Methodology for automated and computer-aided work plan generation using voxel-based accessibility analysis through reinforcement learning in one-off manufacturing

One-off manufacturing is characterized by high workpiece-related planning costs. Current process chains are designed manually and based on implicit knowledge, leading to non-optimal process chains on the one hand, while also disregarding alternative manufacturing options on the other. This paper presents a software-based approach for automated work plan generation that follows a feature-independent process chain design based on identified deficits. For this purpose, the workpiece is divided into discrete voxels to be able to represent the individual intermediate states of manufacturing. The methodology is divided into five phases: The modelling and training of agents, the analysis and discretization of workpiece input information, the initial workpiece definition and technology configuration, the simulation and verification as well as the combination of alternative technology configurations. In the concept reinforcement learning is used for optimization which requires previous training of the agents applied. The agents trained for this approach represent the intelligent clamping system and machining tool. A prototype implementation provides promising results, particularly in the automated identification of clamping positions, regardless of geometry and process-related accessibility, using a prismatic workpiece model. While 97.6 % of the machined elements have been successfully subtracted, there is still a need for optimization to improve the accuracy of the model. In conclusion, the approach offers a promising path to efficient and automated process chain planning of complex workpieces in one-off manufacturing.