Aiming at the problems of high acquisition frequency, heavy load of acquisition system and redundancy of monitoring data in massive condition monitoring data, the limitation of Shannon-Nyquist sampling theorem is broken through by low-dimensional projection of original signals, which greatly alleviates the problem of information overload caused by large data of equipment. Based on the five-layer architecture of PHM, this paper summarizes the existing achievements from the aspects of the concept of compressive sensing (CS) and its application in signal restoration and noise reduction, fault diagnosis, and degradation state identification, points out the existing problems in the existing research, and puts forward the corresponding solutions. This study can provide some reference for the research of compressed sensing.