Deep Semi-Supervised Just-in-Time Learning Based Soft Sensor for Mooney Viscosity Estimation in Industrial Rubber...

Area of Science:

• Materials Science
• Chemical Engineering
• Data Science

Background:

• Soft sensor technology is crucial for real-time quality monitoring in industrial rubber mixing.
• Developing high-performance soft sensors is challenging due to feature extraction issues and insufficient labeled data.

Purpose of the Study:

• To develop a novel deep semi-supervised just-in-time learning-based Gaussian process regression (DSSJITGPR) model for accurate Mooney viscosity estimation.
• To address the limitations of traditional soft sensor methods in feature extraction and data scarcity.

Main Methods:

• A stacked autoencoder was used for latent feature extraction from historical process data.
• An evolutionary pseudo-labeling approach generated high-confidence pseudo-labeled data to expand the modeling database.
• A semi-supervised just-in-time Gaussian process regression model was constructed for online Mooney viscosity estimation.

Main Results:

• The DSSJITGPR model demonstrated superior performance in Mooney viscosity prediction compared to traditional methods.
• The method effectively extracted latent features and handled the scarcity of labeled data.
• Validation through industrial rubber-mixing process data confirmed the model's effectiveness and superiority.

Conclusions:

• The developed DSSJITGPR model offers significant advantages for soft sensor applications in rubber manufacturing.
• This approach provides a robust solution for real-time quality variable estimation, particularly in data-scarce environments.
• The study highlights the potential of integrating deep learning, semi-supervised learning, and just-in-time learning for advanced process monitoring.