An offshore wind turbine is about as demanding an environment for computing hardware as you can find outside of aerospace. Salt spray, constant vibration, temperature swings from -20°C to 40°C, humidity at 100%, and zero tolerance for anything that might require a boat to fix. And each turbine is generating thousands of sensor readings per second — blade loads, bearing temperatures, oil viscosity, generator output, tower sway — that need to drive real-time decisions without relying on a connection to shore that might be slow, intermittent, or down entirely.

Edge computing isn’t a nice-to-have for offshore wind. It’s the only architecture that works.

Why Offshore Wind Is a Hard Edge Problem

The connectivity picture for an offshore wind farm is more constrained than onshore SCADA systems. Turbines communicate back to an offshore substation via submarine fibre or wireless links within the array. The substation connects to shore via a high-voltage export cable and, separately, via a subsea data link. That link might have good bandwidth on a calm day. In practice, operators can’t rely on consistent low-latency connectivity for real-time control decisions.

Beyond connectivity, the data volumes are prohibitive to centralise. A modern 15MW offshore turbine might have 5,000+ sensors reporting at frequencies from 1Hz (temperatures, pressures) up to 25kHz (vibration analysis for gearbox and bearing fault detection). Transmitting all of that raw data to shore continuously would require bandwidth that isn’t available and costs that aren’t justifiable. The data needs to be processed at source.

The economics are also brutal. An unplanned outage on an offshore turbine costs £50,000-£150,000 per day in lost generation plus the operational overhead of dispatching a crew by helicopter or service vessel. Any advance warning of a developing fault — even a day’s notice — is worth a significant fraction of that.

The Turbine Edge Node Architecture

A typical production edge architecture for an offshore turbine in 2026 has compute embedded at three levels:

Nacelle controller — the real-time control system, typically a hardened PLC or embedded controller running at microsecond to millisecond control loops. This handles pitch control, yaw, power setpoint, and safety interlocks. It is deterministic, isolated from general data networking, and talks to the SCADA layer via OPC-UA.

Turbine edge server — an industrially rated server (Beckhoff, Siemens, or similar) in the nacelle or tower base. This is where the data processing work happens: ingesting high-frequency sensor streams, running anomaly detection models, computing derived metrics (CMS spectral analysis, SCADA aggregations), and storing a local time-series buffer. Connection to the offshore substation is via the turbine’s internal network.

Farm-level edge gateway — at the offshore substation, an aggregation layer that collects from all turbines, runs cross-turbine analytics (comparing performance across the array, detecting curtailment effects, fleet-wide anomaly detection), and manages the connection to shore. This is also where local SCADA historian data sits for compliance and audit.

Predictive Maintenance at the Edge

The highest-value edge AI application for offshore wind is vibration-based predictive maintenance on gearboxes and main bearings. These are the components most likely to cause an unplanned outage and most expensive to replace in the field.

Gearbox fault detection uses high-frequency accelerometer data (typically 25kHz sampling) and applies spectral analysis — FFT, order tracking, envelope analysis — to identify the characteristic frequency signatures of bearing wear, gear mesh deterioration, and oil degradation. These algorithms run continuously on the turbine edge server.

In a production system, the edge node detects an anomaly pattern, logs the event with the raw waveform, flags it in the local SCADA system, and transmits a compressed summary to shore. Human analysts or a remote monitoring centre review the summary and decide whether to schedule a maintenance visit or continue monitoring. The raw waveform data stays on the turbine unless explicitly requested for deeper analysis.

What makes this genuinely powerful is the combination of fleet comparison and local alerting. A bearing that’s running warm by 5°C compared to other turbines in the array is more meaningful than a single turbine reading 5°C above its own historical baseline — the fleet-level context catches developing faults that individual threshold alerts miss.

Software Stack Considerations

For turbine-level edge nodes, the software environment needs to be maintainable over a 25-year asset lifetime. Key choices:

SCADA and data acquisition: OPC-UA is the standard protocol for turbine data, connecting the nacelle controller to the edge server and up to the farm SCADA. Kepware, Matrikon, and native OPC-UA stacks are all in production use.

Time-series storage: InfluxDB and TimescaleDB are commonly used at the farm level. At the turbine level, lightweight alternatives like QuestDB or DuckDB are increasingly considered for space-constrained nacelle hardware.

ML inference: ONNX Runtime is the pragmatic choice for deploying trained anomaly detection models to edge nodes — models trained centrally in Python or R can be exported and run on the nacelle server without a full ML framework.

Remote management: Over-the-air update capability is non-negotiable for offshore. Balena and Mender are both used; the requirement is verified rollback on update failure, since you cannot physically access a turbine for a failed deployment.

Connectivity and Data Tiering

With reliable bandwidth to shore typically ranging from 2-10 Mbps per turbine (and less during weather events), data tiering is essential. A practical approach:

  • Real-time tier: Aggregated SCADA metrics at 10-second resolution, alarm states, production data — low volume, continuous transmission
  • Event tier: Anomaly events, high-frequency waveforms triggered by alert conditions — burst transmission on demand
  • Analysis tier: Full historical datasets for periodic analysis — scheduled transfer during maintenance windows or low-generation periods

This tiering keeps the shore connection usable for operational data while preserving the ability to retrieve detailed diagnostic data when needed.

The UK North Sea is home to over 15GW of offshore wind capacity in 2026, with another 10GW in construction. Every new wind farm being commissioned is deploying more sophisticated edge infrastructure than the last. The engineering challenges are substantial, but the economics of avoiding unplanned downtime make the investment straightforward to justify.