Needs for automation in port surveillance, combating piracy or asymmetrical warfare and in the development of autonomous ships require new algorithms capable of assigning a class (sail boat, zodiac, etc.) to a ship based on a photo or video sequence.

These algorithms rely on automatic learning methods (machine learning).

"We shall present the general problem of machine learning, emphasising the specific difficulties of image classification. We shall then show how classifying ship images falls under previous developments within Naval Group Research, by focusing on the specific difficulties of this problem.

The presentation of the work and findings obtained throughout the thesis will then be used to illustrate how these difficulties can be faced."

The complexity of reality is a major obstacle to achieving a faithful simulation. Although we sometimes know the nature (statistically, for example), we never know the details. In addition, the laws of evolution governing the simulation are normally unstable with respect to deviations from the initial movement conditions and implementing them precisely is extremely costly in calculation time ...

The purpose of this thesis is to predict the impact of many sources of uncertainty on a physical quantity of interest (such as acoustic emissivity), which depends on the vibration response of a submerged structure. As simulating the response is costly in both calculation time and memory, high-dimension statistical learning methods are used.

By exploiting the low-level structure of the solution, we can create an approximation of the variable of interest which then only requires a low number of calls at a costly deterministic code.

Several welding passes in succession (multi-pass welding) are needed to weld very thick parts together. This process can generate specific microstructures and residual stress which can potentially weaken the assembly.

The work carried out simulated multi-pass welding of an HLES steel to predict the level and localisation of residual stress. A thermo-metallurgical and thermo-mechanical model of the steel was deduced firstly from a test campaign. A representative mock-up of a submarine web-plating join was then welded to measure changes in the microstructure and residual stress after each welding pass.

The final step was to compare the measurements with the results obtained by numerical simulation of the welding of this mock-up.

Naval platform control laws are currently validated by simulations and tests, methods which are not particularly suitable for controlling uncertainties. Postulating bounded errors on the settings of the platform dynamics and external interference, the problem envisaged in this thesis is defining a space containing the system trajectory with certainty.

For navigation in confined waters, we shall thus seek to prove the subsequent presence of the submarine in a tube, using more formal methods such as interval analysis. Work to date uses the theory of control and aims to guarantee, despite the uncertainties, certain global performance criteria associated with the system. The guarantee associated with the trajectory will be seen in a second phase.

The purpose of the work is to identify the magnetisation of a submarine hull by measuring the magnetic field near the hull so that the induced field can be corrected using an immunisation system associated to this measurement (closed loop immunisation).

The problem, called reverse, in determining the causes (magnetisation) based on the effects (measuring the field in a network of on-board sensors) is all the more difficult given that the measurements are made in near field.

An original method (regularising the solution) to ensure the uniqueness of the solution has been validated experimentally on an instrumented submarine mock-up (3 m long).

Titanium alloys are envisaged to produce parts intended for a marine environment as the titanium oxide film formed spontaneously in this environment provides natural protection. Nevertheless, in a multi-material system subject to cathodic protection, the production of hydrogen may cause hydrides to form that could weaken the material. The study will cover the hydrogen sensitivity of phases TA6V ELI and T40.

"We shall seek to understand the effect of the hydrogen and the induced alterations to the microstructure on the mechanical properties and especially the fracture mechanics."

It is currently particularly costly in resources and time to obtain experimental fatigue data on welded joints.

There is therefore an obvious economic advantage in developing a rapid method (gain of more than a factor of 10) which can also be used for parametric studies impossible to envisage until now given the necessary testing time (influence of load ratios, effects of history, finish, etc.).

A great deal of work based on measurements of the rise in temperature under cyclic loading has led to the development of a rapid method on homogeneous material test coupon, the so-called "self-heating" method.

"Our project aims to adapt this rapid approach to naval welded DH36 steel assemblies (therefore not homogeneous)."

The purpose of adaptive Human-Machine Interfaces (HMI) is to take into account the variability of the internal state of the operator to optimise the behaviour and performances of human-machine systems. Various fields are included in the design of this type of HMI, including human factors.

One of the human factor challenges in this long-term multi-discipline programme is to capture the internal state of the operator and infer the effects of this state on his performance. This mainly involves creating efficient user models.

The goal of this thesis is to assess the impact of the "effect of synchrony" on performance, by considering its links with awareness of the situation and mental load.

The work of this thesis explores the use of composite materials to improve the effectiveness and eco-compatibility of marine structures and reduce their costs.

The advantages of the specific bending-torsion coupling properties and light weight of these materials are mainly explored using numerical forecasting tools of fluid-structure interaction phenomena applied to deformable structures.

"We shall present the hydrodynamic tunnel tests on a composite profile, using vibration measurements, end displacement and stress under several flow conditions.

These experimental results are used to estimate the ability of the numerical coupling method to predict the fluid-structure interaction phenomena."

Nowadays, Additive Manufacturing (AM) processes normally operate in an open loop, which limits performance. Installing a servo-control loop assumes real time control (monitoring) and full understanding of the numerical simulation of the process.

But available AM modelling is not yet able to predict the behaviour of materials for the manufacture of actual parts and the current monitoring methods also seem insufficient for industrial applications.

A new multi-sensor monitoring methodology is being developed under the joint laboratory (JLMT) between Centrale Nantes and Naval Group. The aim is to control metallic Additive Manufacturing processes by powder spraying (LMD process) and wire deposit (WAAM process). Thermal and geometric monitoring strategies are proposed initially to check the geometry of deposited beads and the accumulated heat within the part and to predict the final microstructure during manufacture.

These strategies are applied on single-bead walls manufactured with the two processes of interest. Future work will study other monitoring possibilities.

Naval Group has been bonding materials to surface ships for about ten years. Full-scale impact tests are necessary to design these assemblies. Although these tests can endorse their mechanical resistance, they cannot test the entire domain of useful behaviour.

For the impact produces very extensive speeds and deformations for the adhesive that are difficult to understand for that. Thus, it is difficult to model the mechanical behaviour of the adhesive joint; there are few works that address this in the literature and it brings laws of behaviour as varied as viscoelasticity or hyperviscoelasticity into play.

The purpose of this thesis is to remove this technological barrier and suggest a generic impact design of bonded assemblies by applying a flexible adhesive.

Flows in areas with high energy potential are often very turbulent, which affects the operation of marine turbines.

Optimising a marine turbine farm therefore requires flow modelling around the turbine that is precise, takes account of the ambient turbulence and provides an understanding of the far wake generated by the marine turbine.

"To address this problem, we use instationary modelling of the fluid flow, the lattice Boltzman method (LBM) coupled with a large eddy simulation approach (LES). These tools are then used to model a marine turbine in a turbulent environment.

The calculations performed at two different upstream turbulence rates are then compared with experimental results and the results from a more traditional method. They are also used to analyse the marine turbine wake. The main observation is that the turbulence rate impacts significantly the propagation of eddies from blade end."