Online stability boundary drifting prediction in milling process: An incremental learning approach

Abstract

The machining stability is an essential prerequisite for achieving required dimensional accuracy and high-quality surface of the machined parts. The stability boundary will drift with the changing machining dynamic characteristics due to tool wear, cutting temperature increase, etc. The online milling stability prediction is of great significance for improving machining efficiency and surface quality. In this article, we focus on the online monitoring of the position and shape of the stability boundary while milling titanium alloy wherein severe tool wear occurs. The machine learning binary classification algorithm is introduced to update the stability boundary of the machining process. The difficulties lie in: (a) identifying changes of the stability boundary with high sensitivity, (b) providing a high generalization capability to accommodate different characteristics of chatter state, (c) reducing the repetitive learning during the model updating and improving real-time performance. To tackle these challenges, a label-propagation-comparison-based incremental sequential minimal optimization (LPC-ISMO) model is proposed for stability boundary drift prediction. During the tool wear process, the stability domain change is treated as concept drift. The stability boundary updating is considered as a decision boundary training from support vector machine (SVM). The LPC-ISMO method is used to identify whether the stability boundary needs to be updated first, and if needed then update the boundary in the variation directions of the stability domain. Results show that the accuracy of the proposed approach is above 85%, which indicates that we provide a new method for online prediction of stability domain.