Ensemble diagnosis method based on transfer learning and incremental learning towards mechanical big data

Abstract The advancement of big data in industry has prompted the development of intelligent fault diagnosis technologies. Nevertheless, several challenges of big data are notable, such as extracting valuable features from original data, exploring general methods to accommodate data with different distributions, and handling streaming data. This paper proposed an ensemble learning method to handle these challenges. Firstly, a convolutional neural network (CNN) is trained by source domain directly. Then, a transfer learning method with pre-trained CNN is employed to extract valuable information from quite different datasets, which is useful to utilize knowledge of CNN learned in previous object and reduce modeling time. At last, an online learning method named incremental support vector machine is applied to classify various conditions. Five mechanical fault datasets acquired from three test rigs are used in the case study. Comparison study shows that the proposed method can be applied in mechanical fault diagnosis under big data era.