AI based Topology Opmization

Mechanism topology optimization using a ground structure model is a useful design tool for simultaneously determining the topology and dimensions of mechanisms. However, this approach requires a large design variable space to be able to synthesize various mechanisms, resulting in a high computational cost and complexity to find converging results.

 

To overcome these difficulties, we propose a method to reduce the design variables in the ground structure model by using a spring-connected rigid block model which is capable of representing various mechanisms with a single model.

 

Our approach involves two key steps to effectively reduce design variables.

First, based on the ground structure model, we create 2D images that vary according to the design variables. Second, we apply these generated images to a beta-variational autoencoder to reduce the dimensionality of the design space. This dimensionality reduction facilitates the efficient synthesis of mechanisms under various design conditions.

 

To demonstrate the effectiveness and validity of our method, we perform topology optimization based on several numerical examples using the reduced design variables. A genetic algorithm is employed to address multiple optimization problems, including a discrete optimization problem. Using the proposed method, we successfully synthesized mechanisms that met the specified design conditions within the reduced design space.