Search for a command to run...
Early detection is the major emphasis of predictive maintenance, which is essential for minimizing downtime and guaranteeing efficient operation. In this situation, identifying defects in wind turbine blades is crucial. In order to forecast the severity of these fissures by classifying them into multiple groups this study presents a novel idea called the ANN-Enhanced RocketStack Ensemble Architecture. In contrast to conventional stacking frameworks, RocketStack presents a multi-phase refinement procedure created especially to capture the non-linear degradation gradients of blade structural health. To tackle class imbalance and complex blade degradation, the framework incorporates a special hybrid feature space by combining temporal feature engineering to map operational trends, employing an Isolation Forest-derived novelty feature to quantify anomalies, using RFE to reduce redundancy, and applying custom Targeted Undersampling (TUS) strategy with sample weighting to improve minority-class learning. The RocketStack ensemble combines ANN SVC and reduced-capacity XGBoost base classifiers whose probabilistic outputs are sequentially refined using L1-regularized Logistic Regression and a Random Forest meta-classifier to achieve stable and generalizable crack severity prediction. When the frameworks effectiveness is first evaluated using an industrial dataset from companies GE and Siemens that is accessible to the public and it attains an outstanding accuracy of 98%. After that it was applied to a brand-new private dataset with more challenging domain-specific problems and it still performed admirably achieving an accuracy of 85%. Performance is evaluated using accuracy, class-wise F1-score, macro recall, moreover confusion matrices reveal class imbalances. This comparison highlights the actual challenges in predicting fracture severity in industrial situations while also demonstrating the theoretical durability of the RocketStack architecture. These results show that the TUS-enhanced RocketStack Ensemble offers precise and scalable crack severity estimates in wind turbine components, promoting AI-driven industrial maintenance advancements and guaranteeing sustained production of sustainable energy.