Co-authors​ :

 Johan MERZOUKI (FRANCE), Anaële LEFEUVRE , Julien GOSTEAU (FRANCE) 

The recent advances in computational engineering and data-driven approaches, applied to materials and processes, have enhanced the efficiency of material design. For example, Takada et al. 2021, optimised a blend of PPS and elastomer processed with a twin-screw extruder thanks to Machine Learning (ML) (Random Forest algorithm). To accelerate discovery of 3D printing materials, Erps et al., 2021, applied Bayesian Optimization (BO) to automatically guide the experimental design, ultimately creating materials with higher performance. With the same intention, Chen et al. 2023, developed an adaptive framework to accelerate optimization of flame-retardant composites using ML. Data from experiments, handbooks and published papers (called domain knowledge) were used to train the models, allowing them to get feedback on or predict the properties of interest. In the end, whatever the data-driven method used, the main challenge is to work from an exhaustive, reliable, well structured and clean dataset. For this purpose, Zhao et al. 2018, designed an ontology-enabled knowledge graph framework for nanopolymer systems that led to an actual platform for data storage and retrieval while ensuring compatibility with ML tools for material discovery and design.
In a recycling context, relationships between composite material constituents, semi-product and processing parameters are even more important to enable material recycling and circularity opportunities. Indeed, aged materials as a starting material design point need new methods & tools for suitable material and semi product specification. Consequently, within the Airbus Central R&T project named CREDITS (Composite REcycling DIgitalization Tailored Semi-product), a data driven method is deployed to reveal hidden relationships between recycled material ingredients, semi-product & product properties accelerating their development and facilitating their specifications.
Specifically, the approach aims at proposing a common polymer and composite ontology which is a form of structured knowledge representation and to transform it into a usable graph database. The data collection, coming from past projects, literature, simulation and experiments, is handled through a Graph DBMS (DataBase Management System), from collection steps to exploitation by statistical data analysis and ML.

In the frame of CREDITS, the method is applied on two 3D-printing use cases : a recycled plastic alloy A350 duct-bracket and a recycled composite helicopter evaporator outlet connection. This presentation will focus on showing the path employed to minimise the number of experiments leading to conform recycled 3D printed polymer and composite parts, revealing our ontology proposal, data collection and storage methods, data analysis tools as well as our smart design of experiments approach.