<p>The health status of sliding bearings, as the foundation of advanced manufacturing technology, directly impacts the advancement of this technology in industrial development. However, research on bearing fault diagnosis predominantly focuses on rolling bearings, with relatively few studies addressing sliding bearing fault diagnosis. Existing methods are often limited to specific operating conditions, while the complex and variable conditions in practical engineering restrict their applicability. For early-stage weak fault diagnosis, the lack of prominent sample features makes it challenging for deep learning models to extract effective features, resulting in relatively low prediction accuracy. Therefore, in response to the complex and variable nature of actual operating conditions and the subtle characteristics of early-stage faults, we proposed a transformer-based fault diagnosis framework with multi-sensor fusion (transformer with multi-sensor fusion, MSFT) grounded in the improved variational mode decomposition (IVMD) signal processing method. Additionally, common sliding bearing-rotor system fault types were designed and constructed for relevant fault data collection, addressing the current deficiency in sliding bearing fault datasets. Experiments demonstrate that this method meets the fault diagnosis needs for various degrees of fault mixing under multiple operating conditions, making deep learning fault diagnosis methods more applicable to complex practical engineering scenarios.</p>