Ph. D. Project
Wood Products Digital Twin: From Design to Use
2020/10/01 - 2023/09/30
Environmental awareness in France has re-energized the forest forest sector as a whole. This wood processing sector is organized and industrialized progressively in order to guarantee to its customers a quality of its products in line with the specific demands of the market. This important challenge is made difficult, on the one hand, by the extreme heterogeneity of the raw material (the trees), but also by the loss of information between the multiple actors of the logistics chain (from the forest to the customer ). This indirectly induces the use of detection means at each stage of the transformation process in order to collect partially disappeared data. The information thus generated, if it allows a posteriori qualification of the products after a transformation (or a priori but on a limited horizon), does not allow any prediction as to the fate of the products and to an optimized use of the raw material.
For example, the measured characteristics of trees in the forest with Lidar are not used on log yards, the measured characteristics of logs on log yards by RX scanners are not used in sawmills, etc. This lost but necessary data needs to be collected again by other means at other points in the processing chain. This necessarily represents significant financial investments, without making it possible to anticipate the future of products.

This thesis is intended to be a contribution to this problem of breaking the informational continuum at different stages of transformation of the wood product. It clearly poses the difficulty of traceability of products in complex transformation processes, where divergent and convergent cycles of transformation follow each other.

The contribution of this thesis will focus on the concept of Digital Twin Product and its potential for use in tracing products with a fingerprint approach (biometric signature).

The hypothesis posed here, and which we will seek to validate, can thus be expressed: the digital twin of the wood product makes it possible to know by anticipation its future characteristics. All or part of these characteristics define a biometric signature of the product. The signature of the product after a virtually applied transformation can be known in advance, without the physical product having been transformed, through the digital twin. It thus establishes the link between the signatures before and after an operation on the physical product.
In addition, the digital twin can be used as a support for a decision support system as to the fate of the product, based on the virtual product that it can represent in anticipation. This part exploiting the concept of virtual product is in continuity with the thesis of Jérémy JOVER.

Validation of this hypothesis can be carried out at a particular point in the chain of transformation of wood products namely sawing. This operation gives birth to various products from only one (a billon). The problems of choice of sawing scheme and the traceability of the products are particularly important. The modeling of the singularities influencing the choice of the sawing starting from the 3D tomographic images, the constitution of a base of images RX associated with the corresponding multimodal images (Color, Scatter, Depth, ...), and the extraction of these singularities of the Real and virtual wood products will be stages of the work to compare the qualities obtained on the real product and the virtual product. The creation of a biometric signature of the product with a view to its traceability will thus rely on heterogeneous data fusion methods (Choquet integrals, Fuzzy-AHP, ...), hierarchical classification (AG, RdN, blur,), even if the batches of data allow deep learning.
An experimental verification on CRAN's PIM platform (Epinal) will be a key point of this thesis.
digital twin, traceability, wood, virtual product, data fusion, learning, classification
Eco-Technic systems engineering
Demande de financement par contrat doctoral en cours