Defect detection in wind turbine blades applying Convolutional Neural Networks to Ultrasonic Testing

Abstract

The significance of wind-turbine blade safety operation has risen in the context of recent advances in wind energy generation. In this context, Non-Destructive Inspection Technologies (NDT), in particular those derived from Ultrasonic Testing (UT) methods, have proven to be key. Non-destructive evaluation (NDE) analysis has traditionally been performed by a qualified inspector who interprets the acquired signal. However, the emerging digital revolution has brought with it many advances in Artificial Intelligence (AI) and has demonstrated its potential in the field of NDE. AI has allowed to automate and improve traditional techniques in the tasks of data pre-processing, defect detection, defect characterization, and property measurement. Moreover, it has proven to be highly valuable in situations where it is not possible to apply traditional gate methods.

In this paper, the feasibility of using Deep Learning (DL) techniques for the detection of defects in wind-turbine blades (in the Cap zone and in the Cap-Web zone) is analyzed. For this purpose, supervised learning techniques have been used and three case studies were analyzed: two-class classifications for Cap zone, two-class classifications for Cap-Web zone, and four-class classifications have been performed. Several Convolutional Neural Network (CNN) architectures have been proposed, reaching 90% accuracy in all three case studies. These results lay the groundwork for the initial steps in applying AI techniques during the automated inspection of complex wind blade components.