A Fault-related Measurement Point Identification and Early Warning Method for Vertical Strip Tobacco Sorting Machine Based on KPCA-MSET

With the intelligent development of logistics equipment in the tobacco industry, the importance of equipment health status monitoring and fault warning systems in actual production has become increasingly prominent. This paper proposes an early warning method based on KPCA-MSET (Kernel Principal Component Analysis-Multivariate State Estimation Technique) for vertical strip tobacco sorting equipment. The method first analyzes the complex nonlinear relationships between multiple measurement points of the equipment and proposes an automatic extraction method for fault-related measurement points based on KPCA and a dynamic threshold adaptive mechanism. By using KPCA to extract the most representative key measurement points from the original data, data redundancy is significantly reduced while retaining information content, thereby lowering hardware requirements and computation time for large-scale deployment of comprehensive warning models. Subsequently, an MSET state estimation model is constructed based on the optimized measurement points, combined with an improved MAD adaptive threshold algorithm to achieve fault warning. Experimental results demonstrate that the proposed KPCA-MSET method can provide early warnings for potential equipment failures 2 to 14 hours in advance in a vertical tobacco sorting machine at a tobacco logistics center. Additionally, the key measurement points identified through contribution analysis offer effective support for fault diagnosis, significantly enhancing the practical value of early warnings and aiding subsequent diagnosis.