Data-driven modelling approaches combined to physical models for real-time realistic simulations

Speaker:
Rodat, Damien; Airbus Group Innovations; France

Authors:
Rodat, D.; Airbus Group Innovations; France
Guibert, F.; Testia; France
Dominguez, N.; Airbus Operations S.A.S.; France
Calmon, P.; CEA, LIST; France

ID: ECNDT-0534-2018
Download: PDF
Session: Meta Modeling
Room: J2
Date: 2018-06-12
Time: 15:20 - 15:40

From manufacturing to in-service inspections, aircraft structures are under close monitoring through Non-Destructive Testing (NDT) techniques. A large number of these inspections relies on human operators, especially in the context of in-service airplane maintenance. The resulting signal fluctuations and diagnosis errors can be limited by careful operator training, and must be taken into account anyway when it comes to the qualification of NDT procedures or the development of diagnosis assistant algorithms. Managing human-related variability requires a large number of experiments involving different operators and flawed parts. For expensive aeronautical structures, the related costs are drastically increased. To tackle this challenge, the concept of operational NDT simulator was proposed. The simulator is equivalent to a classical NDT equipment but the displayed signals are generated using a mathematical model: expensive flawed parts are replaced by a representative mock-up and defects are numerically added where required. The underlying model must ensure a realism and a computation speed high enough to replace reality. Classical approaches to numerical simulations cannot meet both requirements; data-driven modelling techniques are thus investigated.

A proof of concept is proposed in the case of ultrasound inspections of composite structures. From the modelling perspective, the investigated strategy relies as much as possible on experimental data to ensure a high level of realism. Two different defect types are considered. A model of impact damage C-scans is built from a database of real impact inspections. The case of Flat Bottom Hole (FBH) defects is also studied with a meta-modelling approach applied on real data and enhanced by a physical knowledge. The resulting FBH model is able to efficiently replace real signals: certified NDT operators did not distinguish simulations from real signals. Thus, the operational NDT simulator can become a valuable tool to improve significantly any operations affected by human factors.