Aiming at the fault diagnosis of rotating machinery in industrial production, which faces the problems of signal noise, variable speed and other interferences that lead to inconspicuous fault features, and the traditional methods rely on manual experience for feature extraction and lack of robustness, the present study introduces a novel fault diagnosis approach utilizing the TimesNet model. This method takes the temporal multi-periodicity as the starting point, and the one-dimensional original signal is divided into multiple two-dimensional signals by different sizes of the period respectively. Subsequently, Inception-Attention modeling is applied to extract intricate time and space information. The experiments use the planetary gearbox dataset to verify the influence of different parameters and structures of TimesNet on the diagnostic effect, and carry out comparative tests with five time-series classification models, which provide some reference value for the application of TimesNet in fault diagnosis.