Online Stress Monitoring Technique Based on Lamb-wave Measurements and a Convolutional Neural Network Under Static and Dynamic Loadings

Abstract

This paper presents the development of an online stress monitoring technique based on Lamb-wave measurements and a convolutional neural network (CNN) for metallic plate-like structures under static and dynamic loadings. Monitoring stress levels in structural components is crucial because stress concentrations and stress redistribution can be precursors of structural damage and failure. Moreover, the stress variations owing to dynamic loadings are directly related to the fatigue life of a structure. First, an aluminum plate specimen is fabricated, and piezoelectric transducers are installed on the specimen. Different levels of static loading are applied to the specimen with a hydraulic loading machine, and the ultrasonic responses are obtained at each loading level. The applied loads (ground truths) are measured by a load cell built into the loading machine. Then, a CNN is designed and trained by defining the Lamb-wave time responses as the input and the measured stress levels as the output. The performance of the trained CNN is evaluated using the blind test data obtained from various static and constant-amplitude cyclic loading conditions. The uniqueness of this study lies in (1) the automated stress estimation without feature extraction and (2) the stress monitoring capability under constant-amplitude cyclic loadings up to a frequency of 5 Hz.