Posters

Abstract

In the quest for operations and maintenance (O&M) efficiency, maintenance reporting has the potential to translate into meaningful intelligence. By encouraging and educating stakeholders to properly complete maintenance documents, wind farm operators are building a gold mine of information. Too often overlooked, this gathered unstructured data is a key element to improve KPIs precision and reliability. Properly analysing this data is now possible with artificial intelligence (AI) techniques. In the past years, efforts in AI and machine learning in the wind industry mostly concerned the development of tools for production optimization (wake steering optimization, yaw error optimization) or anomaly detection (early failure detection, underperformance detection). The current presentation will address a new application for AI and machine learning: the analysis of maintenance records. These records can be of distinct types (work orders, work tasks, cases, etc.) The common characteristic of these data sources is that they are unstructured, often in the form of free text; previous analytics were focused on time series data. This presentation will first define the challenge presents when working with text data. Then, an overview of the methods used to extract value from the maintenance records will be addressed. Work on maintenance records analytics is still at an early stage. However, results comparing the O&M work on wind farms will be presented. The presentation will conclude with an opening on what could be possible in a near future.