Scan-to-Map: Handheld Rebar Finder with Magnetic Sensors + AI Copilot
Project Description
Reinforced concrete is everywhere—buildings, bridges, car parks. Inside the concrete are steel bars (“rebar”) that give the structure strength. Over time, rebar can rust/corrode, which can weaken the structure and increase repair costs. The problem is: today’s inspection tools are often slow, expensive, or require damaging the concrete to check what’s happening inside.
In this UROP, you will help build a portable “rebar scanner” that can see rebar through concrete using magnetic sensing. The core idea is simple: steel rebar changes the magnetic field slightly, and a small sensor can measure those changes. You’ll mount a small magnetic sensor on either:
a handheld sliding scanner (like a ruler/slider you move across a wall), or
a small rail/mini-robot scanner that moves in a neat grid.
By scanning an area, you can turn the measurements into a 2D map that shows where the rebar is (and possibly how deep it is). You’ll also explore whether certain magnetic patterns can give early hints of corrosion.
You’ll work on the full “scan → map” pipeline:
collect magnetic data while scanning,
clean it up (reduce noise, drift, and errors),
and convert it into a rebar map using simple physics ideas + lightweight AI/ML.
At the same time, you’ll learn an AI-copilot engineering workflow inspired by Gabriel Petersson:
Project-first: build a working prototype early (even if it’s rough),
Recursive gap filling: when something breaks or is unclear, identify exactly what you need to learn next and learn it fast,
Teach-back validation: explain your understanding back (to the AI and to teammates) to confirm it’s correct.
By the end, the goal is a working first version of a practical inspection kit: a scanner + scanning method + mapping code + test results—something realistic enough to become a bigger research project or future field trial.
In this UROP, you will help build a portable “rebar scanner” that can see rebar through concrete using magnetic sensing. The core idea is simple: steel rebar changes the magnetic field slightly, and a small sensor can measure those changes. You’ll mount a small magnetic sensor on either:
a handheld sliding scanner (like a ruler/slider you move across a wall), or
a small rail/mini-robot scanner that moves in a neat grid.
By scanning an area, you can turn the measurements into a 2D map that shows where the rebar is (and possibly how deep it is). You’ll also explore whether certain magnetic patterns can give early hints of corrosion.
You’ll work on the full “scan → map” pipeline:
collect magnetic data while scanning,
clean it up (reduce noise, drift, and errors),
and convert it into a rebar map using simple physics ideas + lightweight AI/ML.
At the same time, you’ll learn an AI-copilot engineering workflow inspired by Gabriel Petersson:
Project-first: build a working prototype early (even if it’s rough),
Recursive gap filling: when something breaks or is unclear, identify exactly what you need to learn next and learn it fast,
Teach-back validation: explain your understanding back (to the AI and to teammates) to confirm it’s correct.
By the end, the goal is a working first version of a practical inspection kit: a scanner + scanning method + mapping code + test results—something realistic enough to become a bigger research project or future field trial.
Supervisor
LEE Yi-Kuen
Quota
5
Course type
UROP1000
UROP1100
UROP2100
UROP3100
UROP3200
UROP4100
Applicant's Roles
1 - Build the scanning setup (hardware + motion): Mount a small magnetic sensor on a handheld slider or a simple rail/mini-robot, Make sure the scan is repeatable (scan spacing, consistent sensor height)
2- Collect data + make a rebar map: Record magnetic signals while scanning
Write code to clean up the data (reduce noise/drift) and turn it into a 2D map that shows rebar location
Compare different mapping ideas (simple physics-based rules + basic ML if needed)
3- Test, improve, and validate: Create a simple test rig (concrete sample or mock setup with known rebar positions)
Measure “how good it is” using clear metrics (e.g., position error, depth estimate error, robustness)
4- Use AI as your engineering copilot (Gabriel-style): Project-first: build a rough working prototype early, Recursive gap filling: when stuck, identify exactly what you don’t know and learn only that, Teach-back: explain your solution back to AI / teammates to confirm understanding
2- Collect data + make a rebar map: Record magnetic signals while scanning
Write code to clean up the data (reduce noise/drift) and turn it into a 2D map that shows rebar location
Compare different mapping ideas (simple physics-based rules + basic ML if needed)
3- Test, improve, and validate: Create a simple test rig (concrete sample or mock setup with known rebar positions)
Measure “how good it is” using clear metrics (e.g., position error, depth estimate error, robustness)
4- Use AI as your engineering copilot (Gabriel-style): Project-first: build a rough working prototype early, Recursive gap filling: when stuck, identify exactly what you don’t know and learn only that, Teach-back: explain your solution back to AI / teammates to confirm understanding
Applicant's Learning Objectives
Technical competencies
1 Understand how magnetic sensing “sees through” concrete
2 What a magnetometer measures, and why rebar changes the magnetic field, (Optional) how an AC coil can “excite” a clearer signal
3 Use simple physics to predict what you should see: Make quick back-of-the-envelope estimates: “If rebar is deeper / sensor is higher, how does the signal change?”
4 Build and test a real sensor system: Connect a magnetic sensor to a small board / logger, learn basics of noise, drift, calibration, and how to improve measurement quality
5 Build a repeatable scanning/mechatronics setup: Learn how to scan in a grid (like mowing a lawn), Keep scan spacing and sensor height consistent for better maps
6 Turn raw data into a 2D map + simple rebar detection: Write code to filter data and generate a clean 2D heatmap, try simple algorithms (and optional lightweight ML) to estimate rebar position + depth, Report results with clear metrics (error, success rate, robustness)
AI-augmented engineering competencies (Gabriel-style)
1 Project-first building: Make a “first working version” early, then improve it step-by-step
2 Recursive gap filling: When you get stuck, learn only what you need next (one small gap at a time)
3 Teach-back validation: Practice explaining your understanding back to AI/teammates and checking it with plots/tests
1 Understand how magnetic sensing “sees through” concrete
2 What a magnetometer measures, and why rebar changes the magnetic field, (Optional) how an AC coil can “excite” a clearer signal
3 Use simple physics to predict what you should see: Make quick back-of-the-envelope estimates: “If rebar is deeper / sensor is higher, how does the signal change?”
4 Build and test a real sensor system: Connect a magnetic sensor to a small board / logger, learn basics of noise, drift, calibration, and how to improve measurement quality
5 Build a repeatable scanning/mechatronics setup: Learn how to scan in a grid (like mowing a lawn), Keep scan spacing and sensor height consistent for better maps
6 Turn raw data into a 2D map + simple rebar detection: Write code to filter data and generate a clean 2D heatmap, try simple algorithms (and optional lightweight ML) to estimate rebar position + depth, Report results with clear metrics (error, success rate, robustness)
AI-augmented engineering competencies (Gabriel-style)
1 Project-first building: Make a “first working version” early, then improve it step-by-step
2 Recursive gap filling: When you get stuck, learn only what you need next (one small gap at a time)
3 Teach-back validation: Practice explaining your understanding back to AI/teammates and checking it with plots/tests
Complexity of the project
Challenging