ULTRASONIC HIGH-STRENGTH BOLT AXIAL FORCE EVALUATION BY MACHINE LEARNING-BASED TO WAVEFORM ANALYSIS

Abstract

 Bolt axial force of high-strength bolted joints decreases due to deterioration. One of the conventional methods to evaluate the bolt axial force is ultrasonic testing. However, quantitative evaluation of the bolt axial force is controversial in terms of accuracy, and some problems remain for practical use. This study attempts the accurate evaluation of the bolt axial force by ultrasonic testing. The time zone analysis of the ultrasonic waves indicated that the waveform in the initial time zone includes bolt axial force information. The parasitic discrete wavelet transform (P-DWT) was applied to improve the evaluation accuracy. As a result of targeting bolts manufactured in the same lot, machine learning using linear regression evaluated the unknown bolt axial force within an error of 6% or less. Therefore, it was shown that the proposed method of this study corresponds to the existing bolt and can evaluate the bolt axial force with high accuracy.