GridWatch AI — Utility Pole Risk Profiling | HackMichigan 2026 | DTE Energy Challenge

In this video, I walk through GridWatch AI — an end-to-end AI-powered utility pole risk profiling system built for the DTE Energy challenge at HackMichigan 2026.

What we built:
GridWatch AI uses machine learning, real-time data pipelines, and IBM watsonx.ai to predict which utility poles in Metro Detroit are most likely to fail — before they actually do. Traditional pole inspection is manual, slow, and reactive. We made it predictive.

How it works:
The system pulls live power pole locations from OpenStreetMap, real-time weather data from Open-Meteo (wind speed, gusts, precipitation), and enriches each pole with 29 structural and environmental features including tilt angle, crack detection, flood zone, soil type, AQI, and soil moisture. Three XGBoost models then run in parallel, one predicts a risk score (0–100), one classifies failure probability, and one estimates storm failure probability. IBM watsonx.ai Granite generates human-readable maintenance recommendations, work orders, and 5-year maintenance schedules for field crews.

Key features shown in this video:
- 🗺 Live Leaflet map with real OSM pole locations and color-coded risk markers
- 🤖 XGBoost ML model trained on 3,000 synthetic Michigan utility poles (R²=0.955, AUC=0.829)
- ⚡ Storm simulation mode: simulate 60–80mph wind events and watch failure predictions update live
- 📷 Computer vision inspection: upload a pole photo and get structural defect analysis
- 🧠 IBM watsonx.ai Granite integration: maintenance recommendations, work orders, crew instructions
- 📊 Analytics dashboard with feature importance, district risk breakdown, and model metrics
- 💰 Budget optimizer: knapsack algorithm allocates maintenance budget across highest-priority poles
- 📅 Risk history tracking and 5-year predictive maintenance scheduling
- 🎯 Custom predict tab: enter any pole parameters and get a live ML prediction instantly

Tech stack:
Python · FastAPI · XGBoost · scikit-learn · pandas · IBM watsonx.ai · Open-Meteo API · OpenStreetMap Overpass API · Leaflet.js · Chart.js · HTML/CSS/JavaScript

Problem this solves:
70% of power outages in the US are weather-related. Utilities like DTE Energy spend billions on reactive maintenance — fixing poles after they fail instead of before. GridWatch AI shifts that to predictive maintenance, reducing outages, optimizing crew deployment, and saving millions in emergency repair costs.

Evaluation criteria we addressed:
✅ Impact: proactive vs reactive maintenance, outage reduction
✅ Innovation: CV + ML + LLM + real-time weather fusion
✅ Accuracy: XGBoost with 29 features, R²=0.955
✅ Usability: field-crew-friendly work orders, dispatcher summaries
✅ Scalability: works across entire DTE service territory
✅ Explainability: every risk score shows top contributing factors, no black box

🔗 GitHub: https://github.com/Manognya86/Hack-Michigan

#machine learning, #IBM watsonx, #utility poles, #predictive maintenance, #XGBoost, #computer vision, #DTE Energy, #HackMichigan, #hackathon, #FastAPI, #Python, #power grid, #AI, #smart grid, #infrastructure
#HACKMI, #IBM

Видео GridWatch AI — Utility Pole Risk Profiling | HackMichigan 2026 | DTE Energy Challenge канала Vanshika Sangtani