Abstract: The deposition process of robotic wire arc additive manufacturing (WAAM) is intense and complex, and the process parameters fluctuate significantly. It is difficult for a single sensor signal to accurately reflect the forming state, and forming quality control still faces great challenges. To address this problem, a multimodal monitoring platform integrating vision, electrical signals, and temperature images was constructed. Based on deep learning and image processing technologies, the morphological features of the molten pool were extracted, and the distribution information of current, voltage, and temperature during the cladding process was collected, thereby realizing the synchronous extraction and fusion of multi-source information features. By designing multiple groups of additive manufacturing process experiments, a multi-source feature dataset containing 642 groups of samples was established, and a cladding layer quality prediction model was constructed based on the ensemble learning method of light gradient boosting machine (LightGBM). The results indicate that the proposed method can effectively improve the accuracy of forming quality prediction; the macro-average accuracy of the prediction model reaches 90.4%, which provides strong technical support for realizing intelligent quality control in the robotic WAAM process.