Digital shadow dedicated to resin infusion filling process of composite parts
Topic(s) : Special Sessions
Co-authors :
Philippe LE BOT (FRANCE), Damien LECOINTE , Nihad SIDDIG (FRANCE), Olivier FOUCHÉ (FRANCE), Yves LE GUENNEC (FRANCE), Ibrahim ABDULLAH (FRANCE), Florent NIGET , Christophe MARCHAND (FRANCE)Abstract :
On-line monitoring of resin infusion processes reduces the risk of scrap parts and improves the quality of produced parts. Different solutions are usually used to follow the filling phase, such as Infrared analysis, optic fibers, dielectric sensors, thermocouples or heat flux sensors and many other technologies [1]. These approaches give partial information about the process health. Another approach is the use of digital shadow making it possible to combine information collected in real-time with prior knowledge of process physics to provide decision support for operators [2]. The French institute IRT Jules Verne is leading the research project MONOCLE, with a consortium of industrials (NAVAL GROUP, SICOMIN, PREDICT, BUREAU VERITAS and PCMI). This project aims to develop a digital methodology for monitoring LRI process for large, thick, unique or near-unique parts, and thus provide operators with a decision-making tool. The monitoring solution is based on the integration of different functional layers within a dedicated framework process: instrumentation (flow), data collection and user restitution of observable elements, real-time communication using OPC UA protocol, simulation model, self-calibration, and filling time prediction. The reliability of the prediction depends on the relevance of physical assumptions, the initial and boundary conditions, and the knowledge of model parameters.
As a first validation, the prediction model is based on a physical formulation derived from Darcy's law for flow in porous media to estimate the evolution of the filling front position over time. This requires knowledge of the permeability of the reinforcements, which can be highly variable. Consequently, a real-time self-calibration approach using a Kalman filter is implemented inside the digital twin. The prediction model is also continuously updated to take account of actual process evolution. Validation tests on simple thin plates shows that the prediction horizon increases as the infusion goes on: prediction accuracy is rather weak in the first few moments, then becomes more relevant and highly accurate from halfway through the infusion.
The second step consists of implementing real-time process simulation instead of the analytical solution, to consider 3D flows as well as complex geometry and to predict non-quality effects such as dry spots. Model reduction techniques have been specifically developed for computation efficiency. In the end, some validations on complex shapes show the accuracy of the in-line predictions.
As a first validation, the prediction model is based on a physical formulation derived from Darcy's law for flow in porous media to estimate the evolution of the filling front position over time. This requires knowledge of the permeability of the reinforcements, which can be highly variable. Consequently, a real-time self-calibration approach using a Kalman filter is implemented inside the digital twin. The prediction model is also continuously updated to take account of actual process evolution. Validation tests on simple thin plates shows that the prediction horizon increases as the infusion goes on: prediction accuracy is rather weak in the first few moments, then becomes more relevant and highly accurate from halfway through the infusion.
The second step consists of implementing real-time process simulation instead of the analytical solution, to consider 3D flows as well as complex geometry and to predict non-quality effects such as dry spots. Model reduction techniques have been specifically developed for computation efficiency. In the end, some validations on complex shapes show the accuracy of the in-line predictions.