Misalignment monitoring between electrodes and rivet in three-sheet resistance element welding based deep neural network using dynamic resistance

Abstract

As use of aluminum alloys with steel materials in automobile parts increases with the reduced weight of car bodies, the demand for dissimilar welding of these materials is also increasing. Welding of these materials is limited by problems including formation of an intermetallic compound; thus, use of resistance element welding (REW) as a countermeasure is increasing. In the REW process, a rivet of steel material is inserted into an aluminum alloy and welded with resistance heat generated by current flowing between the steel material and the rivet. A nugget is formed between the steel materials, suppressing formation of intermetallic compounds. However, REW quality deteriorates as misalignment between the rivet and the electrode increases. As the alignment is uncertain, good welding quality cannot be guaranteed. Thus, a technique for monitoring the misalignment of rivets and electrodes is essential for application of three-sheet REW. In this study, three-sheet (steel-aluminum-steel) REW was performed on a riveted aluminum alloy and two types of steel, and welding quality was evaluated according to the misalignment distance. In addition, a pre-contact process was introduced before welding to monitor the misalignment between the electrode and the rivet, and the dynamic resistance waveform based on the misalignment distance was analyzed. The dynamic resistance waveform was used in a deep neural network (DNN)-based misalignment prediction model to monitor misalignment of electrodes and rivets; the prediction accuracy of the DNN model was 97.7%. In addition, as a result of checking the mean absolut error (MAE), root mean square error (RMSE), and mean percentage error (MPE) to evaluate the performance of the deep neural network model, they were 0.09, 0.01, and -0.11%, respectively, and the deep neural network model showed good performance.