Inline image-based reinforcement detection for concrete additive manufacturing processes using a convolutional neural network

Abstract

Within the scope of additive manufacturing of structural concrete components, the integration of reinforcement provides an inevitable opportunity to enhance the load bearing capacity of the components. Besides the rebar integration itself, ensuring as-planned concrete cover is key to achieve a stable and long-term legally permissible integration. The thickness of the as-built concrete cover however is unpredictably altered during printing by the varying material behaviour of the printed concrete. In addition, the lack of opportunities to anchor reinforcement elements before printing can lead to a displacement of reinforcement during printing. In this publication, we present an approach for determining the position of reinforcement elements within additively manufactured components without post-process measurement steps. During the printing process, RGB images and depth camera data are recorded by a camera mounted to the print head. Subsequently, a neural network is employed to distinguish between reinforcement structures and the deposited material within the coloured image. By overlaying the colour image data with the depth information a 3D point cloud is generated, within which the reinforcement is marked.