Pattern Recognition and Damage Detection in Wind Turbine Monitoring Using Cepstral Coefficients

Traditional damage detection methods for dynamic systems like wind turbines often rely on basic classification approaches that identify whether a structure is damaged or intact. However, these methods fall short in providing detailed information on varying levels of damage or identifying emerging damage patterns, limiting their diagnostic utility. To address these challenges, this paper introduces a novel approach utilizing cepstral coefficients to improve the precision and efficiency of damage detection in wind turbine monitoring. Cepstral coefficients, widely recognized for their success in speech and speaker recognition, offer a powerful tool for structural health monitoring by capturing the unique vibrational characteristics of dynamic systems. When applied to wind turbines, cepstral coefficients serve as damage-sensitive features that can reveal subtle changes in the structural behavior, making them highly effective for monitoring systems subjected to complex loading conditions. This study validates the proposed approach using experimental data from the Aventa AV-7 wind turbine, owned by ETH Zurich. Nearing the end of its design life, this turbine has been used as a platform for research on system identification, operational modal analysis, and fault detection since 2020. The methodology simplifies the analysis, reduces computational complexity, and enhances the sensitivity to various damage scenarios by employing cepstral coefficients in the damage detection process.