The Inspection Challenge
Automotive hose assemblies — coolant, fuel, vacuum, brake and power-steering lines — are manufactured and washed in mixed batches and must be segregated post-washing. Manual sorting above 20 units per minute typically degrades accuracy below 95%, creating risks of mis-binning, assembly delays, and field failures. For safety-critical components like brake lines, single errors can trigger recalls.
| Challenge Area | Observed Impact | Risk Level |
|---|---|---|
| Mixed-batch post-wash flow | Part numbers intermixed on conveyor; no automatic separation | Critical |
| Near-identical part profiles | Operators cannot reliably distinguish OD variants within 1 mm | Critical |
| Colour-code ambiguity (wet parts) | Water film desaturates stripe colours, defeating visual coding | High |
| Manual ASRS bin assignment | Operator-entered bin codes cause mis-routing and inventory errors | High |
| Throughput vs. accuracy trade-off | Speed ramp-up forces accuracy compromise above 20 units/min | High |
Why Traditional Methods Fall Short
Machine vision reads the physical part directly, making its classification immune to label damage, lighting shifts and operator fatigue. Traditional approaches all fail in post-wash environments — manual sorting, barcode scanning, color charts, hand gauging, weight sorting, RFID, and batch declarations.
| Method | Limitation | Failure Mode |
|---|---|---|
| Manual visual sort | Fatigue-driven error rate 3–8% after 2 hours continuous | Mis-bin, line stop |
| Barcode / QR scanning | Labels wet, torn or absent post-wash; scan rate drops sharply | Unread parts, holds |
| Colour chart comparison | Ambient light variation and wet surfaces defeat naked-eye grading | Mis-classification |
| Dimensional hand gauging | Contact measurement at line speed is unsafe and impractical | Throughput loss |
| Weight sorting | Hoses of different types may share weight ranges within tolerance | False accept |
| Operator batch declaration | Relies on paperwork integrity; no unit-level verification | Audit non-conformance |
Suggested Machine Vision Architecture
Two area-scan cameras (overhead and lateral) under structured LED illumination capture hoses at the wash exit. A convolutional neural network with ResNet backbone processes dual-view images to identify part type, outer diameter, length, color stripe patterns, and surface anomalies within 120 milliseconds. Results are published via OPC-UA to SCADA and ASRS systems for real-time bin routing.
| Detection Parameter | Method | Accuracy | Cycle Time |
|---|---|---|---|
| Part-type classification | CNN (ResNet backbone) | > 99.5% | < 120 ms |
| Outer diameter category | Calibrated metrology | ± 0.5 mm | < 80 ms |
| Hose length range | Pixel-count metrology | ± 2 mm | < 80 ms |
| Colour stripe pattern | Colour vision + ML | > 99% | < 100 ms |
| Kink / crush defect | Contour morphology | > 97% | < 120 ms |
| Surface mark / cut | Anomaly detection DL | > 96% | < 140 ms |
| Wrong-part / mix-up flag | Classification + catalogue match | > 99.5% | < 120 ms |
Expected Outcomes & ROI
A single automotive hose recall can exceed ₹5–50 crore in direct costs and OEM penalty charges. Vision-guided segregation typically delivers payback within 12–18 months.
| Outcome Metric | Baseline (Manual) | Target (Vision) | Improvement |
|---|---|---|---|
| Part-type mis-sort rate | 3–8% | < 0.1% | > 97% reduction |
| Throughput (hoses/min) | 15–20 | 30–60 | 2–3× increase |
| Operator headcount (sorting) | 2–4 per shift | 0–1 (oversight) | 2–3 FTE redeployed |
| ASRS bin accuracy | 92–95% | > 99.5% | Near-zero mis-bin |
| Part traceability coverage | < 30% (batch level) | 100% (unit level) | Full per-hose record |
| Payback period | — | Typically 12–18 months | Positive ROI |
