A study on real-time monitoring of GMA welding quality continuous hidden markov model

Welding technology is a fundamental basis that determines the performance and life of nuclear power plants. The piping is more complicated than normal industrial plants and has more welded parts. It is necessary to have overall technical capabilities ranging from design, manufacturing, and construction to cultivate a competitive foreign nuclear industry. As the power generation equipment manufacturing part consists of welding more than half of the whole process, welding technology has an absolute influence on enhancing competitiveness. In particular, high-performance materials and higher-level welding techniques are required to prevent corrosion defects, etc., occurring in the use of equipment. When manufacturing nuclear power plants of nuclear reactor pressure vessels and piping systems, welding techniques in which welding defects are extremely reduced are required. In this paper, a prediction model to determine the correlation between welding parameters and spatter in the GMA welding process is developed and verified. Based on model using the CHMM of machine learning, a model is developed to predict the defects using correlation analysis of process variables and spatters. To verify the reliability of the developed defect prediction model, the welding spatter obtained from the actual welding experiment and the prediction model are compared and analyzed, and the accuracy is evaluated. © ICROS 2020.