Main_file

  In view of blast furnace ironmaking process (BFIP), the co-existence of dynamics and nonstationarity causes it extremely difficult to build an effective fault detection model for securing safety and reliability. First, to explore the hybrid properties in the dynamic nonstationary system more explicitly, we established an inferential observation decomposition strategy by combining independent nonstationary, static, and dynamic components. Motivated by the foregoing analysis, we purposefully develop a novel method called adaptive dynamical interpretable analytic stationary subspace analysis (DiASSA) to discover BFIP dynamic consistent features. In particular, we provide an iterative modeling algorithm to obtain the optimal estimation in a closed region and effectively segregate the dynamic and static parts. Subsequently, the static part is further modeled by ordinary ASSA for constructing static consistent features and eliminating the interference of nonstationary information. Theoretical explorations, including fault detection sensitivity and quantified detection bounds, are presented immediately afterward. Case studies based on practical BFIP confirm that our proposal exhibits superior robustness to nonstationary disturbances, while the more sensitive fault detection capability proves its feasibility.