Les exigences de qualité des produits ainsi que des normes environnementales et énergétiques de plus en plus drastiques nécessitent le développement de technologies de fonctionnalisation de surface en particulier celles qui s’appuient sur les procédés de revêtement par dépôt de couches minces. Le contrôle de la qualité de surface revêtue présente un enjeu industriel d’envergure. En effet, il n’existe pas à l’heure actuelle de technique d’inspection non destructive qui allie à la fois rapidité, fiabilité et flexibilité pour le contrôle de l’uniformité de revêtement. Pour répondre à cette problématique, ce travail de thèse porte sur le développement d’une technique d’inspection basée sur la thermographie active. Après étude expérimentale et numérique de la réponse thermique de surfaces bicouches, nous proposons une première méthodologie d’exploitation des mesures issues d’une excitation ponctuelle (laser) et surfacique (flash(s)). L’approche mise en place s’appuie sur l’implantation d’un algorithme des moindres carrés partiels (PLS NIPALS). Celui-ci a été testé sur plusieurs matériaux conducteurs et non conducteurs et dans différentes configurations expérimentales puis comparé à la méthode de contrôle conventionnelle par courants de Foucault (pour les matériaux conducteurs). La méthode développée permet d’extraire la signature thermique intrinsèque de l’hétérogénéité d’épaisseur du revêtement. Une deuxième approche a été explorée, elle s’appuie sur la mise en œuvre des nouveaux outils que nous offre le « Deep Learning ». Les premiers résultats obtenus semblent prometteurs. L’ensemble des résultats ouvre le champ vers une exploitation industrielle de la thermographie infrarouge pour le contrôle non destructif de revêtement. / Product quality requirements as well as increasingly drastic environmental and energy standards require the development of surface functionalization technologies, particularly those based on thin film coating processes. The quality control of coated surface presents an important industrial challenge. Indeed, actually there is any non-destructive inspection technique that combines speed, reliability and flexibility for coating uniformity inspection. To respond this challenge, this work focuses on the development of an inspection technique based on active thermography. After experimental and numerical studies of thermal responses of bilayer surfaces, we propose firstly a measurement methodology based on a point (laser) and surface excitation (flash (s)). The approach is based on the implementation of a partial least squares algorithm (PLS-NIPALS). It was tested on several conductive and non-conductive materials and in various experimental configurations and compared to the conventional eddy current control method (for conductive materials). The developed method makes it possible to extract the intrinsic thermal signature of the coating thickness heterogeneity. A second approach has been explored, based on the classification algorithm based on Deep Learning tool. The first results seem promising. The overall results open the opportunity to an industrial exploitation of infrared thermography for non-destructive coating testing.

A fundamental problem of microwave (MW) thermal processing of materials is the intrinsic non-uniformity of the resulting internal heating pattern. This project proposes a general technique to solve this problem by using comprehensive numerical modeling to determine the optimal process guaranteeing uniformity. The distinctive features of the approach are the use of an original concept of uniformity for MW-induced temperature fields and pulsed MW energy as a mechanism for achieving uniformity of heating. The mathematical model used to represent MW heating describes two component physical processes: electromagnetic wave propagation and heat diffusion. A numerical solution for the corresponding boundary value problem is obtained using an appropriate iterative framework in which each sub-problem is solved independently by applying the 3D FDTD method. Given a specific MW heating system and load configuration, the optimization problem is to find the experiment which minimizes the time required to raise the minimum temperature of the load to a prescribed goal temperature while maintaining the maximum temperature below a prescribed threshold. The characteristics of the system which most dramatically influence the internal heating pattern, when changed, are identified through extensive modeling, and are subsequently chosen as the design variables in the related optimization. Pulsing MW power is also incorporated into the optimization to allow heat diffusion to affect cold spots not addressed by the heating controlled by the design variables. The developed optimization algorithm proceeds at each time-step by choosing the values of design variables which produce the most uniform heating pattern. Uniformity is measured as the average squared temperature deviation corresponding to all distinct neighboring pairs of FDTD cells representing the load. The algorithm is implemented as a collection of MATLAB scripts producing a description of the optimal MW heating process along with the final 3D temperature field. We demonstrate that CAD of a practical applicator providing uniform heating is reduced to the determination of suitable design variables and their incorporation into the optimization process. Although uniformity cannot be attained using“static" MW heating, it is achievable by applying an appropriate pulsing regime. The functionality of the proposed optimization is illustrated by computational experiments which show that time-to-uniformity can be reduced, compared to the pulsing regime, by up to an order of magnitude.