A Consideration of Mechanical Kansei: Latent Representation of Topology-Optimized Structures in Variational Autoencoder

It is illustrated in general that the design process at the conceptual stage is highly dependent on individual skills and aesthetic sense. However, even in a subjective phase, objective aspects such as mechanical rationality are implicitly considered, and these two perspectives are inseparable. A concept of Mechanical Kansei defined as “the capability to evaluate and judge impressions in terms of unconscious, intuitive, and integrative information in mechanics,” has been proposed as a clue for interpreting the relationship between such impressions of objects and structural rationality in civil engineering and architecture fields. In this paper, we considered the Mechanical Kansei as the ability to intuitively recall shapes and forms that reflect the flow of forces from the given design domain and boundary conditions. The topology-optimized shape was taken as the structure that visualizes the flow. A Variational Autoencoder (VAE) was used to learn a set of shapes topology-optimized to the randomly located supports of a simple beam. The result shows that a two-dimensional latent variable space can derive the rational structures through the decoder without the mechanical model. This latent representation of topology-optimized shapes can be a clue to understanding the Mechanical Kansei.