Designing method for Japanese traditional joinery using machine learning and its application to mechanical joint design.

(Designing joints based on clusters’ boundaries defined by weighted spectral clustering and optimum weight selection from stress tensor matrix)

Abstract

Green design is becoming a one of the major concerns of product design due to growing impact of global warming and to attain carbon neutrality. The Tsugite which can be seen in historical buildings in Japan is traditional Japanese joinery designed to enable replacement of worn parts to extend the life of temples and pagodas. It is inherently sustainable technique. In this paper, a new joint design method based on concepts of the Tsugite, which targets enabling replacement of parts for frame-structure automobile, is proposed. The weighted spectral clustering is selected to design a joint since a characteristic of the minimum-cut of spectral clustering is suitable for guiding and designing matching-surfaces of the Tsugite joinery. To apply this method to more complex shaped mechanical joints, coordinate transformation based on the principal axes of the stress-strain tensor is introduced. By selecting the shear stress in the plane defined by the two maximum principal axes as weights for clustering, an appropriate clustering results can be obtained, and boundaries of clusters show good guidance for designers to design mechanical joints for frame structures.