Author ORCID Identifier

https://orcid.org/0000-0002-9117-4465

Semester

Spring

Date of Graduation

2023

Document Type

Dissertation

Degree Type

PhD

College

Statler College of Engineering and Mineral Resources

Department

Mechanical and Aerospace Engineering

Committee Chair

Ever Barbero

Committee Member

Bruce Kang

Committee Member

Eduardo Sosa

Committee Member

Victor Mucino

Committee Member

Joe Bedard

Abstract

Structural health monitoring spans many decades of research across multiple engineering fields. However, typical monitoring processes for damage detection of complex structures usually prohibit real-time or fast detection of debilitating damage to the structure. One of the major issues of real-time detection of damage is the enormity of data that needs to be processed, which is worsened by the relative inability of fast relaying of data to structural engineers. With the rapid advancement of Machine Learning, both issues can be overcome, and detection of failure is achieved with non-invasive techniques. This dissertation explores the applicability of Machine Learning as a non-invasive technique for early critical failure detection in two separate examples. The first example is a column under buckling load. The column includes various imperfections, which can cause early failure compared to theoretical solutions reporting ideal buckling load. Using synthetic data, a large data set is used to train a Machine Learning algorithm to predict which columns are imperfection sensitive. The service load on these columns does not exceed 30% of the nominal critical load. The second example considers the aeroelastic response of an aircraft and specifically the extreme case of flutter. A combination of inertial, aerodynamic, and elastic forces, flutter is a dynamic instability on a vehicle, most observed on aircraft wings. A modal analysis on a wing reveals that the leading mode shapes are susceptible to flutter. The mode shapes, along with air velocities, structure damping values, and various experimentally tractable features are used to train a Machine Learning algorithm to recognize possible flutter. Further data on a wing with different material properties are then used to test the trained Machine Learning algorithm and observe its accuracy in early flutter detection. A linear regression model to find the damping values of the wing is also explored.

Embargo Reason

Publication Pending

Share

COinS