Modified neuro-fuzzy system for online classification of wind turbine blade defects

This study considers the current problem of the automated detection and classification of defects in wind turbine blades in the context of the development of modern wind energy. A critical analysis of the application of existing diagnostic methods, particularly convolutional neural networks and LIDAR technologies, was conducted, emphasizing their advantages and limitations. As an alternative solution, an adaptive neuro-fuzzy system based on the modified architecture of the Wang–Mendel system is used, which is optimized for classification tasks under conditions of limited data and nonstationary information flows. A detailed description of the system structure, features of its online training, and the implementation of a modular architecture considering the effective processing of features and preservation of the interpretability of the results are presented. The experimental results proved the high classification accuracy (94.6%) of the proposed system, with its faster processing time (2.2 times) compared to classical approaches, making it a promising candidate for implementation in practical systems for the technical monitoring of wind turbines.