Signal processing and feature engineering for automated Multi-Type damage assessment of wind turbine blades via Machine vision

Ensuring the structural integrity of wind turbine blades (WTBs) is paramount for operational safety and optimal energy generation, yet diagnosing the diverse range of potential damages incurred during service presents significant challenges for conventional inspection techniques. Reliable and automated damage assessment is crucial for effective condition monitoring. This paper presents an automated diagnostic approach leveraging machine vision, focusing specifically on robust signal processing and tailored feature engineering to address these challenges. The proposed pipeline integrates several key stages: (i) optimized image segmentation employing a combination of K-Means clustering and OTSU thresholding to effectively isolate potential damage regions even against variable backgrounds found in field images; (ii) refined morphological closing operations, incorporating a minimum connected domain area threshold (a), specifically designed to suppress imaging noise and artifacts common in UAV-captured WTB imagery while preserving true damage contours; and (iii) extraction of a carefully selected set of geometric and grayscale features, engineered to capture the distinct morphological and textural characteristics of different damage types. Based on these engineered features, a rule-based classifier achieves accurate localization and subsequent classification of five critical WTB damage types: sand holes, excoriations, cracks, oil stains, and breakage. Experimental validation performed on a comprehensive dataset of 1250 blade images, including samples augmented to simulate challenging real-world conditions, demonstrates the effectiveness of the approach, achieving an average detection and classification accuracy of 91.00%. This integrated machine vision system offers a valuable tool for automated WTB condition monitoring, facilitating more efficient, objective, and potentially earlier identification of various damage modes, thereby supporting informed maintenance strategies and enhancing wind farm reliability.