Automated data processing for efficient development of multimodal machine learning models in tool wear detection

Document Type

Article

Publication Date

7-7-2025

Department

Department of Mechanical Engineering-Engineering Mechanics

Abstract

This study aims to develop an automated data processing (ADP) framework that automatically processes multimodal data for machine learning (ML) diagnostics of manufacturing processes. The objective is to contribute to the state-of-the-art automated ML (AutoML) domain by developing a novel all-in-one ML development framework that automatically processes data, trains ML models, and conducts evaluation based on prior process-based knowledge. For this purpose, the ADP framework was designed based on a higher-level decision-making unit and a lower-level machine learning unit to select the best set of data processing methods, ML models, and training strategies, defined as policies, for the machine tool monitoring. The ADP framework was tested with data collected from orthogonal tube turning experiments for cutting tool wear classification. The data were gathered using acceleration, acoustics, and temperature sensors mounted on a Computer Numerical Controller (CNC) lathe machine. The results demonstrate that the solutions selected by the ADP framework consistently exhibited similar performance as the global optimal policy with improved computational efficiency. Consequently, the proposed ADP framework was demonstrated to automatically conduct multiple steps within the ML development process without manual intervention for tool condition monitoring to facilitate democratization of ML development for manufacturers.

Publication Title

International Journal of Advanced Manufacturing Technology

Share

COinS