A tool condition monitoring method based on two-layer angle kernel extreme learning machine and binary differential evolution for milling. (15th December 2020)

• Record Type: Journal Article
• Authors: Zhou, Yuqing; Sun, Bintao; Sun, Weifang
• Is Part Of: Measurement. Volume 166(2020)
• Publication Date: 2020-12-15
• Subjects: Tool condition monitoring -- Milling process -- Two-layer angle kernel extreme learning machine -- Binary differential evolution; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2020.108186
• Languages: English
• ISSNs: 0263-2241

Abstract

Highlights: In comparison with KELM, two-layer angle KELM has a higher prediction accuracy. TAKELM don't require manual presetting of the kernel function and its hyperparameter. TAKELM-BDE can obtains the fewest feature parameters with a small prediction error. Abstract: Tool condition monitoring (TCM) in numerical control (NC) machines plays a significant role in ensuring manufacturing quality. TCM based on multiple sensors can provide more information related to tool condition and is a topic of great interest. A novel TCM method for milling processes based on a two-layer angle kernel extreme learning machine (TAKELM) and binary differential evolution (BDE) was proposed. The TAKELM could enhance inherent feature extraction in microarray data and don't require manual presetting of the kernel function or optimization of its hyperparameter. The BDE algorithm was applied to search the optimal feature parameter combinations in multi-domain alternative feature parameters to achieve the fewest feature parameters satisfying a very small prediction error. Finally, the performance of the proposed TAKELM- BDE method was verified on two milling TCM experiments (one open-access benchmark TCM data set and one TCM experiment) by comparison to the KELM- BDE, PCC- TAKELM, and mRMR- TAKELM methods. The results indicated that the mean absolute error, mean absolute percentage error, and root mean square error of the proposed method are less than 0.005 on two milling TCM experiments, and outperforms the other three methods.