Multi-class data augmentation and fault diagnosis of wind turbine blades based on ISOMAP-CGAN under high-dimensional imbalanced samples

The application of deep learning algorithms in fault diagnosis has become increasingly widespread. However, complexity of wind turbine blade monitoring data and scarcity of fault samples severely hinder diagnostic accuracy in practical industrial scenarios. To address this challenge, a multi-class data augmentation and fault diagnosis method is devised. This method employs Isometric Feature Mapping (ISOMAP) and Conditional Generative Adversarial Nets (CGAN) to reduce dimensionality, enhance icing and cracking sample amount and improve diagnosis accuracy. An ISOMAP dimension reduction model is formulated to capture overarching feature distribution of icing and cracking samples, further a CGAN network model is designed to generate blade fault samples in low-dimensional manifold space. Experimental comparisons under multiple classifier models with SMOTE and ADASYN algorithms demonstrate that proposed method effectively addresses issues of high dimensionality and multi-class sample imbalance, significantly enhancing classifier performance. Comparative results with spatial-temporal multi-learner network method and ResDenIncepNet-CBAM algorithm further highlight superiority and robustness of proposed approach.