Harnessing interpretable machine learning for holistic inverse design of origami

Area of Science:

• Computational mechanics
• Machine learning
• Materials science

Background:

• Origami-inspired systems offer unique reconfigurable properties.
• Inverse design of functional origami is complex due to multi-objective performance criteria.

Purpose of the Study:

• To develop an interpretable machine learning framework for the inverse design of functional origami.
• To generate human-understandable rules for designing origami with specific performance targets.

Main Methods:

• Utilized a decision tree-random forest workflow to analyze origami design features and functional performance.
• Handled complex interactions between categorical and continuous design features.
• Extended shape-fitting algorithms to incorporate non-geometrical performance metrics.

Main Results:

• Established a workflow for fitting origami databases and generating inverse design rules.
• Demonstrated the ability to tackle multi-objective problems for functional origami.
• Enabled consideration of non-geometrical performance in origami design.

Conclusions:

• The proposed framework facilitates holistic inverse design of origami, integrating shape and function.
• Enables the creation of novel reconfigurable structures for applications in metamaterials, robotics, and biomedical devices.