Classification of weld quality in ultrasonic composite welding using vibration information

With global consensus on energy conservation and emission reduction, researching an efficient weld quality detection technology is crucial for promoting the application of ultrasonic welding in joining carbon fibre reinforced thermoplastic polymer (CFRTP) composites in the automotive lightweight design. In this research work, weld quality is classified based on lap-shear strength and weld indentation characteristics. The vibration signals from the surface of the workpiece during the welding process are captured. To extract meaningful information from the vibration signals, the Variational Mode Decomposition (VMD) method is employed, followed by the application of the Hilbert-Huang Transform (HHT) to obtain the HHT spectrum. By introducing the concept of information entropy, the differences in the HHT spectrum for various welds are accurately quantified. Extracting marginal spectrum amplitudes at various frequency bandwidths throughout the entire duration constructs the feature dataset used to train the Support Vector Machine (SVM) model. Then, the SVM model makes use of the Bayesian Optimization (BO) algorithm to find the optimal parameters for weld quality classification and recognition. The results exhibit outstanding performance, achieving an accuracy rate of 95.35%. This indicates that there is great potential for online detection of CFRTP welding quality through ultrasonic vibration signals.