This paper has addressed a data fusion approach for bearing fault diagnosis based on NCA and CHMM. For bearing or machinery condition monitoring, many indices can be used. NCA-based FE approach is proposed to reduce the dimensionality of the original features and extract useful information. Furthermore, with the extracted features by NCA, CHMM is developed for each state, thus a CHMM database can be built. New test samples can be classified with these trained CHMMs. The results of the first experiment illustrate that NCA-CHMM can successfully distinguish different bearing fault type. In the second experiment, early weak fault in bearing is difficult to detect and diagnose. The NCA-CHMM approach can successfully recognize the healthy, early fault, degraded and failure stage of bearings. Compared with other
existing methods, the proposed approach performs better in both experiments. In total, the proposed NCA-CHMM approach can fuse multichannel data and improve the diagnosis results of bearing or machine