CASE 003 / EDGE AI / PREDICTIVE MAINTENANCEPROTOTYPE
ARAN
Edge-AI predictive maintenance for MSMEs, built at MSME Hackathon 6.0.
01 · Problem & Constraints
Most MSMEs can't afford predictive maintenance — cloud-based industrial IoT platforms are too expensive per machine. Build something that works at roughly ₹3,000 per machine.
- Low per-machine cost (~₹3,000)
- All processing local — no cloud dependency
- Alerts need to be plain-language, not raw sensor data
02 · Architecture & Flow
Explicit Perimeters & Component Flow
01Temperature, vibration, fan RPM and current sensors (DS18B20, ADXL345, SCT-013) per machine
02Raspberry Pi gateway processing all data locally
03Per-machine baseline learning: z-score, moving average and Isolation Forest anomaly detection
040–100 health score with plain-language alerts
Technology Stack
PythonScikit-LearnSQLiteRaspberry PiIoT sensors
03 · Security Decisions
Trust Boundaries Made Legible
- All sensor data stays on-device — nothing leaves the local gateway
04 · Outcome & Roadmap
Verified Results & Real-World Testing
Working prototype at MSME Hackathon 6.0; only flags machines that actually need servicing instead of blanket maintenance schedules.
Next Iterations
Expand the baseline-learning window and validate health scores against real failure data over a longer deployment.