Resumen de Development of innovative methodologies to reduce railway rolling noise through genetic algorithm-based shape optimization techniques
Rolling noise phenomenon is produced due to the wheel/track interaction and induced by the small unevenness present in their surfaces. Such unevenness, known as "roughness", causes that vibrations arise in both the wheel and track when the train passes by with a certain speed, that consequently leads to the appearance of acoustic radiation. This kind of noise is one of the most relevant sources of annoyance and the principal focus of the railway acoustic pollution produced by trains operating through highly populated urban regions.
Thus, the main goal of the present Thesis is the development of a comprehensive methodology to achieve suitable railway wheel designs through the use of Genetic Algorithms (GAs) with the aim of minimizing the associated rolling noise.
When developing the aforementioned optimization algorithms, the Finite Element Method (FEM) is combined with the geometric parametrization of the different wheel design typologies analysed, described as a function of those parameters most relevant for the current research. In order to describe the dynamic behaviour of each component involved in the wheel/track interaction, use is made of linearised models in the frequency domain capable of solving the whole coupled dynamic response from the corresponding cross-section meshes.
Subsequent derivation of the radiated sound power from the dynamic information is carried out by applying a semi-analytical formulation that allows for the wheel acoustic efficiency computation, on one hand, and by making use of an Equivalent Sources Model (ESM) in the track, on the other hand. Besides, such theoretical development is validated with the reference commercial software in the field, TWINS, on which it is based.
Throughout the optimization procedures, a fatigue analysis is performed on every wheel design considered to assure structural feasibility, that acts as a "death penalty" constraint in the algorithm. Furthermore, a modal identification procedure is developed, which allows to characterize modeshapes and to classify them according to their number of nodal diameters and circumferences. Then, two different formulations of the objective function are explored: one focused on directly reducing radiated noise, named LA;W-min; and another centred on decreasing rolling noise by maximizing the average natural frequency of the modeshapes, called NF-max. Hence, in the LA;W -min methodology, the sum in energy of the wheel Sound poWer Level (SWL) expressed in dB(A) is minimized. For the NF-max case, natural frequencies are shifted to frequency regions where the roughness amplitude content is lower.
Different approaches are considered: the inclusion of perforation schemes in the wheel and the variation of its cross-sectional shape, setting the radius as a constant value in one case and using it as an optimization parameter in another. Moreover, the influence on the noise of changing the rail geometric and track viscoelastic properties is also studied.
As a result of the present Thesis, several quieter models of railway wheels have been achieved, with rolling noise reductions of up to 5 dB(A). When the whole railway system with all the components is considered, improvements in the radiated sound power remain achieved with the resulting wheel designs. Besides, correlations between maximization of natural frequencies and SWL mitigation are analysed, establishing the NF-max as a suitable methodology for cases when computational efficiency is prioritized. The sensitivity of the problem to selected design domains and the suitability of the use of GAs are also studied with the obtention of Response Surfaces (RSs) for the geometric parameters used. Additionally, correlations are established between the variation of the geometric parameters and the decrease in the associated acoustic radiation, while the shifting of the modeshapes along the frequency domain is proposed as a physical mechanism responsible of the observed sound power decrease.
