Purpose <p>The purpose of this study is to solve the problem of how to effectively separate and extract different state features under the influence of coupling of different state features and interference noise.</p> Methods <p>The multi-layer scanning Ramanujan decomposition (MLSRD) method is proposed in this paper. MLSRD method adopts order statistics filter (OSF) method to realize the adaptive division of frequency bands to avoid the destruction of state feature structure. Meanwhile, sparse energy ratio (SER) index is defined in this paper to achieve accurate evaluation of state features corresponding to different frequencies and avoid missing detection of state types.</p> Results <p>The results of vibration signal analysis of composite fault of rolling bearing show that the method proposed in this paper successfully separates the composite fault features and accurately extracts the fault feature frequency information.</p> Conclusion <p>The analysis is to realize the compound fault diagnosis of rolling bearing by Ramanujan theory. These findings have potential application prospects in condition monitoring of high-end equipment.</p>