Title: Post-doctoral position: Machine learning approaches for monitoring and prediction in time and statistical domain for the ocean engineering and maritime industry
Date: 2022-06-01 09:54
Author: Benjamin Bouscasse
Job_Duration: 18 mois (extension possible)
Job_Employer: Ecole Centrale de Nantes
The shipping industry is an old established and rather traditional industry, but it has now to evolve fast to comply with new regulations. This implies developing prediction tools both in the time and statistical domain, in order to be able to better route the ships, control the risks at seas, and order some maintenance operations in a timely manner.
Ocean engineering is more recent but it also faces new challenges. The development of large floating wind turbine farms is a key response to the “green society” challenges and reducing the cost of production is possible by improving the maintenance operations and the efficiency of the turbine energy production. This can be also achieved with better prediction tools.
Both the Naval and Offshore industries deal with large and complex structures interacting with a multi-physics environment. In the LHEEA lab, large competencies have been developed over the years to model hydrodynamic interactions between ships or structures and the ocean environment. The approaches are mainly through physical or mathematical models, from the more complex and expensive “high fidelity” ones to the simplest and cheapest “low fidelity” ones. The validation phase is crucial for all models and solvers, as it determines their cost and accuracy. LHEEA’s large experimental hydrodynamic facilities are useful to provide validation datasets.
On both sides there are limits, high fidelity models are good but costly, and anyway still unable to fully accurately predict some highly-coupled and multiscale problems. Low-fidelity models are widely used but usually with large safety factors and they largely fail for deterministic predictions.
Machine learning is an interesting alternative framework. It is now used by multiple actors. At ECN the approach to machine learning focused on neural network to study a couple of problematics where the classical approaches have clear limitations.
Can we learn the behaviour of a marine structure with machine learning techniques? Is it more efficient with respect to classical approaches? There is a series of questions needing answers in this regard.
The postdoctoral position proposes to develop machine learning algorithms and codes, expanding their use in typical ocean engineering applications. The results will be compared to those obtained with traditional methods on a set of validations test cases. The principal application targeted are wave prediction and motion prediction.
At LHEEA some workflows were developed based on neural networks, the first task will be to further improve these models. It will be anyway important to assess the suitability of alternative techniques.
In order to start meaningful work on this, simplified problems will be considered, and parametric studies will be conducted to get control of the neural network capability.
Then, more complex cases will be considered, using experimental datasets.
Some info on Ecole Centrale Nantes:
Centrale Nantes is a French « grande école d’ingénieurs » engineering school, part of the « Écoles Centrales » group (Lille, Lyon, Marseille, Nantes, Paris). It delivers high level teaching for selected students.
The Research Laboratory in Hydrodynamics, Energetics and Atmospheric Environment (LHEEA) at Centrale Nantes is a CNRS mixed research unit (UMR 6598). The LHEEA is tasked with both advancing theoretical knowledge and solving concrete problems around four scientific themes: free-surface hydrodynamics, fluid-structure interactions, dynamics of the atmosphere and systems approach for ground and marine propulsion systems.
The post doc is hired in the IIHNE research group which is studying complex hydrodynamic flows in interaction with marine structures, through numerical and experimental techniques.
Keywords: Hydrodynamics, Waves, Artificial intelligence; deep learning; Time series; Motion prediction;
Required qualification :
• PhD in fluid mechanics or in Computational science
• Very good English level
• Experience/interest in Deep Learning with the standard libraries of the domain (scikit-learn, tensorflow, pytorch,…)
• Interest/experience in software development
• Experience/Interest in Naval or Ocean Engineering
• Autonomy and ability to work in a team
Details on position :
• Start date as early as possible
• Fixed-term contract 18 months (with possible extension)
• Office based in Nantes, France
End of advertisement period: Until the position is filled
Gross Salary: 2600 Euros/month