Improved Meta-learning Neural Network for the Prediction of the Historical Reinforced Concrete Bond–Slip Model Using Few Test Specimens

• DOI: 10.1186/s40069-022-00530-y
• Publication Date: 2022-07-28T23:07:30Z
• Authors: Chengwen Zhang; Qing Chun; Ao Sun; Yijie Lin; HaoYu Wang
• Keywords: Mahalanobis distance; Rebar

Abstract The bond–slip model plays an important role in the structural analysis of reinforced concrete structures. However, many factors affect behavior, which means that a large number tests are required to establish accurate model. This paper aims data-driven method for prediction historical with few test specimens and features. Therefore, new Mahalanobis-Meta-learning Net algorithm was proposed, can be used solve implicit regression problem few-shot learning. Compared existing algorithms, achieves fast convergence, good generalization without performing tests. applied task square rebar-reinforced concrete. First, first pretraining database model, BondSlipNet, established containing 558 samples from literature. BondSlipNet provide priori knowledge Then, another database, named SRRC-Net, obtained by 16 groups pull-out rebar. SRRC-Net posteriori knowledge. Finally, based on databases, not only successfully predicted concrete, but also other 23 types research results scientific basis conservation structures contribute