Failure Diagnosis for Dental Air Turbine Handpiece with Payload Using Feature Engineering and Temporal Convolution...

Area of Science:

• Biomedical Engineering
• Machine Learning in Healthcare
• Dental Technology

Background:

• Dental air turbine handpieces (DATHs) possess complex internal mechanisms, necessitating advanced monitoring for operational reliability and patient safety.
• Traditional diagnostic methods for DATHs may not adequately capture subtle changes indicative of impending failure.

Purpose of the Study:

• To develop and evaluate predictive models for monitoring the health and diagnosing failures in DATHs.
• To utilize temporal convolution networks (TCNs) for their capabilities in time-series analysis, memory transmission, and fast convergence.

Main Methods:

• Vibration signals were captured using an accelerometer during simulated milling operations with a DATH.
• Data were used to train and compare prediction models, including TCN, 1D convolutional neural network (CNN), and long short-term memory (LSTM).
• A diagnostic health classification (DHC) framework was established to evaluate model performance.

Main Results:

• The TCN model demonstrated lower prediction error compared to 1D CNN, benefiting from its memory framework and avoiding the vanishing gradient problem.
• TCN outperformed LSTM, requiring less historical data for effective diagnostic ability.
• High accuracy was achieved in predicting DATH health and failure modes using vibration signals.

Conclusions:

• TCN-based predictive models can accurately assess the health status and predict failures of DATHs before clinical use.
• The developed model serves as a valuable tool for predicting deterioration patterns and estimating the remaining useful life of dental handpieces.
• Integration of embedded sensors with TCN models can significantly enhance the proactive maintenance and safety of dental equipment.