Back to all projects
Defense / DRDO

Predictive Maintenance & Repair Dashboard — WLR Swathi

An AI-driven predictive repair system that forecasts failures and optimizes maintenance cycles for radar subsystems using historical telemetry and operational data.

Role ML Engineer
Domain Predictive Analytics / Radar
Status Deployed

Overview

A predictive maintenance platform for the WLR Swathi weapon-locating radar. The system ingests historical telemetry, maintenance logs, and operational data to forecast subsystem failures before they happen — letting operators shift from reactive repairs to planned, condition-based maintenance. A dashboard surfaces risk scores, time-to-failure estimates, and recommended service windows for each subsystem.

The Problem

Radar subsystems fail in the field with little warning, forcing unplanned downtime during operations. Fixed-interval preventive maintenance either over-services healthy units or misses early failure signatures. A data-driven approach that learns the actual failure signatures from historical data can predict problems days or weeks in advance — turning unplanned failures into scheduled service.

My Role & Contribution

  • Built the data ingestion and cleaning pipeline for heterogeneous radar telemetry and maintenance logs
  • Trained and validated failure-prediction models against held-out historical incidents
  • Designed the operator dashboard and integrated the serving layer with the existing maintenance workflow

Approach

  • Clean and align multi-source telemetry, maintenance records, and operational logs onto a common time base
  • Engineer features that capture drift, anomaly, and degradation patterns across subsystems
  • Train gradient-boosted and time-series models (XGBoost, rolling-window classifiers, survival analysis) to predict failure probability and time-to-failure
  • Serve predictions through a FastAPI service consumed by a React dashboard with Plotly visualizations
  • Close the loop — capture operator feedback on each prediction to retrain and recalibrate

Tech Stack

Python scikit-learn XGBoost Pandas Time-series analysis FastAPI React.js Plotly

Results & Impact

  • Shifted the maintenance posture from reactive to condition-based for covered subsystems
  • Deployed and in active use by maintenance operators
Note: Certain implementation details are covered by DRDO confidentiality. The case study describes the approach at a level appropriate for public sharing.
// TODO: add diagrams / screenshots
← Previous MARL + MANET Simulation