Electromechanical impedance-based concrete structural damage detection using principal component analysis incorporated with neural network

This paper proposed an electromechanical admittance (EMA, inverse of impedance)-based concrete structural damage detection (SDD) approach using Principal Component Analysis (PCA) algorithm incorporated with neural network (NN). First, the PCA algorithm was applied to extract essential feature of the original EMA data obtained from an active piezoelectric lead zirconate titanate (PZT) transducer through data compression. Second, the extracted principle components were adopted as input of a generated NN model to enhance damage prediction ability of the EMA technique. Third, after a sufficient training of NN model using the PCA-processed data, structural damage in term of severity and location could be automatically predicted. Feasibility of the approach was evaluated via conducting an experiment on detecting crack damages in concrete cubic structure, performance of the NN model was analyzed using randomly varied different ratios of training dataset as well. Experimental results demonstrated that the proposed approach gives around 100% accuracy for crack identification, which provided an alternative paradigm of data-driven damage detection in concrete structures.