Self-starting monitoring schemes for small-sample poisson profiles based on transfer learning

In profile monitoring applications, the small-sample phenomenon is a common challenge. Most existing studies address this issue by employing self-starting control charts. However, in practical manufacturing processes, a significant amount of data similar to but not identical with the small-sample process usually exists, presenting an opportunity to enhance real-time monitoring. To fully exploit such similar data and improve small-sample Poisson profile monitoring, a transfer learning-based method for parameter estimation of generalized linear models is proposed. Furthermore, a transferable source domain selection approach based on similarity is introduced to identify the most similar profiles from the dataset of similar processes, thereby preventing negative transfer effects. Leveraging the characteristics of small-sample profiles, the estimated parameters are utilized to construct self-starting T2 2 and MEWMA control charts for real-time monitoring of small-sample Poisson profiles. Simulation results demonstrate that the proposed transfer learning parameter estimation and self-starting profile monitoring method effectively detect shifts, outperforming traditional self-starting control charts, particularly in the detection of moderate and small shifts. Furthermore, the influence of parameters in the transferable source domain selection approach on monitoring performance is discussed. Finally, a real-data example of automobile warranty claims is presented to illustrate the efficacy of the proposed control charts.