A hybrid data-driven model for in-process monitoring of peripheral milling under the variable process parameters

Abstract

Improper radial and axial depths of cut (DOC) during the milling process are one of the main causes of machining defects such as overcutting and machining chatter. A real-time DOC monitoring method is significant for improving the milling efficiency and quality. However, the coupling effect of DOC on physical signals such as force and vibration would deteriorate the monitoring algorithm. Therefore, a hybrid data-driven model is proposed to identify the DOC with multiple sensors, including force, vibration, current, and noise. Firstly, milling experiments with variables such as DOC, feed rate, and spindle speed are conducted to collect multi-source signals. Secondly, the average correlation coefficient method is proposed to select features that are sensitive to DOC. Then, a hybrid model that integrated Convolutional Neural Network (CNN), Bi-directional Long Short-Term Memory (BiLSTM), and Attention module (AT) is established to achieve DOC monitoring under multiple working conditions. The superiority of the proposed model is confirmed by milling experiments and comparisons with multiple state-of-the-art methods.