Application of tensor singular spectrum decomposition based on Morlet wavelet ridges in weak fault diagnosis of bearings

Rolling bearing faults are a major source of rotating machinery failures, and detecting their weak early-stage signatures is challenging due to heavy noise interference. Adaptive variable-sampling tensor singular spectrum decomposition has recently shown promise for fault diagnosis, but its interpolation complementation may introduce spurious components, weakening feature extraction. To overcome this limitation, we propose a collaborative diagnostic method that integrates Morlet wavelet ridge extraction with adaptive tensor singular spectrum decomposition (MT-SSD). Wavelet ridges and their maximum mutation points are first used to identify suspicious fault-related frequency bands, guiding feature-group selection. A third-order tensor model is then constructed from multichannel signals via trajectory matrix formation and tensor order-preserving multiplication.With adaptive sampling and interpolation-based inversion, MT-SSD efficiently extracts fault-related components within the selected bands. Experimental and engineering evaluations verify that MT-SSD effectively identifies early weak fault features under strong noise. Compared with VMD, EMD, and classical T-SSD, MT-SSD offers improved fault-frequency localization, stronger noise suppression, and competitive computational efficiency, providing a reliable tool for early bearing fault diagnosis.