Health Assessment and Prognosis of Offshore Wind Turbine Blades Under Variable Operating Conditions via Optimized Bi-LSTM

Blades are essential components for harnessing wind energy in turbines. Their failure can result in severe accidents, costly maintenance, and high replacement expenses. This study introduces an integrated model combining principal component analysis (PCAs), Tree Seed Optimization [tree-seed algorithm (TSA)], and bidirectional long short-term memory (BiLSTM) for identifying blade damage and predicting health status. By integrating vibration mechanics, signal processing techniques, and machine-learning algorithms, we have developed a robust and generalizable detection tool. PCA is employed to extract multiple fault indicators and determine the weights of significant indicators, forming a health index that quantifies the extent of blade damage. The critical feature parameter matrix serves as input to construct the PCA-TSA-BiLSTM prediction model, which accomplishes short-to-medium-term predictions of the health index by leveraging TSA’s global search capability and BiLSTM’s superior ability to capture nonlinear temporal relationships. The effectiveness of fault indicator extraction is validated through a time-frequency comparative analysis of healthy and damaged states. Comparative analysis with other LSTM-based combination algorithms confirms that the proposed prediction model exhibits higher accuracy, with a root-mean-squared error (RMSE) of less than 0.064 and a mean absolute error (MAE) of less than 0.409. Through exponential fitting, the overall degradation trend of the system was obtained, and four health levels were delineated. The health index below 0.3 indicates a critical fault. This facilitates real-time monitoring of blades and forecasting of future health index trends, thereby supporting decision-making for maintenance strategies.