Structural optimization of reciprocating engines for minimum noise radiation
A comprehensive suite of FE based structural optimization computer programs has been developed to minimise vibration and radiated noise of internal combustion engines. Finite element based vibration and noise prediction techniques, dynamic substructuring capabilities and design optimization methods are the key elements of the program suite. The entire engine noise generation process, starting from the combustion pressure and ending with the radiated noise, is within the scope of the programs. Engine structural member sizes, material properties, structural damping and the gasket stiffness and damping between engine components may be used as design variables. Limited shape optimization capabilities are also developed, including removal and addition of structural parts, and geometric modifications without FE mesh regeneration. The design optimization problem, which may involve complex design modifications, is finally transformed into a simple numerical optimization problem with a few design variables. The programs are so structured that any established numerical optimization method may be used to solve the final numerical optimization problem, although straight forward iterative optimization algorithms are shown to be inefficient for this application. The numerical optimization can be carried out either as an integrated part of the whole procedure or as another separate process. Extensive studies have been carried out on the various factors influencing engine noise optimization, including the characteristics of the radiated sound power as a function of structural design variables, the effects of damping, excitation and FE modelling. A comprehensive analysis starting from crank train loads and ending with radiated sound power level has been shown to be the basis of a practical optimization scheme. Sound power level has been identified as a suitable candidate to be used explicitly as the objective function of the optimization. Excitation models which fail to include correctly phased loads at main bearings and cylinders are shown to be inadequate for this application, although the conclusion may not apply at high frequencies (say, above 1 KHz) because the phase relationship at high frequencies might not be correctly predicted. Because engine FE models inevitably have large numbers of degrees of freedom, the sound power evaluation process is computationally very intensive. Therefore, each element of the procedure has been considered carefully to minimise the total computational burden without sacrificing important physical characteristics. The programs have been tested on a few realistic engine FE models, although this thesis will only include the results based on models of a four cylinder in-line diesel engine with up to 6000 degrees of freedom. The tests suggest that the noise optimization scheme is not only theoretically sound but also computationally viable, although further work is required in the related specialisations to fully realise its potential.