An automated feature extraction method with application to empirical model development from machining power data. (1st June 2019)

• Record Type: Journal Article
• Authors: Pantazis, Dimitrios; Goodall, Paul; Conway, Paul P.; West, Andrew A.
• Is Part Of: Mechanical systems and signal processing. Volume 124(2019)
• Page Start: 21
• Page End: 35
• Publication Date: 2019-06-01
• Subjects: Feature extraction -- Machining -- Hidden Markov model (HMM) -- Response Surface Methodology (RSM) -- Power data -- Time-series segmentation -- Energy decomposition -- Hierarchical clustering -- End-milling -- Empirical model; Structural dynamics -- Periodicals; Vibration -- Periodicals; Constructions -- Dynamique -- Périodiques; Vibration -- Périodiques; Structural dynamics; Vibration; Periodicals; 621
• DOI: 10.1016/j.ymssp.2019.01.023
• Languages: English
• ISSNs: 0888-3270

Abstract

Highlights: A novel method to automate feature extraction from RSM DoE data is presented. Hidden Markov Models, Hierarchical Clustering and Dynamic Time Warping are utilised. The method is validated through Central Composite Design, end-milling experiments. The material cutting energy is extracted with 1.12% mean absolute error. The spindle acceleration energy is extracted with 3.33% mean absolute error. Abstract: Machining shop floor jobs are rarely optimised for minimisation of the energy consumption, as no clear guidelines exist in operating procedures and high production rates and finishing quality are requirements with higher priorities. However, there has been an increased interest recently in more energy-efficient process designs, due to new regulations and increases in energy charges. Response Surface Methodology (RSM) is a popular procedure using empirical models for optimising the energy consumption in cutting operations, but successful deployment requires good understanding of the methods employed and certain steps are time-consuming. In this work, a novel method that automates the feature extraction when applying RSM is presented. Central to the approach is a continuous Hidden Markov model, where the probability distribution of the observations at each state is represented by a mixture of Gaussian distributions. When applied to a case study, the automated extracted material cutting energies lay within 1.12% of measured values and the spindle acceleration energies within 3.33% of their actual values.