EaLDL: Element-Aware Lifelong Dictionary Learning for Multimode Process Monitoring.

Area of Science:

• Industrial Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Traditional static monitoring models in industrial systems fail to adapt to evolving production environments and operating conditions.
• Retraining these models for new modes incurs significant computational complexity.
• Data-driven process monitoring is crucial for modern industry and artificial intelligence applications.

Purpose of the Study:

• To propose a novel multimode process monitoring method that addresses the limitations of traditional models.
• To enable continuous adaptation to new industrial process modes without compromising historical data representation.
• To achieve efficient and robust process monitoring in dynamic industrial settings.

Main Methods:

• Developed an element-aware lifelong dictionary learning (EaLDL) framework.
• Treated dictionary elements as fundamental units and measured their global importance.
• Introduced a novel surrogate loss to constrain dictionary element updates, balancing new mode learning and historical mode retention.

Main Results:

• The proposed EaLDL method demonstrated a favorable balance between learning new modes and retaining historical mode representations.
• Experimental results on numerical simulations and an industrial process validated the method's effectiveness.
• The EaLDL method showed insensitivity to initial points, yielding satisfactory results across various conditions.

Conclusions:

• The EaLDL method offers an effective solution for multimode process monitoring in dynamic industrial environments.
• This approach overcomes the computational complexity and adaptability issues of traditional static models.
• The method's ability to continuously update dictionaries ensures robust and adaptive industrial process monitoring.