Reinforcement learning and optimization based path planning for thin-walled structures in wire arc additive manufacturing

Abstract

A well-designed deposition path is one of the basic prerequisites for the successful fabrication of a component by deposition based additive manufacturing processes. Three main approaches are currently used to determine the deposition path. First, these are general path templates that are applied to the entire geometry. Nevertheless, this approach suffers from poor adaptability to the geometry. Second, they are algorithms where it is necessary to divide the geometry into sub-parts, which are then filled either by general path templates or by paths derived, e.g., from the signed distance function. These often require human intervention and may fail to find a suitable deposition path. Third, there are planning strategies that deal only with a particular topologies, and are not transferable to other geometries. A developed path planning framework named RLPlanner, which makes use of reinforcement learning as well as automatized prepossessing and Sequential Least Squares Programming optimization method, addresses these drawbacks. This solution enables fully automatic deposition path planning for thin-walled structures in wire arc additive manufacturing. In addition, the framework is able to vary the welding speed with the wire feed rate and thus influence the size of the weld bead leading to better adaptability to the geometry.