Robust Optimization and Digital Twins for Maritime Transition – MATritime

The maritime sector faces significant challenges: imposed reductions in the energy footprint of maritime transport, the advent of new modes of propulsion (sail, hydrogen), automation, and digitization ... At the same time, the numerical/digital revolution in naval design processes requires a great mastery of multiple complex domains specific to uncertain environments made up of the sea, the atmosphere, and their interface. New advanced procedures are needed to meet the challenges of a more sustainable, greener, and robust maritime industry. Meeting these challenges requires a considerable evolution of engineering practices with the establishment of dedicated processes in Computational Science and Engineering (CSE) based on advanced digital simulation technologies combining physical and statistical models. Indeed, even if the computing resources increase, the limitations of the physical models and the cost of high-fidelity approaches limit the simulations to a few nominal configurations. However, concentrating the simulation effort on a nominal system may be insufficient if the real-world system differs from the simulated one (due to manufacturing tolerances, random intrinsic effects, model error, poorly known environments, ...). In these situations, it is crucial to objectively quantify the uncertainties of the numerical predictions induced by the system's specification errors and model and to account for all these uncertainties during analyses and decision-making processes. This characterization makes it possible to design more robust systems reaching better levels of performance in actual conditions. The project proposes to develop a holistic approach to uncertainties by equipping numerical predictions with probability laws. Depending on the quality of the probabilistic representation, the computational overhead to estimate the prediction uncertainty can be very large. For example, Monte Carlo sampling methods require many simulations to estimate the variance of predictions, with a prohibitive cost when applied directly to detailed physical models. To overcome these limitations without renouncing precise physics, one has to resort to efficient approaches to produce probabilistic predictions at an acceptable cost. For this, we plan to develop methodologies closely associating physical and statistical modeling (e.g., multi-fidelity, multi-level Monte-Carlo, surrogate models, design of numerical experiment). All these methods, as opposed to purely statistical methods (such as Artificial Intelligence), incorporate physical simulations into the statistical processing producing the prediction; in return, they require a great deal of interaction with the experts of physical simulations to be developed. Our objective will be to deploy these numerical approaches and propose advanced uncertainty analyses, robust predictions, and design strategies for maritime applications. These complex applications will lead to developing research in robust multidisciplinary (approaches by subsystems) and multi-objectives design strategies to cover ship design, from component to system optimization. We will also set up a prototype of a ship's digital twin, integrating models and data to support the digitization of the maritime world and prepare future tools for operational issues (optimization of missions, routes, maintenance operations, ...).