A novel current sensor indicator enabled WAFTR model for tool wear prediction under variable operating conditions

Abstract

The health indicators (HIs) were extracted from the current sensor to represent the tool wear progression. The extracted HIs were found poorly correlated with the progression of tool wear as the raw current sensor signal was susceptible to the influence of other parts and structures in the machine tool. Hence, this paper proposed a novel current sensor-based HI that utilized the mean of inverse hyperbolic cosine function fitted to an envelope of the current signal to improve the correlation. Using the extracted HIs, many bespoke machine learning (ML) models have been developed by researchers. However, these models have many hyperparameters, difficult to interpret and especially poor prediction accuracy has been observed under variable operating conditions. This study overcame these issues by proposing a Weibull Accelerated Failure Time Regression (WAFTR) model, which combines process parameters data with HI for improving the prediction accuracy under variable operating conditions. This model mapped a functional relationship with tool wear in the form of probability density function to identify best HIs and acceleration/deacceleration factors which makes it interpretable. The acceleration/deacceleration factors are useful to deaccelerate the tool wear evolution by controlling the specific values of the machining parameters.