Time–Frequency Analysis for Planetary Gearbox Fault Diagnosis Based on Improved U-Net++

Planetary gearbox plays an important role in many industrial fields and is also a vulnerable component. It is of great significance to develop the time–frequency analysis method of planetary gearbox for ensuring the safe operation of equipment. To analyze the fault characteristics quickly and automatically in time–frequency information, this paper proposes a time–frequency analysis method based on improved U-net++. In the proposed method, the modified U-net++ is used to compress and expand the vibration time–frequency data generated by the normalized S transform, and the original overlap-tile strategy is adjusted. In addition, Tversky loss is introduced into the U-net++ model as an optimization objective. The improved U-net++ is evaluated on the gearbox vibration dataset of in-service wind turbines. The experimental results show that the Dice coefficient of the feature area analysis reached 0.949. The convergence speed and calculation efficiency are also significantly improved, which proved the effectiveness and progressiveness of the improved U-net++ in the planetary gearbox time–frequency analysis.