|
Title |
Bayesian Regularized Artificial Neural Network Model for Predicting the Gauge Factor of Self-Sensing Cementitious Composites
|
|
Authors |
안호현(Ho Hyun An) ; 이상훈(Sang-Hoon Lee) ; 주효은(Hyo-Eun Joo) ; 김강수(Kang Su Kim) |
|
DOI |
https://doi.org/10.11112/jksmi.2026.30.4.201 |
|
Keywords |
인공신경망; 베이지안 정규화; 게이지계수; 부분 의존도; 퍼콜레이션 이론; 자기감지 시멘트 복합체 Artificial neural network; Bayesian regularization; Gauge factor; Partial dependence plot; Percolation theory; Self-sensing cementitious composites |
|
Abstract |
This study presents a Bayesian regularized artificial neural network (BR-ANN) model to predict the gauge factor of self-sensing cementitious composites (SSCC). An experimental database of 84 datasets with 9 input variables was constructed from existing literature. After confirming the highly nonlinear relationship between mix parameters and the gauge factor, the BR-ANN model was developed, demonstrating high predictive accuracy (R² > 0.9) and effectively preventing overfitting. Furthermore, permutation feature importance and partial dependence plots (PDP) were employed to enhance model interpretability. The analysis visualized a nonlinear, inverted U-shaped relationship between the carbon nanotube-to-binder ratio and sensing performance, aligning with percolation theory. Ultimately, this model can facilitate the optimal mix design of smart composites for structural health monitoring.
|