First Steps Towards an Intelligent Laser Welding Architecture Using Deep Neural Networks and Reinforcement Learning

摘要: Abstract To address control difficulties in laser welding, we propose the idea of a self-learning and self-improving welding system that combines three modern machine learning techniques. We first show ability deep neural network to extract meaningful, low-dimensional features from high-dimensional laser-welding camera data. These are then used by temporal-difference algorithm predict anticipate important aspects system's sensor The third part our proposed architecture suggests using these predictions learn deliver situation-appropriate power; preliminary results demonstrated simulator. intelligent introduced this work has capacity improve its performance without further human assistance therefore addresses key requirements industry. knowledge, it is combination Nexting with general value functions also usage for specifically production engineering general. This provides unique example how can be explicitly learned reinforcement support welding. believe would straightforward adapt approach other applications.