Skip to main content
Hero Light

​
Predictive Maintenance System

An end-to-end Predictive Maintenance system that monitors industrial assets (motors, pumps, compressors) in real-time and predicts maintenance needs before failures occur.
Real-time sensor monitoring • Dual Isolation Forest anomaly detection • 100Hz batch feature ML • Health scoring • PDF/Excel reporting

​
Get Started

Get your predictive maintenance system up and running in minutes.

Quickstart

Get the system running locally with Docker in under 5 minutes

Architecture

Understand the system architecture and ML pipeline

API Reference

Explore the REST API endpoints and integration options

Dashboard

Try the live demo with real-time monitoring

​
Key Capabilities

Sensor Ingestion

Real-time voltage, current, power factor, vibration data via REST API

Feature Engineering

Rolling means, spike detection, efficiency scores, RMS calculations

Anomaly Detection

Isolation Forest model trained on healthy baseline data

Health Assessment

0-100 health score with risk classification (LOW → CRITICAL)

Fault Simulation

Configurable severity levels (MILD/MEDIUM/SEVERE) for targeted testing

Explainability

Human-readable explanations: “Vibration 3.2σ above normal”

Dashboard

React + Recharts real-time visualization with glassmorphism UI

Reporting

Role-specialized reports: Executive PDF, Multi-sheet Excel, Industrial Certificate

Operator Logs

Ground-truth maintenance event logging with InfluxDB persistence

Baseline Benchmarking

Live status cards display baseline target values for instant comparison

Purge & Re-Calibrate

One-click system reset: wipes InfluxDB data + DI state, returns to IDLE

Cumulative Prognostics

Monotonic Degradation Index (DI), Damage Rate, and Remaining Useful Life (RUL)

​
What Makes This System Unique?

​
Dual-Model ML Pipeline

The system runs two Isolation Forest models trained during calibration:
ModelFeaturesInputF1 ScoreAUC-ROCJitter Detection
Legacy (v2)61Hz averages78.1%1.000❌
Batch (v3)16100Hz windows99.6%1.000✅
The batch model is primary for inference, catching variance-only faults invisible to traditional models.

​
Cumulative Degradation Index (DI)

Monotonic damage accumulation that reflects real-world physics: industrial equipment damage never decreases. A quiet minute doesn’t erase a bearing crack.
Key properties:
  • Dead-Zone: Batch scores below 0.65 (healthy noise) accumulate zero damage
  • Hydration: DI recovered from InfluxDB on restart—state survives process restarts
  • Purge Reset: One-click reset writes DI=0.0 to InfluxDB
  • Prognostics: Health score, damage rate, and Remaining Useful Life (RUL) all derived from DI

​
Explainable AI

Every alert comes with human-readable explanations: 
High vibration variance: σ=0.17g (5x normal baseline)
Voltage spike detected: 3.2σ above rolling mean
Power factor degradation: 0.78 (target: 0.92)

​
Engineering Philosophy

This system prioritizes TRUST over theatrics, PHYSICS over hype, and ENGINEERING RIGOR over vanity metrics.
We optimize for the metrics that matter in production:
  • Can maintenance teams TRUST the alerts?
  • Can auditors VERIFY the decisions?
  • Can engineers EXPLAIN the reasoning?
If the answer to any of these is “no”, the feature doesn’t ship.

​
Live Deployment

ServiceURLStatus
Frontendpredictive-maintenance-ten.vercel.app✅ Live
Backend APIpredictive-maintenance-uhlb.onrender.com✅ Live
API Documentationpredictive-maintenance-uhlb.onrender.com/docs✅ Live

​
Next Steps

Get Started with Docker

Follow our quickstart guide to run the system locally

Explore the Architecture

Learn how the ML pipeline and data flow work

Build docs developers (and LLMs) love

Get started for freeTalk to us