Predicting specific energy absorption capacity in modified auxetic re-entrant honeycomb structures via symbolic regression

Machine Learning (ML) models are widely used in contemporary scientific and engineering research to model complex phenomena with high accuracy. However, they lack interpretability and rarely provide additional understanding of the phenomenon at hand. These models are therefore referred to as black-box models.Based on the case study of modified re-entrant auxetic structures, a symbolic regression (SR) framework is introduced using the PySR library to derive an explicit algebraic formulation coupling geometry describing parameters to specific energy absorption capacity. Consequently, a human readable and interpretable formula is proposed describing highly non-linear structure- property coupling, bypassing black-box limitations of conventional machine learning models. The predictive performance is compared to conventional Artificial Neural Networks (ANNs). Additionally, sensitivity analysis is performed to assess the contributions of individual features. Ultimately, the PySR derived equation achieved a MAPE on unseen test data of 22.79% ± 17.21%, which competes with the predictive performance of conventional ANNs with a MAPE of 24.74% ± 22.57%.

Version

Published version

Access right on openHSU

Metadata only access