Sintering Quality Prediction Model Based on Semi-Supervised Dynamic Time Feature Extraction Framework.

Area of Science:

• Materials Science
• Chemical Engineering
• Process Control

Background:

• Real-time monitoring of key quality variables in sintering is challenging due to data limitations.
• Conventional soft sensor models perform poorly with insufficient labeled data.

Purpose of the Study:

• To propose a novel semi-supervised dynamic feature extraction framework (SS-DTFEE) for improved real-time process guidance.
• To enhance the prediction accuracy of soft sensor models in industrial sintering processes.

Main Methods:

• Utilizing a dynamic feature extraction (DTFEE) model to extend and extract time-series features.
• Designing a weighted bidirectional LSTM (BiLSTM) unit for latent variable extraction.
• Implementing an encoder-decoder model for unsupervised pre-training and a fine-tuning strategy for prediction improvement.

Main Results:

• The SS-DTFEE framework demonstrated significant improvements in prediction accuracy for FeO content estimation in actual sintering.
• The proposed method outperformed traditional soft sensor models in estimating critical process variables.

Conclusions:

• The developed semi-supervised framework effectively addresses the challenge of limited labeled data in sintering.
• SS-DTFEE offers a robust solution for real-time quality control and process optimization in industrial applications.