Open Collections

Tai, Chung-Yu

University of British Columbia

Spindle health monitoring in machine tools is paramount for optimizing performance and preventing costly downtimes due to repairs. The spindle, a critical component influencing machine tool efficiency, undergoes gradual wear and crack development, often stemming from long-term usage or serious collisions, especially in joint locations like bearing balls, raceways, and the tool holder taper contact interface. These faults result in significant vibrations and poor surface finish during machining operations. This thesis proposes a hybrid approach by integrating a physics-based digital model of the spindle with machine learning principles to enhance spindle health monitoring. Bearing faults, worn tool holder taper contact interfaces, and spindle imbalance are mathematically modeled and integrated into a dynamic digital model of spindles. These faults induce changes in preload and dynamic stiffness, leading to observable vibrations at the spindle speeds and ball passing frequencies. The digital spindle model predicts vibrations caused by spindle faults at specific measurement locations An analysis of multiple spindle fault couplings is implemented to recognize critical signal features used for training. Vibration spectra, natural frequencies, and dynamic stiffness changes are correlated to faults, experimentally validated before and after repair of spindles, as well as under different health conditions of tool holders. To monitor the spindle health, gate recurrent unit (GRU) neural network algorithms are employed for spindle fault detection and examination of its acceptable status. Pre-trained on vibration spectra generated by the physics-based spindle simulation model and fine-tuned by a few measurements, the GRU classifiers for addressing the fault locations and predictors for determining the acceptable levels achieve an accuracy of 96.74% and 94.10% on experimental datasets not used in training. The proposed integrated approach combines physics-based modeling and data-driven techniques, contributing to optimal spindle fault diagnosis performance, minimizing downtimes, and adhering to ISO standards for spindle status evaluation. All proposed methodologies are experimentally validated, offering a promising solution for enhancing the reliability and efficiency of spindle health monitoring systems in the manufacturing industry.

Text

2024-04-29

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/88074

Mechanical Engineering

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/4.0/