Date of Award

Fall 2023

Document Type

Open Access Thesis

Department

Mechanical Engineering

First Advisor

Victor Giurgiutiu

Abstract

Structural integrity is a pivotal consideration in the field of engineering applications, and understanding the behavior of materials and joints under various loading conditions is critical. This thesis presents a novel approach to detect and analyze fatigue induced cracks in various test specimens using Piezoelectric Wafer Active Sensors. This approach is widely used in Structural Health Monitoring applications involving acoustic emissions that can be captured using PWAS which in turn gives the surface response to applied stresses. The ultimate goal of the research is to identify the length of a fatigue crack originating from lap joints and to develop a methodology to classify the crack lengths using a machine learning approach. This thesis is divided into five major chapters as discussed below. The first chapter gives a brief introduction to fracture mechanics, stress concentrations and piezoelectric sensors. Methods to determine the fatigue testing load by identifying the stress concentration factor and stress intensity factor of a specimen are discussed and verified by theoretical and experimental approaches. The second chapter introduces the types of specimens used for testing including dimensions and material properties and the third chapter introduces the concept of fatigue testing followed by the results of in-situ fatigue tests conducted using the proposed specimens. The fatigue life of the specimens are also discussed and proper fatigue loads for each specimen type are identified. The fourth chapter deals with sensor instrumentation and application, thus digging deep into the realm of acoustic emission capture using Piezoelectric Wafer Active Sensors. Methods for sensor bonding and quality check are discussed. The fifth chapter introduces the in-house data acquisition system used for AE monitoring and discusses the relevance of machine learning approaches to monitor the structural health of the specimen. It is shown that AE signals have distinct characteristics based on the crack size and the type of event such as crack growth or fretting. The machine learning approach uses a GoogLeNet model which is convenient and efficient in learning large datasets with minimal processing time. This model utilizes a simple convolutional neural network (CNN) that can identify various features in a Choi-Williams Transform image and distinguish between different wave signatures thus forming a methodology to determine the fatigue crack length. These machine learning models can be used with both active and passive PWAS for structural health monitoring in various conditions with reduced manual efforts to keep up constant monitoring.

Rights

© 2024, Siddharth Kannan

Download

Share

COinS