Sviluppo di un modello teorico e computazionale per l'analisi di propulsori e velivoli ad ala rotante, basato su tecniche di sintesi di sistemi complessi mediante reti neurali.

The present work addresses the development of theoretical and computational models for the analysis and preliminary design of rotary-blade based propulsive systems used in aeronautical and marine vehicles. From a general standpoint, the improvement of propulsor efficiency is a challenging problem in most aircraft/vessels design and has received considerable attention in the last decade; the reduction of the operating costs accounting for environmental issues (fuel consumption and N Ox emissions, for instance) represents one of the most urgent needs for designers. The availability of fast and accurate numerical prediction tools during the pre-design stage is a crucial point to achieve effective propulsion systems optimization strategies. In fact, guidelines derived at early stage may result into faster and less expensive iterative procedures within the classical design spiral steps. In despite of the previuos considerations, numerical tools currently used in the pre-design stage are often based on empirical or semi-empirical simplified theories for the prediction of propeller performance; this is due to the lack of fast and reliable numerical prediction tools compared to time consuming advanced models that, in turns, are not suited for design purpose. In view of the above considerations, the main objective of the present thesis is to develop a fast and automated numerical tool for the optimal design. In detail, a suitable procedure is assessed and applied for the design of rotating-blade propulsive systems. Numerical models are managed synergically through: . an automated generation of different configurations; . an accurate estimation of the behaviour of the physical system to be investigated; . the synthesis of the response (hyper)-surface describing the system behaviour by a surrogate model for a fast exploration of different configurations; . an interface between the approach used to estimate the response (hyper)-surface with a numerical optimization model. For the optimal solution seeking procedure, a set of design variables is firstly defined. Different geometric configurations may be obtained by changing their values within a suitable bounded domain: hence, a virtual systematic propeller series is defined and the optimal solution is searched in its context. The optimal solution is obtained through the maximization of a suitable objective function defining the response surface of the system. The strategy used to find the optimal solution consists of two steps in which two different sets of design variables are allowed to vary, respectively. In detail, globals and local design variables are introduced and, in turns, two optimal global and local optimal solutions are derived. The convergence of the optimization process should require the use of an iterative technique; however, herein this work, a straighforward single-step optimal solution is carried out. This strategy allows the reduction of the maximum configurations number to be analyzed in the solution seeking. Although the modelling environment proposed is general, the application to a problem of practical relevance in the maritime transport context is considered here. Specifically, the optimization of conventional isolated propellers as well as ducted propellers, operating in uniform onset flow, is addressed. The hydrodynamic efficiency is chosen as objective function and contributes to the definition of the response surface of the system. An unsteady, three dimensional hydrodynamics solver based on a Boundary Element Method (BEM) for the analysis of the flow around thrusting propellers is chosen to provide predictions of propulsor performance. Based on inviscid flow assumptions, BEMs are recognized as reliable tools for the hydrodynamic analysis of conventional propellers, and represent an appealing trade-off between accuracy and overall computational burden. Here, an effort to extend the advantages of such a computational methodology to complex propulsive configurations like ducted propellers has been performed. Validation studies are presented in the thesis in order to assess the capability of the developed BEM to predict isolated and ducted propeller performance under a wide range of operating conditions. Within a chosen systematic propeller series, performance data set for isolated or ducted propeller configurations is built-up using BEM simulations; then, the response surface is synthetized through a suitable regression model. This procedure defines the surrogate model that is used for the prediction of the optimal configuration. The regression model used for the synthesis of the response surface consists of a Neural Network (NN) model. Among a variety of NN models, the development of a multilayer feed-forward network trained via a Levenberg-Marquardt model based on a back-propagation technique for the determination of the gradient search direction has been found as an adequate approach in view of the problem of interest. A careful investigation is performed on notional cases in oder to analyse the sensitivity of the developed NN model with respect to different architecture parameters, and the choice of parameter settings to be used is then motivated. The neural network is then interfaced with the optimization model in order to explore the surrogated response surface. Two optimization models are considered here: a Genetic Algorithm (GA) or a simple Parametric Modelling (PM) approach based on a systematic variation of the design variables. The aforementioned numerical approach has been applied to the design of new propellers to be installed on an existing aged fishing vessel; in fact, the replacement of propulsion units on existing vessels represents one of the most effective solution to achieve remarkable trade-off between gains in hydrodynamic efficiency and budget necessary. Nowadays, such a design update strategy (retrofitting) is very popular among ship owners. Thus, an existing vessel is selected as a reference case and the development of a new propulsor with enhanced performance is the final objective of the numerical applications of the proposed methodology. The optimal solution is searched here both under the class of conventional propellers as well as among ducted propellers. Hull boundary-layer induced effect is herein modeled through global parameters known from the characterization of the performance of the existing vessel. Optimal configurations are carried out through the numerical optimization procedure above described, assuming the delivered engine power (at a given blade rotational speed) as requirement. Note that the whole hull-propeller-engine configuration imposes constraints on the thrust exerted by the propulsor. This thrust has to balance, under given operating conditions, the resistance the ship hull determines when moving at sea. Next, the torque absorbed by the propulsor necessarily matches the torque available by the engine rotation. Further constraints are due to minimizing the harmful consequences of cavitation. In this respect, standard formulas to estimate the risk of cavitation-induced erosion and thrust loss are employed via the classical Keller method. Requirements on operative conditions (i.e., minimum advance speed) and geometrical issues (i.e., maximum allowable propeller diameter) close the definition of the design problem through additional contraints. In the thesis, results of the application of the proposed methodology to the selected retrofitting problem are discussed in details. For the optimal configurations, it has been demonstrated that the synthetised response surface determined by NNs is in fair agreement with the response surface that can be derived by using the actual BEM hydrodynamics model. Of course, using the NN-based surrogate, the computational effort is dramatically abated. Next, numerical results derived from GA and PM approaches predict similar configurations. Comparison between the resulting optimal propeller and the baseline propeller shows an improvement, in terms of hydrodynamic efficiency, of about [3 - 5 %]. This gain is fully within the range of estabilished retrofitting aged vessels, as known from designers' practice and is documented in scientific literature on the subject. It is highlighted that largest enhancements in terms of hydrodynamic efficiency are achieved when working points (tipically the advance speed) different from those of the baseline propeller are enforced. In the optimization procedure, the ammissibility of optimal guesses where different working conditions are determined is evaluated against the matching of imposed constraints. Different propeller configurations are considered in the global design problem, depending on the number of blades. Optimized conventional propellers are characterized by small differencies in terms of both diameter and expanded area ratio (i.e., chord) whereas lower values of nominal pitch are observed. Although a single objective optimization is performed in which a given operating condition is addressed, the solutions reveal some robustness in that the hydrodynamic efficiency is increased over a range of the operating conditions around the imposed ones. Similarly to conventional propellers, optimized ducted propellers are characterized by limited differencies in terms of geometrical features. The numerical results demonstrate that i) for a given working point, the higher the gain obtained, the higher the probability to reduce it in other working points, inducing, as a limit case, undesired loss; ii) the lower the gain, the higher is the probability to confirm such a tendency over a wide range of working points. As mentioned above, the major contribution for the efficiency increasing comes out from a re-definition of the working point whereas, if it were fixed (as requirement) during the design process, the maximum achievable gain in efficiency, would decrease to [1 - 2 %]. In conclusion, model validation studies and numerical applications to a retrofitting design exercise of practical interest described in the thesis show that the proposed methodology can be applied to the preliminary design of screw propulsors. The interplay among accurate hydrodynamics models, predictions acceleration obtained via Neural-Network based surrogated models, and numerical optimization tools provides a fully automated environment. Nevertheless, some important modelling aspects have been not included into the scope of the present work and are worth to be investigated in further studies. A first way to improve global performances should be achieved increasing the accuracy of sub-models used during the design process. In particular, a development of the regression model used for the synthesis of the response surface of the system, as well as of the hydrodynamics flow solver is desired. The need to accurately modelling the interaction between hull and propeller by including the presence of the effective wake induced by hull boundary layer will lead future research. In fact, the unsteady interaction between hull and propeller blades deeply affects the hydrodynamic behaviuor of the propeller system inducing transient cavitation on blade surfaces that, in turns, causes annoying sound radiation in the field and induced-vibrations on the hull-plate. The inclusion of more sophisticated and comprehensive system modelling aspects yields that the developed architecture is still valid, provided suitable interfaces with a multi-objective optimization process are introduced.