Development of data-driven prediction model for CFRP-steel bond strength by implementing ensemble learning algorithms

Bonding carbon fiber reinforced polymer (CFRP) laminates has been broadly utilized in steel structure rehabilitation. As for the final strengthened capacity, the bond strength between CFRP and steel usually plays a dominant role instead of the CFRP's mechanical property. However, the bond behavior of the CFRP-steel (CS) interface is very complicated with various failure modes and consequently the bond strength is hard to estimate leading to the CFRP strengthened steel structure insecure. Under this circumstance, in order to accurately predict the bond strength of CS, efficient data-driven models were developed through implementing ensemble learning (EL) algorithms named by "gradient boosting decision tree (GBDT)" and "random forest (RF)" as two representative ones on a collected CS single-shear test database. These models' performances on bond strength prediction were compared and also three representative machine learning algorithms "artificial neural network (ANN)", "support vector machine (SVM)" and "classification and regression tree (CART)" are utilized for validating the necessity. The comparison results indicate that the model generated by the GBDT algorithm attains the best accuracy for CS interfacial bond strength prediction (R2 = 0.98) among the ensemble and machine learning algorithms. Through model explaning analysis, the mechanism behind the GBDT based prediction model was also carefully verified. After these tests and analyses, the GBDT based model was proved to have the potential to facilitate the design and evaluation of CFRP strengthened steel structures.