Fault identification technology of wind turbine blade damage sound source localization and multimodal data fusion based on voiceprint monitoring

Abstract Fault Identification Technology of Wind Turbine Blade Damage Sound Source Localization and Multimodal Data Fusion Based on Voiceprint Monitoring proposes an innovative wind turbine blade damage identification method combining voiceprint monitoring and multimodal data fusion. Acoustic signals from multiple sensors around the blades are collected in real - time and analyzed alongside vibration, temperature, and pressure data. Wavelet transformation ensures data stability and accuracy. Voiceprint features extracted using Fast Fourier Transform (FFT) and Dynamic Time Warping (DTW) algorithms are compared with normal - working - condition signals to identify abnormalities and locate damaged areas. Kalman filtering fuses multimodal data to optimize health assessment. The Time Difference of Arrival (TDOA) algorithm and Convolutional Neural Network (CNN) are used for damage sound source localization and fault classification. Experimental results show the method achieves accuracy and recall rates of over 85% under various fault conditions, with a localization error within 0.15 m. A real - time monitoring system based on an embedded hardware platform is designed for online fault identification and timely warnings.