Enhanced Particle Filtering for Parameter Estimation

Parameter estimation for trending analysis is a generic problem with broad applications. In manufacturing, trending analysis can be applied to tool wear estimation, which not only affects the tool life but also and the quality of a machined product. Consequently, in-situ tool wear monitoring and remaining tool life prediction play a significant role in ensuring precision and cost-effective manufacturing. This paper presents a Bayesian approach to flank wear rate prediction by linking vibration data measured during machining to actual tool wear. Variation in the measured data is aggregated based on the Kullback-Leibler divergence, which provides a measure for the distance between the current and initial probability distributions of measurement data, when the tool is new and has no wear. Estimation of the tool wear in the state space based on the distribution distance is realized by particle filtering (PF), which has shown effectiveness for non-linear and non-Gaussian system estimation. A new resampling scheme is proposed to overcome the sample impoverishment problem in sequential importance resampling (SIR). The proposed scheme has shown to more accurately define the confidence interval and improve prediction accuracy. Based on the distance calculation, the remaining useful life (RUL) of the tool for a given threshold is recursively predicted. The developed method is evaluated using a set of benchmark data measured from a high speed CNC machine performing milling operations, and good results are obtained by comparing the predicted wear state with off-line tool wear measurement. The paper further presents a comparative study between the proposed method and standard SIR method.