Study On The Application Of Machine Learning Methods To Interfacial Bond Strength Of FRP Reinforced Concrete

As an alternative to steel reinforcement under bad conditions,FRP reinforcing bars provides an effective solution to the corrosion problem of steel reinforcement.Interface bond performance has been considered as an important basis for the cooperative work between FRP reinforcing bars and concrete,which is an important factor to determine whether the structures can be used normally.However,due to the influence of many factors,there are many failure modes of FRP-concrete interface,and the prediction accuracy of the existing theoretical models for the interfacial bond strength is limited.At the same time,in order to get more extensive application of FRP reinforcement in the structural industry,it is necessary to study the decline of FRP-concrete bond strength under extreme high temperature conditions such as fire and the degradation of bond performance in the design service life.In order to get a more accurate and adaptable prediction model of bond strength in these three cases,machine learning methods are used in this paper.The main research contents and conclusions are as follows:(1)The interface failure mechanism of FRP reinforced concrete was analyzed.Based on the literature data,the influencing factors of FRP reinforced concrete interface bonding strength at room temperature and high temperature were discussed respectively.At the same time,several environmental factors affecting the durability of FRP-concrete interfacial bond strength and their influencing mechanism were summarized.The theoretical models and corresponding international structural codes of bond strength at room temperature and high temperature were collected and summarized,as well as the existing prediction models of long-term performance of FRP-concrete interfacial bond strength.(2)GA-BP neural network was used to establish the prediction model of FRP reinforced concrete bonding strength at room temperature.Based on the database containing 292 groups of experimental data,a three-layer neural network model with 13 neurons in the hidden layer was established with 234 groups of experimental data,and the 58 groups of experimental data were substituted into the model to verify the prediction ability of the model.(3)The GEP model was used to establish the prediction model of the residual rate of FRPconcrete bond strength at high temperature.The experimental data of 85 groups of pull-out specimens were collected,and the GEP model was established based on 75 groups of data to generate the explicit formula,and the remaining 10 groups of data are predicted by GEP model.The results show that GEP model,which contains more input parameters,has good prediction ability and can meet the accuracy requirements.(4)RF model was used to establish the prediction model of residual bond strength of FRP reinforced concrete under three common influencing factors in marine environment: seawater solution,alkali solution and aqueous solution.A database containing 210 groups of bond strength durability experimental data was established.168 groups of data were used to train the RF model,and the remaining 42 groups of data were predicted.The predicted values were compared with the experimental values to verify the prediction ability of the RF model.Based on the prediction data of RF model and Litterland’s model,the long-term prediction model of bond strength under the influence of these three environmental factors was established,and compared with the prediction model based on the code,so as to test the prediction ability of RF model in the durability of FRP-concrete bond strength.