Bearing fault diagnosis with compressed data based on two-stage matching pursuit

In recent years, many scholars have devoted their energies to research remote bearing condition monitoring and on-board bearing fault diagnosis, where big data transmission is usually necessary. However, in some cases, the bandwidth of the wireless communication is limited, and the volume of transmitted data is limited by bandwidth. Therefore, raw data are usually compressed before transmission. Different to traditional data compression method, in this study, bearing data is compressed based on Compressive Sensing theory. With this technique, data could be collected and compressed simultaneously, and then it could be recovered in remote side. Unfortunately, when using the compressed data, the accuracy and stability of fault diagnosis could be discounted. To solve this problem, this paper proposes a two-stage matching pursuit algorithm for on-board bearing fault diagnosis. First, we set up a complete database using historical bearing data of all working conditions and fault patterns. Second, Gaussian random matrix was applied to compress the target bearing data and then similarity pursuit was implemented to get the Similar Samples. Third, the Similar Samples were used to construct dictionary matrix. Then, matching pursuit algorithm based on Compressive Sensing was introduced to reconstruct the target sample. Finally, classification reconstruction was conducted to recognize the fault pattern of the target bearing data. In addition, some experiments were conducted to demonstrate that proposed method is working well for bearing under different working conditions.