A new soft sensing method based on serial-parallel GRU with self-attention mechanism for complex multi-unit...

Area of Science:

• Manufacturing Process Optimization
• Artificial Intelligence in Industry
• Soft Sensor Technology

Background:

• Traditional Key Performance Indicator (KPI) testing is time-consuming and costly.
• Digital transformation necessitates timely and accurate KPI prediction for manufacturing.
• Existing methods fail to provide effective real-time production guidance.

Purpose of the Study:

• To develop an advanced soft sensor model for online KPI prediction.
• To address the limitations of traditional destructive testing methods.
• To improve the efficiency and accuracy of KPI monitoring in industrial processes.

Main Methods:

• Proposed a novel Serial-Parallel Gated Recurrent Unit with Self-Attention (SPGRU-SA) soft sensor model.
• Employed SPGRU to extract dynamic features from multi-unit processes.
• Utilized a self-attention mechanism to weigh static and dynamic features for correlation analysis.

Main Results:

• The SPGRU-SA model demonstrated accurate online prediction of KPIs.
• Effectively captured both dynamic and static process features.
• Validated performance on hot rolling strip mill and Tennessee Eastman processes.

Conclusions:

• SPGRU-SA accurately predicts KPIs in complex, multi-unit industrial settings.
• The model offers a viable alternative to costly and slow traditional testing.
• Enhances real-time decision-making and reduces manufacturing losses.