Deep Learning Vision Improves Hot Forging Workpiece Tracking

In industrial settings like hot forging, accurately tracking workpieces is crucial for quality control and process optimization. However, the harsh conditions, including high temperatures and material degradation, make direct tracking difficult. Researchers have developed a novel framework that addresses this by focusing on the equipment handling the workpieces rather than the workpieces themselves. This system employs multiple static 2D cameras to observe handling equipment. Deep learning models estimate the equipment's 3D coordinates and recognize actions like grasping and releasing. These activities are then validated by event-driven finite state machines, which continuously update the workpiece's state and location. A key innovation is a keypoint-guided attention mechanism within a 3D convolutional neural network, enhancing activity recognition by focusing on relevant equipment parts. Evaluated in a real hot forging factory, the framework demonstrated 100% event detection accuracy within a 33-second window, a mean localization error of 317.8 mm, and a system latency of 21 seconds. This approach effectively links vision-based perception with logical, event-driven reasoning, providing a robust solution for visualizing workpiece transfers and analyzing equipment operations.