摘要：AbstractConvolutional Neural Network (CNN) has been widely used in bearing fault diagnosis. Generally, satisfying results can be obtained when training data and test data come from the same domain. However, in practice, a CNN trained with data from one working condition usually has to cope with data from other different working conditions, which inevitably leads to CNN performance degradation. To resolve this problem, three different ways, namely Multi Convolutional Layer (MCL), Data Augmentation (DA) and Signal Concatenation (SC), are proposed in this paper, and then adopted separately or in cooperation to enhance CNN’s ability in working condition adaptation. Experimental results confirm that all three ways can effectively raise CNN’s performance when transferred across working conditions, and the combination of all three methods performs better than the individual one.
