Collaboration Transfer Enabled Standard Rank Control Chart for Cross-Domain Incipient Fault Detection of Bearings With Limited Historical Data

The standard rank control chart, a type of statistical process control (SPC), has been successfully employed in detecting incipient faults in many areas due to its self-starting nature. However, limited historical data as well as varying working conditions hinder its performance in fault detection. To address these issues, a collaboration transfer-enabled standard rank control chart (CTESRCC) is proposed, which aims at boosting the performance of detecting by extracting shared information and target-specific characteristics through transfer learning (TL). First, feature extraction is conducted on vibration data through multitime phase space warping (MPSW), followed by the construction of the corresponding standard rank control charts for the source and target domains, respectively. Subsequently, a bridge domain that links target and source domains is constructed through canonical correlation analysis (CCAs)-enhanced self-taught clustering (STC) to extract baseline-controlled distribution as shared information across domains. Furthermore, to enhance adaptability for the target task, an ensemble transfer strategy is developed by incorporating deviations of newly observed objects in the target domain from the baseline-controlled distribution with transitive knowledge decay. Finally, Monte Carlo simulation is used to establish control limits based on the reconstructed ensemble standard rank control chart. Experimental results show that CTESRCC attains an average missed alarm rate (MAR) of 0.0033 and a false alarm rate (FAR) of 0.036, corresponding to 50% and 85% reductions relative to the comparative method, while maintaining a fault detection rate of 99.7%. These verify the capability of CTESRCC in detecting incipient faults even when historical data are scarce.