Acoustic signal analysis for detecting defects inside an arc magnet using a combination of variational mode decomposition and beetle antennae search. (July 2020)

• Record Type: Journal Article
• Authors: Huang, Qinyuan; Xie, Luofeng; Yin, Guofu; Ran, Maoxia; Liu, Xin; Zheng, Jie
• Is Part Of: ISA transactions. Volume 102(2020)
• Page Start: 347
• Page End: 364
• Publication Date: 2020-07
• Subjects: Arc magnet -- Internal defect -- Variational mode decomposition -- Beetle antennae search -- Naive Bayes classification; Engineering instruments -- Periodicals; Engineering instruments; Periodicals; Electronic journals; 629.805
• DOI: 10.1016/j.isatra.2020.02.036
• Languages: English
• ISSNs: 0019-0578

Abstract

Abstract: An accurate, rapid signal analysis is crucial in the acoustic-based detection for internal defects in arc magnets. Benefiting from the adaptive decomposition without the mode mixing, variational mode decomposition (VMD), has emerged as a promising technology for processing and analyzing acoustic signals. However, improper parameter settings are the root cause of inaccurate VMD results, while existing optimization methods for VMD parameters are only applicable to a single signal with exclusive signal characteristics, rather than different signals with similar features. Therefore, we developed a new acoustic signal analysis method combining VMD, beetle antennae search (BAS), and naive Bayes classification (NBC), and then applied it for detecting internal defects of arc magnets. In this method, multiple optimizations for different signals are simplified to a one-time optimization for the whole signal group by a specially designed parameter-related fitness function. Since the coordinates of the function maximum value in a parameter space correspond to the unified parameter setting generating the overall optimal processing effect for all signals, BAS is introduced to achieve a rapid search of coordinates. With the obtained unified parameter setting, each acoustic signal of arc magnets can be consistently processed by VMD. Next, two modes stemmed from VMD are screened out by an energy threshold, and their specific frequency information is extracted as features representing the internal defects. NBC is carried out to learn and identify the extracted features. The experimental validation of the proposed method was conducted by detecting various arc magnets. Experimental results indicate that the identification accuracy reaches 100% and the detection speed per a single arc magnet approximately ranges between 1.7 and 4.5 s. This work provides not only a new strategy for the parameter optimization of VMD, but also a practical solution for the internal defect detection of arc magnets.