Engineering food packaging safety
Phuong-Mai NGUYEN
Thesis to obtain the degree of Doctor awarded by L'Institut des Sciences et Industries du Vivant et de l'Environnement (AgroParisTech)
Specialty: Food science and processes
Thesis supervisors: Marie-Laure Lameloise (Pr. AgroParisTech) and Olivier Vitrac (CR-INRA)
LNE supervisors: Patrick Sauvegrain, Cédric Lyathaud
Thèse CIFRE (convention N° 45/2011) carried out in collaboration between UMR 1145 "Génie Industriel Alimentaire" (INRA/AgroParisTech) and the "Chimie et Physico-chime des Matériaux" cluster (LNE)
Period: April 2011 - April 2014
Summary
The safety of food contact materials is currently the subject of much controversy. This thesis work is part of the ANR collaborative research project entitled " SafeFoodPack Design: reasoned design of safe packaging" and aim to develop preventive approaches to controlling food contamination by materials in direct or non-direct contact with food. The initial paradigm is that safety can be built using transfer prediction tools, tools to help deform materials and predict their barrier properties. All these methods already exist in the literature. Some of them have been improved and integrated into a single FMECA (Failure Mode, Effects, and Criticality Analysis) design and evaluation approach, which makes extensive use of multi-scale modeling. The molecular scale aims to produce transport coefficients that can be used in models derived from continuum mechanics. Within the framework of an all-atom Flory-Huggins approximation, calculations of activity coefficients and heats of sorption have been generalized to polar and block polymers. Macroscopic modeling is performed using a volume-finite formulation, based on semi-analytical solutions optimized for barrier materials and simulation chaining. At the supply chain level, an inference engine automatically handles dependencies and Bayesian diagrams. The complete approach has been successfully tested for its ability to predict past crises and analyze emerging contamination pathways in the gas phase. To support the creation of a database on the composition of packaging on the French market, a semi-supervised method for deconvolving NMR spectra using a tracking algorithm was also developed. It enables substances from a large dictionary (50 to 400 substances) to be identified and quantified from the 1H NMR spectrum of a polymer extract.
Key words : packaging, food safety, FMECA, diffusion, NMR, molecular modeling
Summary
The safety of food contact materials is subjected to numerous controversies. This thesis work is part of the collaborative project ANR "SafeFoodPack Design: rational design off safe food packaging" and aims at developing preventive approaches for the control of the contamination of foodstuffs by materials directly in contact or not with food. The initial paradigm is that the safety can be enforced with the help of prediction tools: mass transfer simulations, decision tools for the deformulation and estimations of barrier properties. All methods already exist in the scientific literature. They were improved for some of them and integrated within a same approach for design and assessment based on multiscale modeling and using a FMECA (Failure Mode, Effects, and Criticality Analysis) type formulation. Molecular modeling produces transport coefficients to be used in continuum mechanics based models. Within one all atom Flory-Huggins approximation, calculation methods of activity coefficient and sorption heat have been generalized to polar and block polymers. Macroscopic modeling is carried out through a finite volume formulation relying on semi-analytical solutions and optimized for barrier materials and simulation chaining. At the scale of the supply chain, an inference engine enables to manage automatically the dependence and the Bayesian networks. The full approach has been tested with success upon its ability to guess past crises and to infer new emerging contamination pathways via gas phases. In order to support the creation of a large composition database of packaging materials on the French market, a semi-supervised deconvolution method of NMR spectra has been developed using a pursuit algorithm. It makes it possible to identity and to quantify substances of a large dictionary (including from 50 to 400 substances) starting from the 1H NMR spectrum of a polymer extract.