Extremely Low-Speed Bearing Fault Diagnosis Based on Raw Signal Fusion and DE-1D-CNN Network

IntroductionAn effective data-driven bearing fault diagnosis has been moving away from traditional reliance on statistical features toward a focus on raw time-domain analysis. This transition is driven by the need to preserve the critical information contained within the signals.PurposeIn machinery maintenance, a challenge arises when attempting to detect bearing faults under extremely low operating conditions. This is due to the minimal interaction between the rolling element and the fault at such speeds. In response to this, the research aims to tackle the issue of bearing fault detection under these low-speed operating conditions by introducing an advanced fault diagnosis approach.MethodThe proposed method involves the utilization of acoustic emission and vibration sensors to capture the vibration behavior of the bearing component at low operating conditions. Importantly, a data fusion strategy is adopted to fuse the raw signals obtained from both sensor types. These fused signals are then directly input into the novel one-dimensional optimized convolutional neural network (1D-CNN). To further enhance the model's performance, a differential evolution algorithm is leveraged for the optimization of its hyperparameters. To validate the effectiveness of the approach, a preliminary analysis is initially conducted using the CWRU dataset, comparing the diagnosis accuracy of the proposed model to existing literature results. Subsequently, the model's performance is evaluated using experimental data collected under low operating speed conditions, spanning a range from 48 rpm (extremely low speed) to 300 rpm (low speed).ResultThe results showcase the remarkable capability of the DE-1D-CNN model to diagnose the presence of bearing faults under low-speed conditions, with the highest achieved accuracy standing at an impressive 100%. Through the judicious optimization of hyperparameters, the proposed 1D-CNN model proves its competence in bearing fault diagnosis even when the signal was collected under challenging low-speed operating conditions.