Monitoring reinforced concrete structures is of particular importance when it comes to identifying potential anomalies (cracking or excessive deformation, for example) in relation to nominal operation. These anomalies can have consequences both for the overall behavior (strength, etc.) and the functionality (tightness, etc.) of the structure. To meet this challenge (fault detection and prediction of consequences), a strong coupling between measurement data and simulations is essential. Current methodology relies mainly on initial instrumentation of the structure, based on expert opinion or feedback, but the data is not processed and analyzed using numerical calculation codes. The subject of the proposed thesis falls within the framework of a methodological breakthrough, through the combination of machine learning and numerical simulation tools for the detection and diagnosis of anomalies on civil engineering structures, in order to develop intelligent and adaptive instrumentation for monitoring the life of the structure. The methodology revolves around the following axes: processing of measurement data by machine learning, leading to the identification of potentially faulty zones, reconstruction of suitable boundary conditions around the previously detected anomaly by metamodeling, and identification of the fault and its consequences by numerical simulation. The thesis will be carried out jointly by two CEA laboratories: LM2S, specializing in structural mechanics, and LIAD, a unit specializing in Artificial Intelligence and Data Science.
The desired candidate (M2 level) should have an appetite for advanced numerical methods (including machine learning), as well as knowledge of mechanics and/or civil engineering. By the end of the thesis, the candidate will have developed knowledge and skills in numerical simulation, data assimilation and mechanics that can be effectively exploited in both industrial and academic environments.
The thesis may be combined with a preliminary M2 internship.