Time–frequency mode decomposition for wind turbine vibration monitoring under variable speed operation

Wind turbine vibration monitoring under variable speed operation requires separating nonstationary rotor-order components whose frequencies and operating intervals depend on operating state. These components can occupy local support regions in the short-time Fourier transform (STFT) plane rather than fixed spectral bands or continuous ridges. This study presents time–frequency mode decomposition (TFMD), which defines each mode as the time domain signal reconstructed by inverse STFT from one connected support region in the STFT plane. Signal decomposition thereby becomes the estimation of an unknown number of connected support regions. In a synthetic response with six operating states, TFMD separates the component active in each state and produces low reconstruction error without specifying the number of components in advance. In a controlled wind turbine blade strain experiment, the first decomposition reconstructs nine modes whose peak frequencies lie near the nominal once per revolution frequencies and whose energies are concentrated in the corresponding operating intervals. Residual decomposition further reveals the weaker twice per revolution harmonics of all nine operating states. These results support TFMD as a practical candidate for vibration analysis under variable speed operation, while offshore field use requires validation under environmental loading and with measured operating references.