Weld defect identification in friction stir welding through optimized wavelet transformation of signals and validation through X-ray micro-CT scan

Abstract

For control of real-time Internet of Things (IoT)-based remote welding process, continuous detection of defects occurring in the weld sample is of utmost importance so that welding parameters can be changed accordingly to avoid further occurrence of such defects. Time-frequency domain signal processing method, such as discrete wavelet transform (DWT), can be applied for detection of such defects. DWT continuously decomposes a signal into detailed and approximate coefficients through its associated filter banks and provides a time-frequency domain representation of a signal. Different levels of decomposition capture different frequency components, and hence, there is a need for optimization of the level of decomposition of force and power signals recorded during joining of two aluminum sheets by friction stir welding (FSW), for correct identification and localization of defects occurring in the process. Internal defects in the weld samples are further verified by CT scan images. Statistical tools have been used to study the variations in the DWT coefficients due to both internal and surface defects. An attempt has been made to compare between force and power signals as to which gives better defect detection.