Snowflake-Labs/sfguide-ai-powered-predictive-grid-maintenance

Snowflake-Labs/sfguide-ai-powered-predictive-grid-maintenance

Language: Python

License: Apache-2.0

Stars: 10

Forks: 5

Open issues: 0

Created: 2026-01-08T18:37:40Z

Pushed: 2026-02-19T16:20:11Z

Default branch: main

Fork: no

Archived: no

README:

AI-Powered Energy Utilities Predictive Grid Maintenance

A complete predictive maintenance solution for power grid transformers built on Snowflake's AI Data Cloud.

Built By: Sri Subramanian, Senior Solution Engineer GitHub | LinkedIn

---

Overview

This solution demonstrates how utilities can leverage Snowflake to build an end-to-end predictive maintenance system combining:

• Structured Data: Sensor readings, asset metadata, maintenance history
• Unstructured Data: Technical manuals, maintenance logs, visual inspections
• Machine Learning: XGBoost failure prediction, Isolation Forest anomaly detection, RUL estimation
• Cortex Agents: Natural language queries via Snowflake Intelligence
• Real-time Dashboard: Streamlit-powered monitoring and alerting

Business Value

| Metric | Impact | |--------|--------| | Cost Avoidance | $400K+ per prevented failure | | SAIDI Improvement | Reduced outage duration | | Customer Protection | Proactive maintenance for 100 transformers | | ROI | 10x+ return on predictive maintenance investment |

---

Repository Structure

sfguide-ai-powered-predictive-grid-maintenance/
│
├── deploy.sh # Main deployment script
├── clean.sh # Teardown script
├── run.sh # Runtime operations (validate, status, test)
├── requirements.txt # Python dependencies
│
├── scripts/ # SQL Scripts
│ ├── 01_infrastructure_setup.sql
│ ├── 02_structured_data_schema.sql
│ ├── 03_unstructured_data_schema.sql
│ ├── 04_ml_feature_engineering.sql
│ ├── 05_ml_training_prep.sql
│ ├── 06_ml_models.sql
│ ├── 06b_update_score_assets.sql
│ ├── 07_business_views.sql
│ ├── 08_semantic_model.sql
│ ├── 09_intelligence_agent.sql
│ ├── 10_security_roles.sql
│ ├── 10_streamlit_dashboard.sql
│ ├── 11_load_structured_data.sql
│ ├── 12_load_unstructured_data.sql
│ ├── 13_populate_reference_data.sql
│ ├── 14_generate_recent_sensor_data.sql
│ └── 99_sample_queries.sql
│
├── streamlit/ # Streamlit dashboard
│ ├── grid_reliability_dashboard.py
│ └── environment.yml
│
├── data_generators/ # Data generation scripts
│ ├── generate_asset_data.py
│ ├── generate_maintenance_logs.py
│ ├── generate_technical_manuals.py
│ ├── generate_visual_inspections.py
│ └── load_unstructured_full.py
│
└── utilities/ # Utility scripts
└── test_snowflake_connection.py

---

Prerequisites

• Snowflake Account with ACCOUNTADMIN role (sign up for free trial)
• Python 3.8+ (download)
• OS: macOS, Linux, or Windows (with WSL or Git Bash)

Verify installation:

python3 --version
pip --version

---

Setup Guide

Step 1: Install Snowflake CLI

pip install snowflake-cli

Or download installer from Snowflake CLI releases

Verify: snow --version

---

Step 2: Configure Snowflake Connection

snow connection add

| Prompt | Value | |--------|-------| | Connection name | default | | Account | Your account identifier | | User | Your username | | Password | Your password | | Role | ACCOUNTADMIN | | Other prompts | Press Enter to skip |

> Find your account identifier in the URL: https://.snowflakecomputing.com

---

Step 3: Test Connection

snow connection test -c default

---

Step 4: Download Repository

Option A: Clone with Git

git clone https://github.com/Snowflake-Labs/sfguide-ai-powered-predictive-grid-maintenance.git

Navigate into the cloned folder:

cd sfguide-ai-powered-predictive-grid-maintenance

Option B: Download ZIP 1. Go to GitHub Repository 2. Click Code → Download ZIP 3. Extract the ZIP file

Navigate to the extracted folder:

cd /path/to/sfguide-ai-powered-predictive-grid-maintenance

> Note: Replace /path/to/ with the actual location where you cloned or extracted the repository (e.g., ~/Downloads/ or ~/Desktop/). All subsequent commands should be run from inside this folder.

---

Step 5: Deploy

chmod +x deploy.sh
./deploy.sh -c default

> Note: The deployment script automatically creates a Python virtual environment and installs all required dependencies. If dependency installation fails, you can manually run: > ``bash > pip install -r requirements.txt >

| Phase | Description | Duration | |-------|-------------|----------| | 1 | Infrastructure setup | ~1 min | | 2-3 | Data schemas | ~2 min | | 4-5 | ML pipeline | ~2 min | | 6 | Analytics views | ~1 min | | 7 | Data generation & loading | ~5 min | | 8 | ML training | ~3 min | | 9 | Streamlit dashboard | ~1 min |

Total: ~15-20 minutes

---

Step 6: Validate Deployment

./run.sh validate -c default

---

Step 7: Access the Solution

Streamlit App:

• Snowflake UI → Projects → Streamlit → GRID_RELIABILITY_DASHBOARD
• Role: GRID_ANALYST, GRID_OPERATOR, or GRID_ADMIN

Cortex Agent (Snowflake Intelligence):

• Snowflake UI → Projects → Snowflake Intelligence → Grid Reliability Intelligence Agent
• Role: GRID_ANALYST, GRID_OPERATOR, or GRID_ML_ENGINEER

---

What Gets Deployed

Database Schemas

| Schema | Purpose | |--------|---------| | RAW | Sensor readings, asset master, maintenance history | | UNSTRUCTURED | Technical manuals, maintenance logs, inspections | | FEATURES | Engineered ML features | | ML | Model registry, predictions, training data | | ANALYTICS | Business views, semantic model |

ML Models

| Model | Purpose | |-------|---------| | XGBoost Classifier | Failure probability prediction | | Isolation Forest | Anomaly detection | | Linear Regression | Remaining Useful Life (RUL) |

Risk Score

Risk Score = (Anomaly × 0.3) + (Failure_Prob × 0.5) + (RUL_Factor × 0.2)

0-40: Low (routine monitoring)
41-70: Medium (increased monitoring)
71-85: High (schedule maintenance)
86-100: Critical (immediate action)

---

Explore the Solution

Streamlit App Pages

| Page |…