ResearchOS · Review Copilot

Review Copilot

Executive-level research quality analysis for reviewers, area chairs, supervisors, and students. Surfaces critical issues first. Human verification required for all outputs.

Paper text, Markdown, or LaTeX source Upload

# Acoustic-Based Chatter Detection in CNC Milling Using Lightweight Explainable Models

## Abstract
We propose a lightweight, explainable machine learning pipeline for real-time chatter detection in CNC milling using acoustic emission signals. Unlike prior deep-learning approaches that require GPU inference, our method achieves 91.4% F1-score using a gradient-boosted tree with SHAP explanations, running on embedded hardware at 48ms latency. We are the first to combine reproducibility, explainability, and embedded-deployment constraints. We release a reproducible benchmark dataset of 2,400 labelled milling cycles. Random seed: 42. Code available on GitHub.

## Introduction
Machining instability causes surface defects and tool wear. We address three gaps: reproducibility, explainability, and latency. Prior work [Smith et al., 2023] uses ResNet-34 at 420ms—infeasible for closed-loop control.

## Related Work
Smith et al. [2023] use ResNet-34 achieving 94% accuracy at 420ms. Jones & Lee [2022] apply SVM; dataset not released. Brown [2021] proposes 1-D CNN but omits learning-rate schedules.

## Dataset
We collected AE signals from a DMG MORI DMU 50. Sampling rate: 500 kHz. Random seed for all splits: 42. Dataset DOI: doi:10.5281/zenodo.DEMO.

## Method
LightGBM with learning rate 0.05, 300 estimators, random_state=42, batch size 16. SHAP TreeExplainer for feature attributions.

## Experiments
Baselines: threshold (71.2%), SVM (83.6%), 1-D CNN (88.9%), ResNet-34 (94.1%), Transformer (93.4%). Our method: F1 91.4%, 48ms latency, 5-fold CV: 91.4 ± 1.2%.

## Conclusion
Reproducible, explainable, embedded-deployable chatter detection achieving 91.4% F1 at 48ms latency.

## References
1. Smith, A. et al. (2023). Deep chatter detection. Int. J. Mach. Tools, 185.
2. Jones, D., Lee, E. (2022). SVM milling stability. Mech. Syst. Signal Process.
3. Brown, F. (2021). 1-D CNN for AE. J. Manuf. Sci. Eng.
4. Chen, G. et al. (2024). Transformer chatter. Rob. Comput.-Integr. Manuf.
5. Ke, G. et al. (2017). LightGBM. NeurIPS.

292 words · 2,037 chars

Ethics & Safety

This tool supports human review and must not be used as an automated decision system.

Do not submit AI-generated reviews without human verification.

Reviewer remains responsible for correctness, fairness, and confidentiality.

No paper content should be sent to external APIs unless explicitly configured.