Exploration of weak-PGML Method for Efficient Stability Control During Machining Operations

Abstract

Stable control of the machining process can be difficult to maintain in production environments due to many factors, but especially uncertainty about selection of optimal machining parameters. The true underlying stability model for production machines on the shop floor is generally unknown, so parameters are typically selected by operators based on manufacturer recommendations or, in most cases, their accumulated shop floor experience. A class of methods referred to as physics-guided machine learning (PGML) offers a new approach for optimal parameter selection that incorporates theoretical knowledge about machining dynamics into the data-driven machine learning workflow. The focus of this paper is a version of PGML appropriate to shop floor operations in small and medium-sized machine shops (SMMs)—referred to as weak-PGML—that does not require a priori theoretical knowledge of machining dynamics to approximate the true underlying stability model for a particular machining scenario. Rather, the weak-PGML leverages the operator’s accumulated domain knowledge to uncover the true stability model and informs parameter selection. Weak-PGML reflects the limited resources and lack of technical expertise typical of most SMMs with legacy machines and a new generation of workforce without the accumulated experience of operators who have spent many years on the shop floor.