Industrial IoT Sentinel

Autonomous Industrial Telemetry Anomaly Detection & Predictive Maintenance Agent powered by Gemini 3.5 Flash with Real-Time SCADA Digital Twins and Automated ISO Work-Orders.

Comment

Inspiration

In modern industrial automation and power distribution facilities, unexpected equipment downtime is catastrophic—costing manufacturing facilities upwards of $50,000 per hour. Traditional Supervisory Control and Data Acquisition (SCADA) monitoring relies heavily on rigid, single-parameter static threshold alarms. These legacy architectures lead to alarming operator alert fatigue or fail completely to catch complex, multi-sensor cross-anomaly signatures (such as sub-harmonic vibration escalation accompanied by subtle thermal drift and phase current imbalance).

We engineered Industrial IoT Sentinel to bridge the gap between physical SCADA telemetry streams and autonomous generative intelligence. It operates as a 24/7 autonomous reliability engineer capable of ingesting high-frequency sensor streams, determining precise physical root causes, and generating comprehensive ISO 55000 and OSHA-compliant maintenance work orders before catastrophic machinery failure occurs.

What It Does

Industrial IoT Sentinel is an enterprise-grade autonomous predictive maintenance and SCADA anomaly orchestration platform built for The Taskmaster Track:

  • Multi-Sensor Digital Twin Fusion: Continuous ingestion and evaluation of 20 Hz multi-variable telemetry streams, including Vibration Velocity RMS ($v_{rms}$ in mm/s), Bearing/Casing Temperature ($T_c$ in °C), Hydraulic Pressure (PSI), 3-Phase Amperage ($I_{L1, L2, L3}$), and Acoustic High-Frequency Emissions (dB SPL).
  • Deterministic Normative Assessment: Evaluates baseline asset health in real-time against ISO 10816 mechanical vibration severity standards.
  • Agentic Chain-of-Reasoning Diagnostic Core: Leverages Google Gemini 3.5 Flash with strict JSON Schema constraints to perform physical root-cause failure analysis (distinguishing outer-race bearing fatigue from cavitation, misalignment, or electrical unbalance).
  • Predictive RUL & Financial Risk Projections: Accurately forecasts Remaining Useful Life ($RUL_{hours}$) and models potential plant downtime cost exposure ($/hour).
  • Autonomous ISO 55000 / OSHA Work Order Dispatch: Generates actionable maintenance work orders complete with itemized replacement parts (BOM), required technician qualification tiers, estimated Mean Time to Repair (MTTR), and step-by-step OSHA 1910.147 Lockout/Tagout (LOTO) safety protocols.
  • Enterprise Control Plane & Settings Vault: Features a secure client-side API key vault, runtime model tier selector (Gemini 3.5 Flash / Pro), and telemetry sampling frequency tuning.

How We Built It

  • AI Core & Agentic Pipeline: Google Gemini 3.5 Flash API configured with system-instruction contracts and structured JSON Schema validation.
  • Edge & SCADA Workstation UI: High-performance reactive interface built using React 18, Tailwind CSS, real-time SVG vector sparklines, and dynamic telemetry gauges.
  • Cloud Infrastructure & Containerization: Packaged with an optimized multi-stage Docker container and deployed serverless on Google Cloud Run in the us-central1 region.
  • Deterministic Verification Layer: Pre-processes high-frequency telemetry metrics into calculated RMS variances and rate-of-change ($\Delta T/\Delta t$) metrics before passing structured payloads to Gemini.

Challenges We Faced

  1. Bridging High-Frequency Continuous Telemetry with LLM Reasoning: Industrial sensor streams produce continuous time-series data, whereas language models operate on discrete token structures. We resolved this by creating an edge-level deterministic feature extractor that computes rolling RMS velocity and rate-of-change indicators before synthesizing structured JSON payloads for Gemini.
  2. Eliminating Diagnostic Hallucinations in High-Risk Safety Scenarios: Industrial maintenance requires exact component replacement specifications and strict safety procedures. We integrated ISO 10816 vibration classification envelopes and OSHA 1910.147 LOTO compliance rules into the agent's prompt contracts to ensure all generated work orders are 100% actionable and standard-compliant.

Accomplishments That We're Proud Of

  • Sub-second latency for complex multimodal anomaly cross-correlation and root-cause determination using Gemini 3.5 Flash.
  • Zero-latency serverless deployment running smoothly on Google Cloud Run.
  • Automated end-to-end transformation of raw anomalous sensor data into costed, actionable, and printable maintenance work orders.

What's Next for Industrial IoT Sentinel

  • Direct Hardware Integration: Native OPC-UA and Modbus TCP connector daemons for Siemens S7-1500 and Allen-Bradley ControlLogix PLCs.
  • Thermal Imaging Ingestion: Feeding live FLIR radiometric thermal camera streams directly to Gemini for multimodal visual-thermal defect inspection.

Built With

Share this project:

Updates

Submission history