A hierarchical scheme for dynamic monitoring of multi-scale multi-mode systems

Complex chemical plant operations exhibit multi-scale dynamics and multi-mode characteristics. Existing methods typically address either multi-scale, dynamic, or multi-mode monitoring separately. This paper proposes a general hierarchical scheme for dynamic monitoring of systems with multi-mode dynamic behaviors. The core strength of the proposed method lies in its iterative procedure, which comprises two steps: dynamic pattern modeling and mode segmentation. Firstly, dynamic patterns across different modes are captured using latent vector autoregressive (LaVAR) modeling. In mode segmentation, data representing new dynamic patterns are filtered for the construction of the next LaVAR model, guided by two monitoring indices. The hierarchical structure sequentially extracts dynamic patterns, inherently dealing with unbalanced data common in industrial applications. Experiments are conducted to demonstrate the effectiveness of the proposed scheme for multi-mode dynamic system monitoring. • Proposing hierarchical dynamic monitoring for multi-scale, multi-mode systems. • Employing latent vector autoregressive modeling to capture data collinearity. • Performing offline modeling hierarchically for each mode. • Developing dynamic indicators for mode discrimination and fault detection. • Validating effectiveness through comparisons with machine learning methods.