Using machine learning to detect underperforming turbines

2 Min Read

It’s not always clear if a turbine is underperforming, but machine learning can assist in finding out.

VANCOUVER, BC: Clir Renewables has released a new product feature which automatically detects underperforming assets and highlights key actions to rectify underperformance.

A wind turbine will never generate its expected output one-hundred percent of the time, and its performance can and almost certainly will change over time. There are various reasons for this. Some are known but cannot be controlled or managed, such as fluctuating inflow conditions. However, occasionally the opposite is the case, the fault is unknown but can be controlled or rectified. With datasets full of noise from the known reasons, is it possible to extract data to identify the unknown causes? Yes, is the answer.

Using layered machine learning, built on an advanced data model, Clir Renewables has created an underperformance detector for its software solution. This detector works along with other algorithms in the software to analyze the data and classify them based on the reason for the underperformance. The detector creates a synthetic event when turbine power output is well below the historical mean for that wind speed. This invaluable piece of information helps identify ongoing issues at a turbine, not indicated by the SCADA data, inflow conditions under which the turbine does not perform well, and the duration and lost energy associated with the underperformance. It also highlights a hardware or software configuration change that reduces power performance. The detector removes the noise leaving a clean set of data from which the unknown causes can be deduced, and corrective actions created. Alternatively, if the cause is still unidentified, the owner can approach the manufacturer with the cleaned data looking for answers and solutions.

Selena Farris, Data Scientist at Clir Renewables, said “With noise filled datasets the uncertainty of any conclusions that can be drawn on causes of underperformance will increase significantly, and in a lot of cases, issues can be completely masked by the noise. Utilizing the advances in machine learning, a well-structured data model, and deep domain expertise Clir software provides a tool to reduce this uncertainty, generating actionable insights for owners to increase performance and protect their assets from faults and failures.”


About Selena Farris: Selena has over ten years of renewable energy industry experience with a masters degree specializing in wind energy. Working as a developer, consultant, and utility has given her experience and insight into all stages of the wind farm life cycle. She has managed multiple project portfolios and met campaigns, advising on the most financially responsible development path.