A multi-sensor defect detection system combining a USB webcam and thermal infrared camera, deployed as a web-based quality management platform.
Why this thesis exists, and what we set out to prove.
Automotive frames, ship hulls, bridge members, aircraft components — defective steel propagates downstream into systems where failure has real consequences. Manual visual inspection fatigues within minutes, varies between inspectors, and physically cannot match modern production-line speed.
Linear abrasions from foreign objects or interlayer friction during coiling — disrupt protective oxide layers.
Localized depressions from impact pressure — introduce stress concentrations that weaken load-bearing parts.
Premature failure of downstream products, rejected batches, customer-trust erosion, recall liability.
Each objective maps to a distinct artifact — dataset, model, software, hardware, compliance. We'll revisit each one in Chapter 4 with the evidence of how it was met.
What's broken in current inspection, and the system we built to fix it.
Most published systems run a single CNN on a benchmark dataset under controlled lighting. None ship with an audit-grade record system; few survive real factory noise. We needed an inspection platform that works on consumer hardware, fuses two sensors for subsurface detection, and meets ISO 9001 / 10012 expectations out of the box.
USB webcam runs YOLOv11. MLX90640 catches subsurface anomalies. Soft-voting ensemble fuses them. Steel IRIS records every verdict with an immutable audit trail.
Built on the prototype rig, supplemented with a remapped NEU-DET subset. Roboflow handled annotation and augmentation: flips, rotations, contrast, synthetic noise.
Per-class F1: Dent 0.877 / Scratch 0.902. Inference 9.3 ms per image on the mini-PC.
A subsurface dent may not show a visible boundary under overhead LED, but it shifts local heat-conduction. Per-pixel EMA baseline + 2.5σ + 1.0°C + 3-frame persistence flags real anomalies, not sensor noise.
Webcam reaches 90% per-sensor accuracy; thermal 80%. Weights are calibrated empirically — the fused probability gates the verdict at τ = 0.50.
Move the sliders to see how the verdict bin shifts. The composite severity blends visual detections, thermal anomaly, hotspot count, and fusion hits into a 0–100 score.
What the system actually achieved on the held-out test set.
Overall classification accuracy — exceeding the 90% literature threshold for prototype-level systems by 6 percentage points.
How we met each of the five objectives — with the specific evidence.
Every claim in this thesis is bounded by the conditions we actually tested. Every gap has an explicit next step.
Only 8 Good samples — a single FP moves FDR by 12.5 pts. Fix: expand to balanced n ≥ 200.
Trained at 5 cm/s, stable LED. Real factories have variable lighting, vibration, dust. Fix: site-specific re-calibration before deployment.
Webcam underperforms at sheet corners due to barrel distortion and uneven edge illumination. Fix: lens correction + directional side lighting.
Only two defect classes trained. Pitting, scale, laminar cracks excluded. Fix: dataset expansion + retraining.
No formal R&R budget. Fix: per-modality repeatability/reproducibility study to close ISO 10012 §7.3.
One borderline surface-oxidation case (LIVE-003). Fix: fusion veto rule — reject NG when thermal Δ<1°C and YOLO conf<0.55.
It's deployed. You can open it now.
React frontend on Vercel, Supabase PostgreSQL backend, Python hardware acquisition layer streaming from the Mini-PC. The same nine modules you saw earlier — running live.
We've prepared backing materials in case the panel wants to dig deeper.