PhD Oral Exam - Ali Ebrahimi, Mechanical Engineering

When studying for a doctoral degree (PhD), candidates submit a thesis that provides a critical review of the current state of knowledge of the thesis subject as well as the student’s own contributions to the subject. The distinguishing criterion of doctoral graduate research is a significant and original contribution to knowledge.

Once accepted, the candidate presents the thesis orally. This oral exam is open to the public.

Abstract

Due to the significant variability in the fatigue performance of composite materials, their durability assessment remains a major challenge. To achieve reliable evaluations, the in-service damage in each component must first be individually identified; then it should be translated into a quantifiable measure of durability such as residual strength or remaining fatigue life. This study investigates the feasibility of detecting in-service damage in composites using piezoresistivity-based sensing approach and translating these damage-related electrical signals into a measure of durability employing machine learning (ML) techniques, in three different scenarios: (1) in-situ residual strength prediction of composites using their electrical response during fatigue loading, (2) residual strength prediction of composites with unknown fatigue damage histories using their electrical response during a low-stress cyclic loading diagnostic test , and (3) early fatigue failure detection of composites based on their electrical behavior at the initial stages of loading. To pursue these objectives, numerous glass/epoxy specimens, rendered electrically conductive through the incorporation of carbon nanotubes (CNTs), were fabricated and subjected to various experimental procedures to generate data for each case. The resulting data were subsequently processed using ML techniques to assess the possibility of each scenario.

First, to examine the feasibility of in-service residual strength prediction, specimens were exposed to fatigue loading to introduce fatigue induced damage while their electrical responses were continuously monitored. Upon completion of fatigue damage, the specimens underwent quasi-static loading to determine their corresponding residual strength. The analysis showed that a K-nearest neighbor (KNN) meta-model, combining Decision Tree (DT), Support Vector Regression (SVR), and KNN as base learners in an ensemble learning framework, effectively predicted the residual strength of unseen samples using their electrical behavior during fatigue loading, achieving a mean absolute percentage error (MAPE) of 4.7%.

Second, to evaluate the practicality of residual strength prediction of composites with unknown fatigue damage histories, a set of specimens covering a broad spectrum of fatigue damage levels were prepared and subjected to a low-stress cyclic loading diagnostic test, which is designed to introduce negligible additional damage, while their electrical responses were recorded. The specimens were subsequently subjected to quasi static loading to determine their corresponding residual strength. The analysis showed that a KNN meta-model, incorporating DT, SVR, and KNN as base models in an ensemble framework, successfully predicted the residual strength of unseen specimens using their electrical response during the diagnostic test, achieving an MAPE of 5.7%.

Third, to assess the possibility of identifying early fatigue failure in composites, specimens were subjected to fatigue loading until failure while their electrical responses were continuously monitored. Short-life specimens—those failing prematurely—were then identified, and an autoencoder-based anomaly detection technique was employed to detect these samples using their electrical behavior during the loading. The results indicated an F1-score of 95% and an accuracy of 97% in correctly identifying short-life specimens, confirming the effectiveness of the proposed approach for early fatigue failure detection.

Collectively, these investigations highlight the strong potential of integrating piezoresistivity based sensing approach with ML techniques to develop intelligent systems for durability evaluation of composites. Such systems can enhance the reliability of composite structures by reducing the risk of catastrophic failures and support a shift from schedule-based to condition-based maintenance planning, enabling safer and more efficient use of composite materials.