Rigid polyurethane foams are polymeric materials, produced through simultaneous expansion and polymerization reactions, mainly used for thermal insulation purposes. The formulation of chemicals with which these foams are realized has a great impact on the final product quality, determined through time profiles of foam growth velocity and temperature gradients. The aim of the Thesis is the development of multivariate machine learning methodologies for both the characterization of the foams and the formulation of the polyurethane systems. First of all, methodologies for the study of the expansion process are developed by means of a multi-way principal components analysis (PCA, Wise e Gallagher, 1996, Camacho et al., 2008) and a pattern recognition technique based on similarity indices (Ottavian et al. 2012) applied to time profiles of growth velocity and temperature gradient. These methodologies guarantee a in-depth understanding of the polyurethane systems and a science-based method which demonstrate to be much more efficient than the analytical state-of-the-art methods, which are mostly based on operators’ experience. Furthermore, a reduction of 50-75% of the duration of the product quality laboratory analysis is proposed. Finally, based on the inversion (Tomba et al., 2016) of a partial least squares regression methods (PLS, De Jong, 1993) calibrated on data from designed experimental campaigns, a tool is developed to suggest the most appropriate formulation which guarantees the desired final product quality in terms of growth velocity and foam’s height time profiles.
Le schiume poliuretaniche rigide sono materiali polimerici, prodotti attraverso reazioni simultanee di espansone e polimerizzazione, principalmente usati nel campo dell’isolamento termico. La formulazione delle sostanze chimiche con le quali queste schiume vengono prodotte ha un impatto notevole nella qualità del prodotto finale, determinata attraverso profili temporali di velocità di crescita e gradiente di temperatura. Lo scopo della Tesi è l’introduzione di tecniche multivariate di machine learning per la caratterizzazione e la formulazione dei sistemi poliuretanici. Innanzitutto, sono state sviluppate metodologie per lo studio del processo di espansione attraverso l’analisi delle componenti principali (PCA, Wise e Gallagher, 1996, Camacho et al., 2008) e una tecnica di pattern recognition basato su fattori di similarità (Ottavian et al. 2012) applicate ai profili temporali di velocità di crescita e gradiente di temperatura. Queste metodologie garantiscono una profonda comprensione dei sistemi poliuretanici e un metodo scientifico molto più efficiente dei metodi analitici allo stato dell’arte, quasi esclusivamente basati sull’esperienza dell’operatore. Inoltre, viene proposta una riduzione del 50-75% della durata dell’analisi sperimentale della qualità del prodotto. Infine, basandosi sull’inversione (Tomba et al., 2016) di metodi di regressione lineare ai minimi quadrati parziali (PLS, De Jong, 1993), calibrato su dati provenienti da campagne sperimentali pianificate, è stato sviluppato uno strumento in grado di suggerire la formulazione più appropriata per garantire la qualità del prodotto finale desiderata in termini di profili temporali di velocità di crescita e altezza della schiuma.
Formulazione di schiume poliuretaniche rigide mediante tecniche multivariate di machine learning
MATTIELLO, MASSIMO
2022/2023
Abstract
Rigid polyurethane foams are polymeric materials, produced through simultaneous expansion and polymerization reactions, mainly used for thermal insulation purposes. The formulation of chemicals with which these foams are realized has a great impact on the final product quality, determined through time profiles of foam growth velocity and temperature gradients. The aim of the Thesis is the development of multivariate machine learning methodologies for both the characterization of the foams and the formulation of the polyurethane systems. First of all, methodologies for the study of the expansion process are developed by means of a multi-way principal components analysis (PCA, Wise e Gallagher, 1996, Camacho et al., 2008) and a pattern recognition technique based on similarity indices (Ottavian et al. 2012) applied to time profiles of growth velocity and temperature gradient. These methodologies guarantee a in-depth understanding of the polyurethane systems and a science-based method which demonstrate to be much more efficient than the analytical state-of-the-art methods, which are mostly based on operators’ experience. Furthermore, a reduction of 50-75% of the duration of the product quality laboratory analysis is proposed. Finally, based on the inversion (Tomba et al., 2016) of a partial least squares regression methods (PLS, De Jong, 1993) calibrated on data from designed experimental campaigns, a tool is developed to suggest the most appropriate formulation which guarantees the desired final product quality in terms of growth velocity and foam’s height time profiles.File | Dimensione | Formato | |
---|---|---|---|
Mattiello_Massimo.pdf
accesso aperto
Dimensione
7.3 MB
Formato
Adobe PDF
|
7.3 MB | Adobe PDF | Visualizza/Apri |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/43564