A soft sensor model based on long&short-term memory dual pathways convolutional gated recurrent unit network for...

Area of Science:

• Materials Science
• Chemical Engineering
• Artificial Intelligence

Background:

• Specific surface area is crucial for cement quality.
• Industrial cement data presents challenges like time-varying delays, non-linearity, and data redundancy, hindering accurate online monitoring.
• Existing models struggle to effectively capture complex spatio-temporal dynamics in cement production data.

Purpose of the Study:

• To develop an accurate soft sensor model for online prediction of cement specific surface area.
• To address the limitations of traditional models in handling complex industrial process data.
• To improve the precision and generalization capability of cement quality monitoring.

Main Methods:

• Introduction of parameters L and S into the Convolutional Gated Recurrent Unit (ConvGRU) network to modify internal linear constraints and enhance feature extraction.
• Design of dual spatio-temporal feature extraction pathways: a long-term memory enhancement pathway and a short-term dependence pathway.
• Fusion of features extracted from the dual pathways within the L/S-ConvGRU model for prediction.

Main Results:

• The proposed L/S-ConvGRU model demonstrated higher precision in predicting cement specific surface area compared to existing methods.
• The model exhibited superior generalization capability when trained on raw data from a cement plant.
• The dual pathway approach effectively captured long-term and short-term time-varying delay information.

Conclusions:

• The L/S-ConvGRU model offers an effective solution for accurate online monitoring of cement specific surface area.
• The modified ConvGRU architecture and dual pathway design significantly improve feature extraction and prediction accuracy.
• This approach provides a robust tool for enhancing cement quality control in industrial settings.