A nonlinear full condition process monitoring method for hot rolling process with dynamic characteristic

Area of Science:

• Materials Science and Engineering
• Industrial Process Control
• Data Science

Background:

• Hot rolling processes (HRP) are complex industrial operations with distinct loaded and idle phases.
• Existing fault detection methods primarily focus on loaded conditions, neglecting critical idle periods.
• Effective monitoring of both conditions is crucial for overall process efficiency and safety.

Purpose of the Study:

• To develop a novel nonlinear process monitoring model for comprehensive hot rolling process (HRP) analysis.
• To address the limitations of previous research by incorporating monitoring of idle conditions.
• To enhance fault detection capabilities across the entire HRP cycle.

Main Methods:

• Definition of a dissimilarity index (DI) for condition identification.
• Establishment of a Support Vector Data Description (SVDD) model for idle condition monitoring.
• Application of t-distributed stochastic neighbor embedding (t-SNE) for nonlinear principal component (NPC) extraction.
• Utilisation of nonlinear slow feature analysis (NSFA) and nonlinear cointegration analysis (NCA) for dynamic and static variation analysis.

Main Results:

• The developed nonlinear model effectively monitors both loaded and idle conditions in HRP.
• NCA successfully revealed long-run dynamic relationships in nonstationary process data.
• NSFA effectively extracted latent temporal dynamics and static variations from stationary data.
• Validation on a real HRP confirmed the model's superior monitoring performance.

Conclusions:

• The proposed nonlinear full condition process monitoring model offers a significant advancement in HRP analysis.
• Integrating idle condition monitoring enhances the robustness and completeness of fault detection.
• The methodology provides a valuable tool for optimizing complex industrial processes like HRP.