Use of Long Short-Term Memory for Remaining Useful Life and Degradation Assessment Prediction of Dental Air Turbine Handpiece in Milling Process

Abstract

The complexity of the internal components of dental air turbine handpieces has been increasing over time. To make operations reliable and ensure patients’ safety, this study established long short-term memory (LSTM) prediction models with the functions of learning, storing, and transmitting memory for monitoring the health and degradation of dental air turbine handpieces. A handpiece was used to cut a glass porcelain block back and forth. An accelerometer was used to obtain vibration signals during the free running of the handpiece to identify the characteristic frequency of these vibrations in the frequency domain. This information was used to establish a health index (HI) for developing prediction models. The many-to-one and many-to-many LSTM frameworks were used for machine learning to establish prediction models for the HI and degradation trajectory. The results indicate that, in terms of HI predicted for the testing dataset, the mean square error of the many-to-one LSTM framework was lower than that that of a logistic regression model, which did not have a memory framework. Nevertheless, high accuracies were achieved with both of the two aforementioned approaches. In general, the degradation trajectory prediction model could accurately predict the degradation trend of the dental handpiece; thus, this model can be a useful tool for predicting the degradation trajectory of real dental handpieces in the future.

Keywords

• dental air turbine handpiece
• long short-term memory
• logistic regression
• remaining useful life