Application of Deep Learning Approach for Predicting Electromechanical Impedance Signal of Steel-Concrete Bond Failure: Baseline Free Approach

Bond failure in reinforced concrete structures is challenging to identify at an incipient level, leading to loss of mechanical properties and making severity evaluation challenging using PZT sensors via electro-mechanical impedance (EMI) techniques. The prediction of EMI signals poses significant constraints, particularly in the case of historical or existing structures. The present study employs a convolutional neural network (CNN) and long short-term memory (LSTM), a proficient deep learning algorithm to predict baseline conductance and futuristic bond strength data of steel-concrete composite specimens to address this limitation. In this study, a metal wire based PZT sensor was instrumented at the steel interface for continuous bond strength monitoring. Conductance signatures were acquired at healthy and failure load conditions using pullout testing. Furthermore, the piezo-coupled signatures were used to build the model and predict using a deep learning approach. The prediction of the EMI signal was compared against experimental results, demonstrating the ability of the model to predict and forecast EMI signals with higher accuracy. The accuracy of the proposed model was validated through various performance metrics. The algorithm offers a valuable tool for predicting baseline signatures and is useful in health monitoring applications of different critical infrastructure systems.