Damage detection of wind turbine blades via physics-informed neural networks and microphone array

Health monitoring of wind turbine blades is critical for ensuring power system stability and reliability. Conventional non-destructive testing methods often fail to detect early-stage damage due to environmental noise and insufficient sensor coverage, frequently resulting in unexpected failures and costly operational downtime. To address these challenges, this study proposes a novel damage detection method integrating physics-informed neural networks with a microphone array. The proposed method uses a Bluetooth speaker embedded within the blade cavity to generate controlled acoustic excitation, while the microphone array captures subtle, damage-induced signal variations. The proposed method incorporates acoustic energy conservation principles directly into the neural network's loss function, effectively harmonizing data-driven predictions with fundamental sound propagation physics. This physics-constrained methodology significantly reduces training data requirements while enhancing result interpretability and reliability. Numerical simulations demonstrate the method's superior performance, achieving an R2 score of 0.91 and maintaining error indices below 0.5 % using only 5 % of the training data. The proposed method exhibits robust performance even in low signal-to-noise ratio environments and across various frequency excitation scenarios. This work establishes a practical pathway toward microphone array-based damage detection in operational wind farms, ultimately advancing sustainable wind energy utilization and reliability.