Porosity Detection and Prediction Algorithm of Galvannealed High-Strength Steel Welds in Gas Metal Arc Welding Process

Abstract

ABSTRACT

Porosity Detection and Prediction Algorithm of Galvannealed High-Strength Steel Welds in Gas Metal Arc Welding Process

Seungmin Shin Department of Mechanical Engineering The Graduate School Hanyang University

The application of zinc-plated high-strength steel, which is resistant to corrosion and formability, has been increasingly applied in automobile body and chassis parts to lighten the vehicle and improve the safety. Most power-carrying chassis components, such as cross members and lower arms, are assembled using gas metal arc welding (GMAW) processes. However, owing to the zinc vapor generated from the zinc plated layer during the GMAW process, porosities such as a blowhole are formed in the welds, resulting in welding discontinuity. This causes a serious decrease in the strength of the welded portion as well as a decrease in the productivity. Such weld porosities are important factors in judging between good and defective products in actual production lines. In a conventional weld quality inspection, visual inspections and non-destructive inspection systems have been widely applied, despite the occurrence of real cost and production difficulties. Therefore, to cope with the recent smart processes pursued by Industry 4.0, it is necessary to develop a real-time welding quality system by introducing a quality judgment algorithm using artificial intelligence and data processing regarding the welding quality. This thesis is divided into three sections. First, lap joint experiments conducted using a galvannealed high-strength steel sheet, GA 590 MPa grade FB 2.3 mmt, which was applied to the automotive chassis parts during the GMAW process, are described. Zinc residues were confirmed using a semi-quantitative energy dispersive X-ray spectroscopy (EDS) analysis of the porosity in the weld. In addition, tensile shear tests were conducted to evaluate the weldability. Furthermore, the effects of the porosity, such as blowholes and cavities generated in the weld, on the tensile shear strength (TSS) were verified experimentally by comparing the porosity at the weld section of a tensile test specimen with that measured through radiographic testing. Second, using arc voltage and a welding current waveform, feature variables were extracted to predict the sizes of the external pits formed in the galvannealed high-strength steel sheet during the GMAW process. In addition, the feature variable was used as the input value to predict the pit size of the weld metal. As a result, the prediction performance was verified by applying it to a multiple linear regression model and an artificial neural network (ANN). Third, experiments were carried out using galvannealed high-strength steel sheet materials applied to automobile chassis parts through the GMAW process. Specifically, the feature variables were derived using a preprocessing by measuring the welding current and arc voltage waveforms in real time with a data collector. Furthermore, the correlations between the feature variables and porosity ratio were analyzed, and a deep-learning technique was employed to develop and evaluate an algorithm for detecting and predicting the weld porosity.

Keywords: Gas metal arc welding (GMAW), porosity, Tensile shear strength (TSS), feature variables, Multiple linear regression, Artificial neural network (ANN), Deep neural network (DNN).