An Intelligent Failure Mode Identification Model for Wind Turbines

Abstract This article proposes an intelligent failure mode identification model to automatically identify failure modes of wind turbines from textural maintenance records collected from wind farms. Initially, an identity (word) prediction model is constructed based on bidirectional encoder representations from transformers and the conditional random field model to predict identities reflecting the failure symptoms of wind turbines. Subsequently, a failure mode prediction model is created to identify failure modes of offshore wind turbines with the assistance of those of onshore wind turbines. An adaptive resampling mechanism is then constructed to reconstruct the datasets as a basis to highlight the failures with low frequencies. Ultimately, a graph construction model is presented to display the identified failure modes, along with their corresponding components, in a graphical manner. Maintenance records from onshore and offshore wind turbines validate the effectiveness of the proposed method. Overall, the proposed method transforms the failure data analysis from a human-based to a machine-based approach, significantly reducing the need for human interaction in textural failure data analysis and supporting the operation and maintenance of wind turbines in the big data era.