Comparison of Health Indicators Construction for Concrete Structure Using Acoustic Emission Hit and Kullback-Leibler Divergence

Abstract

This paper investigates the construction of health indicators (HIs) for concrete structures using acoustic emission (AE) hit and Kullback-Leibler Divergence (KLD). Health indicator has an important role in the structural health monitoring (SHM) framework through its portrayal of the deterioration process. By harnessing AE nondestructive test, the authors suggest that the HI can be constructed through a deep learning model from the raw data. Prior to the training of the deep neural network (DNN), its parameters are achieved by autoencoder pretraining and fine-tuning. Afterwards, the AE hits and KLD values are extracted from the data to be the training label for two different types of HI. The evaluation of two HIs are done with fitness analysis and remaining useful lifetime (RUL) prognosis, which shows both their capability to present the deterioration process and their drawback in regard to this matter.