Tool wear detection using novel acoustic emission features and a two-stage Mann-Whitney U test

Accurate detection of tool wear is a critical challenge in manufacturing due to the complex and nonlinear nature of acoustic emission (AE) signals generated during machining. Conventional methods, including machine learning and deep learning, often fail to detect the subtle changes associated with early tool wear and suffer from a lack of explainability in their decision-making processes. This paper introduces three novel AE features: rank-based entropy, fractal geometry indicator, and chaos quantifier. These features are specifically designed to capture the intricate dynamics of AE signals that traditional methods overlook. To validate the discriminative power of these features, a two-stage Mann-Whitney U test is employed, combining the new features with other advanced AE features to differentiate between worn and unworn tools. Experimental results from milling and drilling machine tests show that the proposed approach significantly improves detection sensitivity, accuracy, and interpretability compared to existing methods, providing a more robust framework for tool wear detection. The source code for this paper is available at: https://github.com/thuan-researcher/novel-ae-features-two-stage-mu.