An LSTM-based measurement method of 3D weld pool surface in GTAW

The 3D weld pool surface is affected by thermo-mechanical coupling, which significantly determines the quality of the joint. However, it is extremely difficult to achieve its measurement because of the super bright arc and high temperature existing during welding. In this paper, a structured light sensing system is established to measure the 3D weld pool surface in gas tungsten arc welding (GTAW). A novel measurement method is proposed based on the long-short-term memory (LSTM) neural network in the field of deep learning. The training and test sets required for network modeling are output by a pre-built simulation model. The measurement accuracy of the method is verified by convex mirror tests and GTAW tests, and the average time that the model cost to finish onetime calculation is 2.73 ms in GTAW tests, which shows the feasibility and superiority of the proposed method in real-time measurement.

Automated summary

A structured light sensing system was established to measure the 3D weld pool surface in gas tungsten arc welding (GTAW), with a novel measurement method proposed based on the LSTM neural network in deep learning. The feasibility and superiority of this method in real-time measurement was demonstrated through verification of measurement accuracy and computational time.