“…Although transfer learning has achieved promising results in fault diagnosis of machinery, the methods commonly have the following shortcomings: first, most of these them still need a certain amount of labeled data, for example, reference [10] and [11] require more than 10 target training samples to achieve effective recognition accuracy; Second, we need to do a lot of preprocessing work, such as to extract features [12] from spectrum data rather than the original vibration data; and finally, these methods only transfer the simulation experiment data set to another simulation experiment data set [13], and the speed, loading, and fault degree of these data sets changed slightly, so the generalization ability of these methods are limited. To deal with the above-mentioned limitations, a new deep transfer learning network, named the transferred discriminator network (TD), is proposed for fault diagnose of rolling bearings in this paper.…”