Multimodal-based weld reinforcement monitoring system for wire arc additive manufacturing

Abstract

With the rise of big data and artificial intelligence, intelligent welding systems (IWS) are attracting more and more attention in machinery manufacturing. The work focused on weld reinforcement monitoring in wire arc additive manufacturing (WAAM). To track deposition status, a weld reinforcement monitoring or prediction system based on multi-modality was proposed. It is expected to be used as a functional component in the IWS in the future, helping to monitor deposition quality and detect anomalies. The reasons for adopting multi-modality can be summarized as follows. Firstly, the complex deposition process is hard to characterize with a single sensor. Secondly, equipped with a multi-sensor can reduce the impact of strong arc light or splash on the monitoring system. To fuse the welding information from different sensors, an implicit fusion method based on parameter sharing was adopted. The core of the proposed monitoring system is the neural network model. The input of the model includes two parts: molten pool images and infrared temperature field. Experimental results demonstrated that the model's weld reinforcement prediction error is around 0.05 mm. The study proved the effectiveness of multi-modal information for weld reinforcement monitoring systems.