M3-2.3 - A Dataset for Condition Monitoring of Screwdriver Bits based on a Piezoelectric Accelerometer for Machine Learning

H. El Moutaouakil; J. Schauer; A. Schütze

M3-2.3 - A Dataset for Condition Monitoring of Screwdriver Bits based on a Piezoelectric Accelerometer for Machine Learning

Event: 23. ITG/GMA-Fachtagung Sensoren und Messsysteme 2026
2026-06-09 - 2026-06-10
Nürnberg
Band: Vorträge
Chapter: Modellbildung und Signalverarbeitung 2
Author(s): H. El Moutaouakil, J. Schauer - Centre for Mechatronics and Automations Technology, Saarbrücken, A. Schütze - Saarland University, Saarbrücken
Pages: 463 - 468
DOI: 10.5162/sensoren2026/M3-2.3
ISBN: 978-3-910600-11-9
Price: free

Abstract

To ensure high-precision industrial operations, monitoring assembly processes and tool health using machine learning has become critical. In automated screw assembly, screw bits are vulnerable to progressive wear and sudden damage, which can lead to workpiece defects and costly production downtime if not identified early. Conventional monitoring techniques often rely on torque or angle analysis; however, these typically provide post-process diagnostics rather than real-time insights. This paper presents a high-resolution vibration dataset designed to advance research in real-time tool wear detection and process monitoring. The data was captured using a piezoelectric accelerometer mounted on an electric screwdriver, which recorded vibrations during both tightening and loosening cycles. To simulate industrial degradation, controlled wear was artificially induced on the bits across multiple incremental stages. The resulting dataset includes three independent experimental series, comprising varied screw hole manufacturing processes and randomized sequences to ensure model robustness and prevent overfitting. By providing well-annotated time-series data in MATLAB format, this resource enables the development of interpretable machine learning models that predict wear stages and recommend proactive tool replacement. Initial validation using time-frequency feature extraction and principal component analysis demonstrates clear signal separability, confirming the dataset's utility for both classification and regression-based condition monitoring.