Topology optimization of composite structures with data-driven resin filling time manufacturing constraint

Abstract

This paper presents a framework for modeling the data-driven manufacturing constraints and integrating them into the structural topology optimization. The empirical surrogate model of manufacturing constraints is constructed by mining the results of numerical process simulations of massively sampled topologies, using statistical learning. During optimization, the manufacturing constraints are modeled as additional objectives to the structural performance objective, and the multi-objective topology optimization is solved by the Kriging-interpolated level-set (KLS) approach and the multi-objective genetic algorithm (MOGA). The resulting Pareto frontiers offer opportunities to select the designs with some sacrifice in structural performance, yet improved manufacturability. An example of topology optimization of composite structures considering resin filling time showed the process of the proposed framework, and demonstrated its feasibility. Due to the proposed abstract topology features inspired by underlying physics of the filling process, the surrogate model of resin filling time is far more generalizable than the traditional surrogate models based on, e.g., bitmap and local feature representation. In particular, the model can reasonably be applied to the situations with the different inlet gate locations and initial bounding boxes from the training set, while the traditional surrogate models fail in such situations. Three case studies for composite structure topology optimization are discussed with different inlet gate locations and initial bounding boxes in order to show the robustness of the data-driven resin filling time predictive model.