Study on Real-Time Porosity Defect Detection Through Neural Network Structure Optimization using Genetic Algorithm in GMAW

Abstract

Zinc-coated steel sheets are widely applied as automotive chassis parts because they have high corrosion resistance and good compatibility. However, in the gas metal arc welding (GMAW) process, serious porosity defects occur due to zinc vapor generated during welding, which causes problems such as durability or productivity reduction in the welded structure. To secure weldability and productivity, it is essential to secure monitoring technology that determines whether porosity defects are generated in real-time. To solve this problem, this study provides a method of extracting feature variables from arc voltage signals generated during welding and optimizing the hyper parameters of the porosity detection algorithm deep neural network (DNN) which be learned with feature variables by applying genetic algorithm (GA). To verify the performance of the proposed method, as a result of applying it to the optimized DNN model using the experimental data of the GMAW experiment using a high-strength zinc-coated steel sheet, a prediction accuracy of 93.1 % was derived, which is improved by 3.60 % than DNN model from previous research.