Modelling the effect of base component properties and processing conditions on mixture products using probabilistic,...

Area of Science:

• Materials Science
• Chemical Engineering
• Data Science

Background:

• Developing new materials by mixing components is common but predicting mixture properties is difficult without understanding underlying physics.
• Current modeling approaches often rely on assumptions of ideal mixing, limiting their applicability.

Purpose of the Study:

• To propose a new data-based methodology for predicting the properties of mixture-based materials.
• To develop a model that integrates base component properties, mixing proportions, and processing conditions.
• To estimate prediction uncertainty and incorporate expert knowledge.

Main Methods:

• Utilized probabilistic, knowledge-guided artificial neural networks (ANNs).
• Developed a model to jointly predict properties based on component characteristics, mixture ratios, and manufacturing conditions.
• Incorporated monotonicity constraints for expert knowledge integration.

Main Results:

• The methodology accurately predicts pharmaceutical tablet quality attributes (e.g., mass variation, tensile strength).
• The model overcomes limitations of previous approaches by not assuming ideal mixing.
• The approach allows for the estimation of aleatoric uncertainty in predictions.

Conclusions:

• The proposed data-driven methodology effectively predicts properties of mixture-based materials, including pharmaceutical products.
• This approach enhances material design by providing accurate predictions without requiring deep prior knowledge of physical phenomena.
• The integration of expert knowledge and uncertainty estimation improves model reliability and applicability.