Development of a machine learning model for deviation from trajectory detection in multi-pass TIG welding in a narrow gap

This paper presents a novel approach to detect deviations in welding trajectory, using a Machine Learning model that predicts the occurrence of defects during narrow gap operations. A regression model is developed and trained using experimental data acquired from a number of synchronised sensors used in the TIG process with filler metal. Two primary sources of data are employed. The voltage, current and other electrical parameters are monitored as well as the changing dynamics of the weld pool, which are monitored by two cameras stationed behind and in front of it. An image processing algorithm is then used to extract the contour and geometrical characteristics of the weld pool, including its length, width, and area. The data is organized, stored, and referenced in a database for swift and easy processing. The database is created to train, test and validate the Machine Learning (ML) model. During welding, artificial deviations are introduced to examine the ML model's ability to predict the welding trajectory deviation. Various training models are evaluated to analyse their predictability within a defined set of parameters, in order to anticipate the occurrence of defects.