<p>Online monitoring of the welding process significantly impacts final quality and welding defects. Different methods, like vision sensing, are usually employed to monitor the welding process in real time. However, such a method requires special equipment and preprocessing of images during online monitoring, which is time-inefficient for industrial applications. Using generated signals, such as acoustic emission signals, can be an alternative method. However, this method performs poorly in high temperatures and requires high costs and special equipment. When monitoring the processes, the traditional artificial intelligence (AI) models often suffer from high-dimensional data, slow response, poor results in dynamic tasks, and noisy environments. To overcome these challenges, we propose a novel hybrid quantum neural network (HQNN) comprised of quantum and classical architectures using acoustic signals (sounds) to classify welding defects. The HQNN utilizes both the advantages of quantum machine learning and classical machine learning models. In our work, we built three classical machine learning models: artificial neural network (ANN), LightGBM (LGBM), and CatBoost (CB) for classifying welding defects, and compared the performance of our proposed HQNN model with these classical machine learning models. We also employed a data augmentation (DA) technique to increase our training data and improve the generalizability of our AI models. We trained all the models using the sounds generated during the Metal Active Gas (MAG) welding. Our HQNN outperformed all the classical models and achieved an accuracy of 98% on the test dataset with fewer parameters. The promising results of our proposed HQNN demonstrate that it can be applied in the robotic welding industry, which requires fast and precise monitoring.</p>