Uncertainty modeling and applications for operating data-driven inverse design

Abstract

Operating data-driven inverse design (DID) provides updated knowledge for the forward design process. It forms a closed-loop of design enhancements in which uncertainties from different sources, such as the model and environment, have a significant impact on the design and optimisation results. This paper proposes a method for analysing DID uncertainty based on Bayes’ theorem. First, the sources of uncertainty in the forward and inverse design processes are analysed, and the a priori information is defined in the inverse design process. The inverse relationship prediction model is based on correlation information and the a posteriori information. A clustering method is proposed to extract information on different working conditions from the operating data to provide a basis for mass-personalised design. Then, the a posteriori information and working conditions information are fed back to the forward design process for design optimisation, forming a closed-loop interaction process of forward and inverse design. Finally, the feasibility and effectiveness of the above methods are illustrated with a case study on the inverse optimisation of the structural characteristic parameters of a bicycle, and the clustering identification of environmental working condition load parameters for mass personalisation.