Abstract
The online prediction of friction stir welding quality is an important part of intelligent welding. In this paper, a new method for the online evaluation of weld quality is proposed, which takes the real-time temperature signal as the main research variable. We conducted a welding experiment with 2219 aluminum alloy of 6 mm thickness. The temperature signal is decomposed into components of different frequency bands by wavelet packet method and the energy of component signals is used as the characteristic parameter to evaluate the weld quality. A prediction model of weld performance based on least squares support vector machine and genetic algorithm was established. The experimental results showed that, when welding defects are caused by a sudden perturbation during welding, the amplitude of the temperature signal near the tool rotation frequency will change significantly. When improper process parameters are used, the frequency band component of the temperature signal in the range of 0~11 Hz increases significantly, and the statistical mean value of the temperature signal will also be different. The accuracy of the prediction model reached 90.6%, and the AUC value was 0.939, which reflects the good prediction ability of the model.
Highlights
According to the requirements of the engineering application, we evaluated the weld quality based on the tensile strength of the joints
The energy value obtained by the wavelet packet method is used as the influencing variable to evaluate the weld quality
The energy data are used in the genetic optimization algorithm (GA)-least-square support vector machine (LSSVM) prediction model
Summary
Automation and intelligence in the welding process is an important direction for the development of the FSW process. The main focus of this research is how to detect and evaluate the weld quality in the welding process. The traditional methods of FSW inspection and evaluation, such as ultrasonic inspection, X-ray inspection, and coloring inspection, are difficult to implement in the welding process. Accurate online monitoring of the welding process has become an important focus of research [1,2,3]. Verma et al [5] used rotational speed, traverse speed, and tilt angle as input variables to study the ultimate tensile strength of the welding seam using machine learning methods, such as Gaussian regression (GPR), support vector machine (SVM), and artificial neural network (ANN), and realized the purpose of welding process optimization. Sumesh et al [6] used the current amplitude signal to establish a direct connection with the weld quality, extracted the statistical features of the original data through data-mining software, established a J48 and random forest algorithm, and reported the classification effect on the weld quality
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
