Frequency Selection for On-line Identification of Welding Penetration through Audible Sound

Abstract

Audible sound sensing technology is a possible key to real-time monitoring and controlling of welding quality and process for intelligent manufacturing of robotic welding. In this paper, the selection of frequency components from audible sound signal was carefully researched for identifying three types of seam penetration, e.g., under penetration, normal penetration and burning through by means of Principal Component Analysis (PCA). The selected principal components were carefully analyzed. Then, the ability of detecting and identifying different weld defects was thoroughly discussed and demonstrated. At the end, the degree of data redundancy and noise were quantitatively evaluated and discussed. In this paper, PCA has been verified to be able to effectively reduce the feature dimension and accurately identify the different weld defects using the selected features in real-time for robotic welding.