(121a) Machine Learning Approaches to Predict Density and Resiliency of Polyurethane Flexible Foams
- Conference: AIChE Spring Meeting and Global Congress on Process Safety
- Year: 2022
- Proceeding: 2022 Spring Meeting and 18th Global Congress on Process Safety Proceedings
- Group: Industry 4.0 Topical Conference
- Time: Tuesday, April 12, 2022 - 3:30pm-4:00pm
Polyurethanes are used in various applications such as coatings, adhesives, sealants, elastomers, thermoplastic, flexible and rigid foams. During the foaming process, the polyols and isocyanates are mixed with blowing agents, catalysts, surfactants, and additives2. These components' chemical structure and relative concentration determine the density, resiliency, and other mechanical and physical properties of the foam's targeted application. In addition, polyurethane reaction kinetics, dynamic phase separation of soft and hard segments, the nucleation and growth of the gas bubbles, the distribution of cell sizes, and open and closed cell windows also drive flexible foams' properties. Several science-based models have been developed to predict a wide range of descriptors for polyurethanes. However, due to the complexity of polyurethane foams components and rheo-kinetics, it is challenging to develop science-based models3 to predict mechanical and physical properties of foams, which led us to use the machine learning models.4 These science-based descriptors were fed into a wide range of AutoML platforms to select machine learning architectures best suited for the property of interest. Once we selected the best architecture, the hyperparameters of each machine learning architecture were optimized to develop the final model deployed in the Azure cloud for the user interface. In addition, prediction intervals were estimated for each predicted value using the bootstrap approach.5 This led us to develop and optimize models for density and resiliency of wide variety of foams with an accuracy (R2) higher than 0.9. Currently, efforts are directed towards implementing similar methodologies to improve the accuracy of the models that predict hardness values of flexible foams.
- Randal, D.; Lee, S. The Polyurethane Book; John Wiley & Sons, New York, 2002; Engels, H.-W.; Pirkl, H.-G.; Albers, R.; Albach, R. W.; Krause, J.; Hoffmann, A.; Casselmann, H.; Dormish, J. Polyurethanes: Versatile Materials and Sustainable Problem Solvers for Today's Challenges, Angew. Chem., Int. Ed. 52, 9422, 2013
- Baser, S.A. and D.V. Khakhar, Modeling of the dynamics of water and R-11 blown polyurethane foam formation, Polymer Engineering and Science, 34, 642, 1994; Harikrishnan, G. and D.V. Khakhar, Modeling the dynamics of reactive foaming and film thinning in polyurethane foams, AIChE Journal, 56, 522, 2010; Tesser, R., et al., Modeling of polyurethane foam formation, Journal of Applied Polymer Science, 92, 1875, 2004.
- Saunders, J. H. and K. C. Frisch. 1962. Polyurethanes, Vol. 1, Chemistry. New York, NY: Interscience Wiley, p. 264.
- Pugar, J. A.; Childs, C. M.; Huang, C.; Haider, K. W.; Washburn N. R., Elucidating the Physicochemical Basis of the Glass Transition Temperature in Linear Polyurethane Elastomers with Machine Learning, J. Phys. Chem. B 124, 9722, 2020
- Khosravi, A.; Nahavandi, S.; Creighton, D.; Atiya, A. F., Comprehensive Review of Neural Network-Based Prediction Intervals and New Advances, IEEE TRANSACTIONS ON NEURAL NETWORKS, 22, 1341, 2011