser pointer.
• Measuring tape.
• Strong tape or clamp system.
• Notebook or spreadsheet for observations.
• Phone or digital camera for backup photos.

Advanced Materials

• Raspberry Pi Zero 2 W.
• USB camera with known resolution and adjustable focus.
• Laser pointer and mount for fixed alignment.
• Calibrated ruler or digital caliper.
• Plaster of Paris or cement-based mortar samples.
• Reusable masonry mockup molds.
• Tripod or rigid camera mount.
• Lighting box or constant LED setup.
• Image target or checkerboard for camera calibration.
• External battery pack or power meter.
• Computer with Python and model training tools.
• Annotated image dataset of cracks and tilt states.
• Protective gear for handling dust and broken material.

Software & Tools

• Python: Organizes the image workflow, labels results, and runs analysis scripts.
• OpenCV: Prepares images, measures edges, and helps compare frames under different lighting.
• Ultralytics YOLOv8: Trains or tests the object detection model on crack images.
• ImageJ: Measures crack length, width, and visible surface change from photos.
• Google Colab: Lets you train or test models on a free cloud notebook if your computer is slow.

Experiment Steps

1. Define one damage signal first, such as crack presence, crack width, or wall tilt, so your model has one clear target.
2. Build a repeatable mock wall system that lets you create similar samples across trials.
3. Choose a baseline measurement method, then decide how you will label each image for comparison.
4. Plan a photo capture setup with fixed camera position, fixed lighting, and fixed background so image changes come from the wall, not the scene.
5. Set your evaluation metric before collecting data, such as precision, recall, false positive rate, or mean error against the baseline.
6. Map out a test matrix that changes one factor at a time, such as crack size, viewing angle, or surface texture.

Common Pitfalls

• Changing the camera distance between trials, which makes crack size in pixels drift and breaks comparison across samples.
• Training on mockups that all look too similar, which causes the model to fail on new crack shapes or textures.
• Using uneven room lighting, which changes shadow patterns and makes the detector confuse shadows with cracks.
• Labeling cracks inconsistently, which gives the model noisy ground truth and weakens accuracy.
• Comparing the model to a shaky laser setup, which adds error to the baseline and hides whether the vision system actually improved.

What Makes This Competitive

A class-level version of this project only checks whether the model can find cracks. A stronger version tests when, why, and where the system fails. You can compare multiple surface textures, crack types, and camera angles, then use a real statistical test to show which factors matter most. If you also compare the vision system against another low-cost detection method, your project starts to look like engineering research instead of a simple demo.

Project Variations

• Test the same crack detector on brick, concrete, and plaster mockups to see how surface texture changes accuracy.
• Swap YOLOv8-nano for a simpler edge-detection pipeline and compare speed, error rate, and false alarms.
• Add a tilt-only experiment that measures how well the system spots small wall-angle changes before cracks appear.

Learn More

• USGS Earthquake Hazards Program: Search for earthquake damage and building assessment resources, then look for masonry and structural monitoring materials.
• FEMA Building Science: Find free guides on structural damage, wall cracking, and post-disaster inspection basics.
• NIST Technical Notes: Search NIST publications for structural monitoring, crack detection, and image-based inspection studies.
• NIH PubMed: Search review articles on computer vision crack detection, structural health monitoring, and image classification.
• MIT OpenCourseWare: Look for civil engineering, computer vision, or signals and systems course materials that explain image analysis basics.