Vessel Predictive Maintenance Platform

NorthSea Shipping maintained 40 vessels on a calendar-based schedule that had nothing to do with the actual condition of the machinery. Unplanned engine failures were causing unscheduled port diversions at an average cost of €600 K per incident in lost charter hire, emergency dry-docking, and expedited spare parts.

1. 1

   Installed ruggedised IoT edge gateways on each vessel to collect vibration, temperature, lube-oil pressure, exhaust-gas temperature, and shaft RPM readings at 500 Hz.

2. 2

   Built an Azure IoT Hub ingestion layer with per-vessel device twins, handling intermittent satellite connectivity through store-and-forward buffering.

3. 3

   Trained LSTM autoencoder models on 6 years of voyage data per vessel class to learn healthy operating envelopes.

4. 4

   Implemented a real-time Spark Streaming scoring pipeline in Databricks that continuously compares live sensor readings against learned baselines.

5. 5

   Created a fleet health dashboard in Power BI Embedded for shore-based fleet managers, showing per-vessel component health scores and ranked alert queues.

We built a vessel-native edge-to-cloud IoT architecture designed for maritime connectivity. Each ship runs a local edge gateway that buffers sensor data during satellite blackouts. Onshore, LSTM autoencoder models detect pre-failure patterns typically 2–3 weeks before a visible fault. Shore-based maintenance planners get ranked alerts so parts and engineers can be pre-positioned at the next port call.