Inferential estimation of polymer quality using bootstrap aggregated neural networks.

Area of Science:

• Polymer Science and Engineering
• Artificial Intelligence in Chemical Processes
• Computational Chemistry

Background:

• Accurate polymer quality estimation is crucial for batch polymerization reactors.
• Traditional methods often struggle with limited training data, impacting model robustness.
• On-line measurements provide valuable data for inferential modeling.

Purpose of the Study:

• To develop an inferential estimation method for polymer quality in batch reactors.
• To enhance the accuracy and robustness of neural network models using bootstrap aggregation.
• To apply the developed technique to a batch methyl methacrylate polymerization reactor.

Main Methods:

• Utilizing bootstrap aggregated neural networks (BANNs) for enhanced model performance.
• Employing bootstrap re-sampling with replacement to create multiple training datasets.
• Developing individual neural networks and combining them into a BANN.
• Using principal component regression (PCR) to determine optimal weights for combining networks.
• Obtaining confidence bounds for predictions via bootstrapping.

Main Results:

• Successfully applied bootstrap aggregated neural networks to estimate polymer quality.
• Demonstrated enhanced accuracy and robustness of the BANN models compared to single networks.
• Validated the technique on a simulated batch methyl methacrylate polymerization reactor.
• Achieved reliable estimation of number average molecular weight and weight average molecular weight.

Conclusions:

• Bootstrap aggregated neural networks offer a robust approach for inferential polymer quality estimation.
• The proposed method effectively addresses challenges posed by limited training data in batch polymerization.
• This technique provides accurate predictions and confidence bounds, valuable for process control and optimization.