Wind Turbine Fault Diagnosis with Artificial Intelligence Tools

Wind power capacity is expanding globally. Remote condition monitoring of wind turbines helps to maximise uptime and minimise maintenance expenses. Artificial neural models can identify emerging issues in renewable energy conversion systems, such as wind machines. Thus, the current work highlights the significant enhancement in fault detection accuracy for turbines with limited data by implementing cross-turbine deep learning. Specifically, it demonstrates that integrating information from turbines with limited and turbines with abundant data allows for the early identification of faults, surpassing previous methods. Obtaining sufficient and accurate data for training fault detection models can be challenging. Often, the available data is insufficient or does not accurately represent the current operation behaviour. Wind farms that have recently been commissioned are missing data from their previous operation. Some older turbines might not have accurate data due to updates in control software or hardware replacements. Following such events, the operation behaviour of a turbine can undergo significant changes, thus the time series acquired from the monitored process do not represent its actual working condition. To this end, the study emphasises the importance of sharing and applying knowledge from one wind turbine to another to address the issue of limited data and facilitate the early fault diagnosis in wind machines.