Transfer Fault Diagnostics of Planetary Gearbox from Steady to Variable Operating Conditions

Planetary gearbox, as an important role in mechanical equipment, its reliability and safety directly impact the comprehensive performance of mechanical equipment. Intelligent fault diagnosis (IFD) of planetary gearbox plays a critical role in saving economic costs and extending the lifespan of mechanical equipment. However, in practical industrial scenarios, mechanical equipment usually operates at steady speeds under healthy conditions, making IFD under variable operating conditions challenging owing to the lack of fault data. To address this challenge of IFD from steady to variable operating conditions, a novel fast-adaptive angular domain resampling transfer network (ADRTNet) is proposed in this paper. The proposed ADRTNet can achieve more robust intelligent diagnostics through angular domain resampling, a fast adaptive network structure, and a parameter transfer strategy. First, the vibrational data feature is enhanced by angular domain resampling to promote the quality of the data and alleviate the effect of varying speeds. Second, a deep convolutional neural network (DCNN) for IFD is constructed and pre-trained using the available fault data collected at steady speeds. Then, the pre-trained DCNN is further fine-tuned with the parameter transfer strategy on the limited variable operating conditions dataset. Finally, the well-trained ADRTNet will be applied to diagnose the test datasets under variable operating conditions. Experimental results have validated that the proposed ADRTNet possess superior fast-adaptability, generalization performance, and diagnostic accuracy from steady to variable operating conditions, and outperforms several representative IFD methods for planetary gearbox.