Revolutionising Marine Industry Security: The Cyber-Physical Systems Programme

Downtime for offshore wind turbines is very costly as it involves both the lost electricity generation and the costs of bringing crews to the turbines. Currently, there are considerable challenges with reliable failure prediction and condition-based maintenance programmes for generators.

The aim of the project is to create a specialised system using machine learning (ML) and Smart Data technologies that will monitor turbines operational condition, identify performance issues and predict future breakdowns. This will be achieved by collecting data from the wind turbines’ Supervisory Control and Data Acquisition (SCADA) system such as electrical, temperature and pressure, combined with vibration and acoustic data which will be obtained via sensors. A suitable source of existing SCADA data was identified – the Levenmouth Demonstration Turbine.  

This project focuses on carrying out an initial feasibility study on the use of wind turbine SCADA data to predict unhealthy conditions (alarms) in the generator. This will be used as a steppingstone in developing a cutting-edge condition monitoring system for wind turbine generators that reduce lost electricity generation due to malfunction, as well as reducing the need for unnecessary manual maintenance checks. The predicted impact of the project is that enhanced wind turbines should lead to a significant financial saving for windfarm operators, therefore further increasing business interest in the renewables sector.  

Inferential measurement systems aim to model the relationship between primary characteristics that are difficult to measure directly and secondary variables that can be more easily monitored, either using on-board sensors or can be obtained from the ground stations. The situation awareness of unmanned aerial systems (UAS) can be considered as a primary characteristic affected by numerous variables ranging from the levels of perception (e.g., visibility of target and objects) to operational conditions (e.g., torque of the engine). The information obtained through inferential measurements can be used to control and optimise the operation of UAS without the need to install additional sensors on board or to excessively rely on the instrumentation data coming from the ground stations.

Various modelling paradigms may be used for inferential measurement, including data-driven modelling methods, such as time-series analysis, artificial neural network, Bayesian probabilistic modelling, support vector machines and genetic programming. The proposed approach for this project is to choose, build and evaluate an inferential measurement system for a UAS that is capable of enhancing its situational awareness through real-time analysis of sensor inputs and evaluating relevant secondary variables.