A data-driven prognostic approach based on wavelet transform and extreme learning machine

Abstract

The monitoring of a cutting tool is needed for the prediction of impending faults and estimating its Remaining Useful Life (RUL). Implementing a robust Prognostic and Health Management (PHM) system for a high speed milling CNC cutter remains a challenge for various industries to reach improved quality, reduced downtime, increased system safety and lower production costs. The purpose of the present paper is health assessment and RUL estimation of the cutting tool machines. To do so, an approach based the use of Wavelet Packet Transform (WPT) and Extreme Learning Machine (ELM) for tool wear condition monitoring is proposed. Among the main steps is feature extraction where the relevant features of raw data are computed in the form of nodes energy using WPT. The extracted features are then fed to the learning algorithm ELM; the main idea is that ELM involves nonlinear regression in a high dimensional feature space for mapping the input data via a nonlinear function to build a prognostics model. The method was applied to real world data gathered during several cuts of a milling CNC tool. Results showed the significance performances achieved by the WPT and ELM for tool wear condition monitoring.