A novel acoustic emission signal segmentation network for bearing fault fingerprint feature extraction under varying speed conditions

The acoustic emission (AE) technique is known for its sensitivity to early damage and is well-suited for online condition monitoring of rolling element bearings (REBs) in various industrial application scenarios. Nonetheless, identifying weak faults under varying speed and strong background noise conditions still remains challenging. In addition, the comprehensive modelling of AE signal from faulty REB in electromechanical systems is still a pending issue. In light of this, a well-considered model is firstly developed for the AE signal of faulty REBs in this work. After that, a novel bearing fault diagnosis framework based on semantic segmentation networks is devised. Precisely, the proposed framework consists of three main components: a preprocessing step depending on the signal segmentation algorithm, a diagnosis step using fault fingerprint mapping, and a postprocessing evaluation step supplemented by a density peak clustering (DPC) approach. We evaluate the presented procedures through simulation analysis and an experimental case under varying speed conditions. Meanwhile, the comparison with the original threshold-based fault fingerprint recognition algorithm is also conducted. The comprehensive results demonstrate the efficacy of identifying fault-associated fingerprint feature (FPF), indicating that the proposed framework holds promise for condition monitoring.