AI-Based Quality Monitoring of Industrial Screwdriving Processes
Development and evaluation of a machine learning approach to classify defined screwdriving errors using process and vibration data.
Background & Approach
Objective
The thesis aims to develop and evaluate a data-driven approach for automatically identifying different types of errors in industrial screwdriving processes. The central question is whether additional vibration measurements can improve classification compared with conventional process data alone.
Methodology
Torque, angle and time profiles are analyzed alongside vibration measurements from a low-cost sensor operating at a sampling rate of 100 Hz. The dataset includes reference screwdriving operations and deliberately introduced error types at different severity levels. Following preprocessing, temporal alignment and feature extraction, supervised machine learning methods are assessed for error classification. Unimodal and multimodal models are compared to determine the additional information provided by vibration measurements.
My contribution
Designed and developed a Python-based workflow for analyzing industrial screwdriving and sensor data. Main tasks include structured preprocessing, analysis of torque-angle characteristics, investigation of vibration signals and the development and evaluation of classification approaches.
Status & Findings
The thesis is currently in progress. It investigates whether combining conventional process measurements with affordable vibration sensing enables more reliable differentiation between screwdriving error types. Validated model-comparison results will be added once the evaluation is complete.
