Physics-Guided Machine Learning for CNC Milling

This project applies Bayesian machine learning to the selection of optimized milling parameters.


Currently, milling parameters are selected using ad hoc methods such as rules of thumb, past experience, or trial and error. This can lead to reduced efficiency and increased production times.

Proposed Solution

In Year One of this project, the team from the University of Tennessee-Knoxville and industry partner Third Wave Systems developed and validated the physics-guided machine learning approach. Physics-based machining model outputs served as input, or training data, for a Bayesian machine learning model. New observations came from in-process signals. Specifically, a microphone was used to record the machining sound signal.

In Year Two, the same team will work on the project. They will add a physical basis for selection of the next test point rather than using the current approach, which relies on maximized value of information. Also in Year 2, the new chatter frequency-based test point selection approach will be validated using the same materials and integral blade rotor (IBR) design from Year One to show any improvement. (Year One results will provide the baseline.)


This project aims to combine a physics-based model and machine learning to the selection of milling parameters, delivering maximum productivity with the minimum number of tests.