The discrepancy between the output of computer-based structural topology optimization and the final product that is ready for economical production remains a major drawback for the implementation of topology optimization in commercial applications. The checkerboard pattern as the preliminary manufacturability concern has been resolved by several regularization schemes. Among the schemes, the filtering methods are the most common.
However, one-step further, extra manufacturing constraints are needed to generate optimized structures in favor of selected manufacturing processes such as plastic molding and metal forming. Generally, some local modifications are implemented to components generated by the topology optimization in order to enhance their manufacturability that demand empirical expertise.
This practice is likely to lead to suboptimal solutions with respect to either structural performance or manufacturability owing to nonlinearity of responses to the change of topologies. Numerical process simulations have been incorporated into the structural design process in order to evaluate component manufacturability. Unfortunately, they cannot be incorporated directly into the optimization loop owing to the high cost of computation.
Professors Kazuhiro Saitou and his PhD student Yuqing Zhou at the University of Michigan, presented a framework for modeling the data-driven manufacturing constraints and integrated them into the structural topology optimization. They constructed the surrogate model of the manufacturing constraints by mining the outcomes of the numerical simulations of sampled topologies implementing statistical learning. Their research work is published in Structural and Multidisciplinary Optimization.
The authors modeled the manufacturability evaluation by the simulation-based data-driven manufacturing constraint modeling method. They solved the multi-objective topology optimization by the Kriging-interpolated level-set and multi-objective genetic algorithm. In order to demonstrate the robustness of the data-driven surrogate process model, and the feasibility of the proposed framework, the authors discussed a sample of the composite structure topology optimization taking into account resin filling time in the resin transfer molding (RTM) process.
Although the authors implemented the Kriging-interpolated level-set representation as a means of reducing the computational cost related to the non-gradient topology optimization methods, they acknowledged the computational efficiency drawback in their study. The resulting Pareto frontiers enabled the selection of designs with some sacrifice in structural performance, yet with enhanced manufacturability. Several examples of the topology optimization of composite structures taking into account resin filling time exhibited the process of the proposed framework and indicated its feasibility.
Reference to the proposed abstract topology feature motivated by the underlying physics of the filling process, the surrogate model of the resin filling time was more generalizable as opposed to the traditional surrogate models based on local feature and bitmap representation.
The model can be applied to situations with varying inlet gate locations as well as initial bounding boxes from the training set, while conventional surrogate models failed in such circumstances. The selected case studies for composite structure topology optimization were evaluated with varying inlet gate locations as well as initial bounding boxes in a bid to show the robustness of the data-driven resin filling time predictive model. The proposed data-driven approach for the manufacturing constraint modeling is a generic framework, which should also be applicable to other manufacturing processes.
Yuqing Zhou and Kazuhiro Saitou. Topology optimization of composite structures with data-driven resin filling time manufacturing constraint. Structural and Multidisciplinary Optimization (2017) 55(6):2073–2086.
Go To Structural and Multidisciplinary Optimization