Rolling bearings are considered an important central component in rotating machines and fault diagnosis for rolling bearings is crucial in condition-based maintenance to reduce the complexity of different kinds of faults. A prognostic algorithm to classify various rolling bearing faults is proposed, consisting of four phases. The stacked denoising auto-encoder which can be filtered so the noise of large number of mechanical vibration signals is used for deep learning structure to extract the characteristics of the noise. Unsupervised pre-trainning method, which greatly simplifies the process of traditional manual extraction, is used to process the depth of the data automatically. Furthermore, the aggregation layer of SDA(denoising auto-encoder) is proposed to get rid of gradient disappearance in deeper layers of network, mixing superficial nodes’ expression with the deeper layers, avoiding the insufficient express ability in deeper layers. The PCA(principal component analysis) was used to extract different features to make classification processing. Through the experimental data of this method and from the comparison of results show that, the new method for rolling bearing fault classification reached 97.02% correct rate therefore compared to other algorithms, achieved better classification rate.