Parallel topology optimization introducing enclosed cavity manufacturing constraints based on graph search algorithms

We present a distributed method based on Graph Theory for detecting enclosed cavities in density-based topology optimization using the traditional material penalization scheme. We integrate the enclosed cavity detection method into a distributed topology optimization framework that utilizes domain decomposition to leverage parallel computing resources. We introduce enclosed cavities as a manufacturing constraint in a modified topology optimization formulation to control unwanted designs based on the number of enclosed holes. Removing enclosed holes is essential for many additive manufacturing techniques because they can complicate the manufacturing process. The cavity detection method generates a graph with the subdomain design variables represented as nodes and their connections as arcs. The enclosed cavity detection method only uses the void phase part of the corresponding subdomain graph, which is divided into disconnected graphs using a Graph Search algorithm. We then classify them based on the connections between subdomain void graphs. The proposal minimizes inter-node communications by generating a hierarchical distributed graph to obtain a coherent representation of the void phase connectivity during the optimization process. Finally, we evaluate the feasibility and scalability of the computational framework using two- and three-dimensional problems with symmetric and asymmetric simplifications using multi-core computing. • Efficient distributed enclosed cavity detection method based on Graph Theory. • Topology optimization formulation penalizing enclosed cavities. • A distributed framework to remove enclosed cavities in topology optimization. • Validated manufacturability of 3D designs using additive manufacturing. • Strong scaling validation of the proposed distributed framework.