Industrial IoT: Predictive Maintenance & Financial Lakehouse
Project Core Objective
The primary goal of this project was to move beyond simple “threshold alerts” in manufacturing. I developed a Full-Stack Data Engineering Pipeline that translates raw mechanical telemetry (vibration and temperature) into Real-Time Financial Risk. By correlating sensor anomalies with live energy market costs via external APIs, the system enables Profit-First operational decisions.
Methodology & Technical Points
1. Medallion Data Architecture
The pipeline follows a modular Medallion Architecture to ensure rigorous data quality and lineage:
- Bronze Layer (Ingestion): Multi-threaded simulation of high-frequency IoT sensors capturing vibration, temperature, and power metrics.
- Silver Layer (Transformation): Implements a Quality Gate that deduplicates records and applies industrial standards to flag Sensor Drift or critical vibration thresholds.
- Gold Layer (Analytics): High-level business logic integrating an External Energy Market API to calculate the
hourly_energy_cost_eurfor every active machine.
2. AI Agent: “The Plant Manager”
Instead of a static dashboard, this project features an Autonomous SQL Agent acting as a virtual Plant Manager.
- Framework: Built using LangChain and Ollama.
- Reasoning Engine: Powered by Local Llama-3 for 100% offline inference, ensuring sensitive industrial telemetry remains secure.
- Logic (ReAct Strategy): The agent utilizes a Reason/Act loop. It analyzes a user’s natural language question, generates a precise SQL query, executes it against the Gold Layer, and interprets the financial impact.
Sample AI Reasoning Trace
User: “Which machine should I prioritize for repair based on financial loss?”
AI Agent: “Based on current telemetry and energy prices of €198/MWh, you should prioritize COMPRESSOR_03. It currently accounts for 63.14% of total factory hourly loss due to thermal inefficiency.”
Engineering Features
Data Governance & Pipeline Health
To ensure the reliability of the executive dashboard, I implemented an automated Governance Audit (data_governance.py):
- Health Scorecard: The system calculates a Pipeline Health % based on data freshness and the ratio of records passing the Silver Quality Gate.
- Safety Thresholds: If the health score drops below 90%, the orchestrator issues a critical warning to prevent decisions based on corrupted data.
Business-Driven Logic
- Energy Correlation: The system identifies Thermal Waste and multiplies that inefficiency by the current market rate.
- Automated Recommendations: The pipeline generates a
SHUTDOWNrecommendation if a machine is both inefficient and operating during peak-cost energy hours.
Technical Performance Metrics
| Metric | Achievement |
|---|---|
| Pipeline Reliability | Quality-check flags 100% of anomalous readings. |
| Data Integrity | Governance audit ensures 90%+ quality before Gold layer. |
| Inference Security | 100% Local (No external data leaks via Ollama). |
Author: Gitesh Kumar
Project Context: Industrial IoT & Agentic Data Engineering
Tech: Python, SQL, LangChain, Llama-3, Power BI
