This talk is expected to create greater awareness of the challenges in the numerical methods for structure and machines design.
Dr Ranjbar's presentation will be on a comparative study on optimisation in structural acoustics. A combination of a commercially available finite element software package and additional user-written programs is used to modify the shape of the structure. This is done iteratively and without manual intervention to achieve significant improvements of the objective function. The optimisation process continues automatically until the predefined maximum number of function evaluations is reached. The design variables are the structure's local geometry modification values at selected surface key points. The objective of the optimisation includes the minimisation of the root mean square level of structure borne sound (a general measure of the vibrational sensitivity of a structure). In addition, the structural mass remains constant and the allowable ranges of design variable values are restricted by prescribed upper and lower limits. The optimisation procedure is tested on the finite element model of a rectangular plate made of steel. Twelve different optimisation methods are tested against each others. These methods are considered either as approximate or exact. The approximate optimisation methods use either an approximated value of objective function, e.g. hybrid design of experiments and hybrid neural networks, or the approximated values of the first and the second derivatives of the objective function, viz. method of feasible directions, sequential quadratic programming, Newton method, limited memory BFGS method for bound constrained problems, method of moving asymptotes, mid-range multi-points method and controlled random search method. The exact optimisation methods, e.g. genetic algorithms, tabu search and simulated annealing, are derivative-free methods and they use the exact value of objective function. Furthermore, a statistical approach is followed for the comparison of optimisation methods. Advantages and disadvantages of each optimisation algorithm are reported in detail. The rate of convergence (a measure of optimisation speed) and the robustness level of each optimisation method are evaluated. Some optimisation methods are classified as fast, medium and slow. Method of moving asymptotes and mid-range multi-points method are introduced as the fastest methods. An efficiency criterion, namely, the overall performance level, which is a measure of convergence rate and the robustness level for an optimisation method, is introduced. The most effective group of optimisation methods includes method of moving asymptotes, method of feasible directions and simulated annealing method. They are the most suitable candidates for implementing in the hybrid robust shape optimisation algorithms for structural acoustics applications. Also, the overall performance level can be considered as a switching strategy in hybrid optimisation algorithms. However, it is predicted that a hybrid optimisation algorithm with one switching stage is relatively robust and fast for numerical shape optimisations in structural acoustics.
Two parallel programming techniques are also experienced on the optimisation methods. It does not mean that optimisation code itself is parallelised, but rather are the independent optimisation computations distributed to several CPUs to collect more results. Finally, the use of effective structural-acoustic analysis methods can drastically reduce the total optimisation time. If the powerful optimisation methods become equipped with effective (fast and reliable) structural acoustic analysis methods, then they can present more desirable optimisation results in a shorter period of computation time. In this case, they can even be considered as a suitable replacement for the complex and the multi-stages hybrid optimisation algorithms.
About the Speaker
Born in Shiraz, Iran, Dr Mostafa Ranjbar obtained his doctorate degree in Mechanical Engineering from Technical University of Dresden in Germany. He is currently Head of the Faculty of Mechanical Engineering, Islamic Azad University-Pardis Branch, Pardis New City in Iran. His research interest encompasses structural acoustic optmisation, noise and vibration. He is the invited speaker at ACMAE 2010 hosted in Kuala Lumpur.
10.30am Dr Mostafa Ranjbar's Talk
Who Should Attend
This is a highly technical talk for engineers, researchers, managers from the machineries and systems sectors of the PE industry.
This is a non-chargeable event. Pre-registration is required. To reserve a seat, please register online.
For technical enquiries, please contact Dr Lin Wei, Email: wlin@SIMTech.a-star.edu.sg
For general enquiries, please contact Alice Koh, Email: kohth@SCEI.a-star.edu.sg