Date of Thesis

Spring 2019


In this study, a baseball pitch was examined to try to understand its behavior, and make a predictive model of it. A baseball pitch was tested experimentally with a wind tunnel and modeled computationally with COMSOL CFD software. Five input variables (spin rate, sting angle, seam orientation: Y axis, seam orientation: Z axis, and air velocity) were controlled, with force in three axes recorded as outputs. The experimental and computational results were examined and seen to be interdependent for all input variables. Experimental and computational data were both insufficient for predicting system behavior. Experimental data collection would have required an unreasonable amount of time, while computational data collection provided adequate qualitative results, but lacked quantitative accuracy. Both of these attempts to understand the system behavior fell short, indicating that the baseball was a complex system. This lack of system understanding from either experimental or computational results necessitated a different approach to predicting system behavior. This led to the application of a physics based machine learning algorithm, which aimed to combine experimental and computational data. This combination of data improved the predictive ability of the system, showing that a physics based machine learning algorithm can be used to better understand baseball pitch behavior. This result points to the possibility that a physics based machine learning algorithm can help facilitate the understanding of a complex system.


Baseball, Machine Learning, Physics, Engineering

Access Type

Masters Thesis

Degree Type

Master of Science in Mechanical Engineering


Mechanical Engineering

First Advisor

Indranil Brahma