Adaptive multi-synchrosqueezing transform for compound fault diagnosis of gearbox bearings under non-stationary conditions

Rolling bearings operating under variable speed and load conditions usually generate strong non-stationary vibration signals, which make accurate extraction and separation of compound fault features difficult for conventional time-frequency methods. To address this issue, this study proposes an adaptive multi-synchrosqueezing transform (AMSST) for compound fault diagnosis of gearbox bearings. The proposed method introduces time-varying window parameters into the multi-synchrosqueezing framework according to a well-separated condition, so that the initial time-frequency representation can better match local signal characteristics. In addition, a window-parameter prediction strategy is used to reduce the computational burden of adaptive analysis. Based on the resulting adaptive time-frequency representation, fault-related components are more effectively concentrated and separated in the frequency domain. The performance of the proposed method is evaluated using both simulated compound fault signals and experimental gearbox bearing data, and is compared with several representative methods, including ASTFT, WSST, AFSST, and conventional MSST. The results show that AMSST achieves clearer fault-frequency localization, higher energy concentration, and stronger robustness to noise and parameter variations. These findings indicate that the proposed method provides an effective tool for time-frequency analysis of non-stationary vibration signals and for compound fault diagnosis of gearbox bearings.