Assessment Of Delaminations In Fiber Reinforced Polymer Curved Plates

Fiber Reinforced Polymer(FRP)has been widely used in various important engineering fields due to their excellent structural characteristics.Therefore,it is imperative to monitor and detect delaminations which may reduce the load capability and service lifetime of FRP structures,at an early stage.The vibration based damage identification method can predict the damage of the structure based on the changes of the vibration parameters,in particular it can be applied to the on-line structural health monitoring(SHM).Compared with other vibration parameters,frequency measurement has the advantage of simply testing process,data stability,and good reproducibility.After analyzing by intelligent information processing technology,the frequency shifts can evaluate the damage state and safety performance of the structure.The FRP curved plates,which are actually a common application form of FRP materials,have been rarely report of its delamination detection work.Therefore,current project mainly focuses on delamination detection for the FRP curved plates through changes in frequencies,and the inverse algorithms,the artificial neural network(ANN)and genetic algorithm(GA),were developed in order to predict the location,size and interface of the delaminations in FRP curved plates through frequency shifts.The FEM model was established by model updating technique and was used to generate ‘delamination-frequency shifts' databasefor training inverse algorithms-ANN.Meanwhile,this FEM model was also used in GA to caculate the numerical frequency shifts for comparing to the target ones,and the delamination was predicted when the difference between the numerical and targeted frequency shifts comes to minmum.In numerical validation,both ANN and GA can accurately assess the damage with the prediction error less than 2%.However,for the discrete parameter – interface,ANN prediction is very poor,while GA is relatively better.However,the prediction time of GA is too long to be applied in the online health monitoring technology that demands quick identification.In order to improve the efficiency of GA,the surrogate model was built to replace the time-consuming finite element model calculation for frequency shifts.The prediction time is reduced to be 1/163 of the genetic algorithm without surrogates while with the comparable accuracy.One of the main difficulties of intelligent technology application in the structural health monitoring is the influence of noise.To compare the robustness of the two algorithms,this paper adds different level of noise to the numerical frequencies to simulate the errors caused by the experimental measurement or the model deviation,and the performance of these inverse detection algorithms under noises were thus evaluated.The results show that the prediction error of ANN is close to 18% under the 5% noise level,while the prediction error of GA is still within 10%.Therefore,under the impact of the noise,GA has better stability and accuracy,and ANN is sensitive to noise and thus less robust.In order to investigate the accuracy of the two frequency-based algorithms under laboratory conditions,seven carbon fiber reinforced composite(CFRP)curved plate specimens were manufactured of two different radian for the curves.The measured frequency shifs were used to predict the actual delamination of these speciments under FFFF and CFFC conditions.Noise response rate(NRR)was proposed to evaluate the influence of the frequency on the prediction accuracy under noise.The high NRR indicates that the mode of frequency is sensitive to noise,and for the first time,it is proposed to choose those frequency with low NRR for predicting delamination damage.Through experimental vailidation,this is approved to give much more accurate prediction than simply using all the measured frequencies.Moreover,the experimental results also show that the prediction accuracy of the CFFC conditions is higher than that of the FFFF boundary,with 10% prediction error for GA which is close to the sensitivity analysis results for the similar noise level,while ANN has worse performance,with prediction error in 35%.In conclusion,frequency-based method was validated numerically and experimentally to be able to predict delaminations successfully and the surrogate-assisted genetic algorithm was recommended to be applied since it is more robust.