Detection Welding Performance of Industrial Robot Using Machine Learning

Abstract

Automated welding robots have become an essential component in manufacturing industries due to their precision and increased productivity. Furthermore, they eliminate the need for human workers to be exposed to toxic fumes and bright light during welding processes, ensuring worker safety. However, any failure in welding robots can lead to poor product quality and impact a company's credibility. Therefore, it is critical to predict the welding performance of these machines to maintain product quality. This study proposes the use of spectrogram image classification to classify the quality of welding products. Using a TM-1400 WGIII welding robot, 180 welding datasets were collected to determine the conditions of the welding robot based on different welding wire lengths. Current and voltage signals of the welding robot were studied to identify suitable input parameters and machine learning algorithms were utilized to improve the accuracy of the predictive welding quality. The experiment resulted in an accuracy of 89% based on a 70% training sets and 30% testing sets of data, indicating that spectrogram image classification is an effective technique to monitor the condition of welding robots, improve welding quality, and increase product quality control. This method can be beneficial for both small- and large-scale industries such as automotive, energy, and construction.