Topology Optimization and Machine Learning based Parametric Optimization Techniques: A Comparative Study With Physical Validation

Topology optimization (TO) and Parametric Optimization (PO) are two fundamental structural optimization (SO) techniques. The aim of this study is to understand how these two techniques (TO and PO) stack up against each other from a variety of perspectives - (1) time taken for optimization, (2) manufacturability, (3) real-life adherence to simulated results, and (4) overall stiffness - by using the problem case of a cantilever beam as the basis of comparison. For this cantilever beam application, both TO and PO are evaluated and compared in a systematic manner on 3 different volume fractions (VF) of beams - 30%, 50%, and 80%. PO is conducted using surrogate optimization as opposed to simulation in-the-loop optimization. Most importantly and uniquely, the results achieved via software simulation for both PO and TO are then validated in a real-world physical setting. The simulated versus physically validated results are compared for all six TO and PO optimized cantilever beams. Through this comparison, it was found that the percent errors between the simulated and physical displacement for the 30%, 50%, and 80% TO/PO beams were 45.5%, -6.6%, -1.5%, 33.7%, 2.4%, and -0.1%, respectively. It was found that - for the chosen cantilever beam application - (1) TO is faster and shows better compliance, (2) PO is more manufacturable and shows better predictability between software simulated and physically validated results. Moreover, for both TO and PO, the 50% and 80% VF beams show good alignment between simulated and physical results.