Meet the Researcher - Kyle Singleton

Breakout Room: 12

KS_ProfilePhotoResearcher Name: Kyle Singleton
Title of Research: Optimization and Application of Physics Informed Neural Network
Division Representing: Engineering
Institution: Mississippi State University
Institution Location: Mississippi
Home State: Mississippi
District Number: 3
Advisor/Mentor: Sungkwang Mun
Funding Source: United States Department of Army Research Lab

Research Experience:  
Kyle is a senior at Mississippi State University majoring in Mechanical Engineering. Kyle's work history consists of current employment at the Center of Advanced Vehicular Systems researching applications of artificial intelligence in additive manufacturing. As an undergraduate research assistant, Kyle has assisted in optimizing neural networks for various applications, studied applications of thermodynamics in physics based neural networks, and increased his understanding of artificial intelligence and python coding language. In addition, he worked as a co-op for two terms at Nucor Steel located in Huntsville, Alabama. Job responsibilities at Nucor were working in a safety-conscious, team-oriented environment, designing work platforms, under the supervision of a Professional Engineer, using the AISC Manual, and studying failure mechanisms to engineer solutions. Also, Kyle interned for a semester at CF Industries in Yazoo City, Mississippi. His responsibilities included conducting Ultrasonic Thickness readings guided by the Mechanical Integrity inspector, developing safe internal and external predictive maintenance plans for pressure vessels, and conducting business with vendors to facilitate the installation of magnetic drive pump systems. Lastly, Kyle has recently began his senior design project working with Los Alamos National Laboratory where his team will be designing a testing bed for materials exposed to extreme temperatures. Kyle has also served as Vice president of Mississippi States Society of Mechanical Engineers where he spent time tutoring others for the GRE Exam and leading the canned food drive initiate for the JL King Center.

Presentation Experience: 
Kyle has had the opportunity to present through school clubs and class projects.   As the President of Mississippi State Hockey, he has presented updates at scheduled team meetings, team standings to general manager and coach, and options for potential sponsors.  He was also responsible for presenting to the board of the Southeastern Collegiate Hockey Conference on behalf of Mississippi State Hockey.  Lastly, he was invited to speak on his experiences playing hockey at the Mississippi Sports Hall of Fame on behalf of Mississippi State Hockey.  Additionally,  as former Vice President of Mississippi States chapter of the American Society of Mechanical Engineers he was tasked with speaking at club events and presenting club goals to sponsors and faculty.  School projects gave Kyle the opportunity to present technical information to and academic audience.  Kyle and his team researched the failure mechanics of a hockey stick and were tasked with a thirty minute discussion to approximately eighty graduate and undergraduate students on how the hockey stick failed, how the material failed, how the study was conducted, and how to improve the design.  Also, in a undergraduate course studying the 3D modeling software, SolidWorks, Kyle was tasked with modeling a piston from an engine.  He was then tasked with presenting his design and how his design could be further improved to over forty students and multiple professors.

Significance of Research:       
Machine Learning has proved to be a useful tool as it translates observed data into a prediction of the physical circumstance.  One common machine learning system is the artificial neural network that interprets data and produces outcomes. These outcomes are compared to reference data sets and use feedback to train and correct the predictions. However, this process relies heavily on access to large data sets which allow the network to train. To reduce the amount of training data required, laws of physics can be implemented into the neural network. This is referred to as a Physics Informed Neural Network (PINN), and it applies additional constraints to the system which act as a filter for results. As a result, less data is needed to train a PINN, and the PINN can extrapolate results based on the known physics, therefore, allowing PINNs to serve as an approximate solver for partial differential equations. The accuracy and speed in which the network trains are crucial in determining if it is an effective substitute for other solver methods. In this study, a PINN is created to solve the heat equation for a number of sample boundary conditions in 1D and 2D. The accuracy of the PINN and its training speed is analyzed as a function of multiple factors such as the shape and size of the PINN, the number of PINN colocation points, and the size of the training data set. This presentation provides insight into the optimization of a PINN and applications for the network. 

Uniqueness of Research: 
The research on physics informed neural networks provides great significance in further understanding the optimization of the system, and the applications in additive manufacturing.  This study will prove the benefits of machine learning when compared to other solvers.