The traditional one-dimensional Convolutional neural network is difficult to realize adaptive attention to fault characteristics, resulting in lower accuracy and generalization ability of the network model. In order to solve the problem, this paper proposes a hybrid multi attention network model (HMA1DCNN) that includes channel attention mechanism (CAM), time attention mechanism (TAM), and multi-scale attention mechanism (MSAM). The model is validated using a planetary gearbox gear and bearing fault dataset from a circulating water pump (CRF) unit test bench. The results show that the model can accurately classify different health states of planetary gearboxes under strong noise backgrounds. Compared with other models, the accuracy of fault identification has been improved.