An improved UNet model based on adaptive activation function and squeeze-and-excitation module for milling tool wear segmentation

Abstract

The fine monitoring technology of the milling tool wear condition is a crucial prerequisite to ensure both the processing quality and the smooth progress of the machining process. In order to fulfil this requirement, this paper constructs an improved UNet model to achieve end-to-end high-precision segmentation of milling tool wear area. The model uses Resnet as the feature extraction framework, and introduces an adaptive activation function to prevent information loss and minimize the activation function cost. Meanwhile, the squeeze-and-excitation module is introduced in the front and back ends of the feature extraction framework to enhance the important features and suppress irrelevant features. The accuracy and adapt ability of the proposed model is confirmed through the experiment of accelerating milling cutter life and three different failure phenomena.