Identification of piecewise linear dynamical systems using physically-interpretable neural-fuzzy networks: Methods...

Area of Science:

• Mechanics of Materials
• Robotics and Control Systems
• Computational Intelligence

Background:

• Self-locking origami structures exhibit complex, unmeasurable piecewise linear force-deformation relationships.
• Understanding these constitutive relations is crucial for predicting origami dynamics and folding processes.

Purpose of the Study:

• To develop a dynamical identification process for determining the model and parameters of piecewise linear origami structures.
• To create a physically interpretable neural-fuzzy network for analyzing input-output data from these structures.

Main Methods:

• A physically-interpretable neural-fuzzy network was constructed based on the piecewise linear assumption.
• The network's components (neurons, coefficients, validity functions) directly correspond to physical properties (segments, parameters, non-smooth points).
• Training involved local linear optimization, nested optimization for partitions, and Local Linear Model Tree optimization for model selection.

Main Results:

• The neural-fuzzy network successfully correlated measured input and output data for origami structures.
• Demonstrated the physical interpretability of the network's components in representing origami mechanics.
• Validated the approach through examples, illustrating effective data training methods.

Conclusions:

• The proposed physically-interpretable neural-fuzzy network offers an effective and generic method for handling piecewise linear dynamical systems.
• This approach provides significant physical insights into origami structures and advances artificial neural network research.
• The methodology is broadly applicable to systems with unknown piecewise linear characteristics.