Weak fault detection with a two-stage key frequency focusing model

The fault diagnosis of early failure bearing can discover the potential danger in the mechanical equipment in time. It remains a great challenge due to the noise interference. Although many diagnosis methods have been proposed, the characteristics of signal have not yet been fully investigated, which leads to unsatisfactory diagnosis results. To solve this problem, a weak fault detection method with a two-stage key frequency focusing model is designed. Translation invariance in time domain and translation variance in frequency domain are systematically considered. The effectiveness is verified on four constructed datasets. The results show that the designed network has the best comprehensive performance comparing to the state-of-the-art methods.