Machine vision system for digital twin modeling of composite structures

Abstract

Although the structural design of composite structures has already been carried out on a virtual level, composite mechanical properties remain sensitive to fiber orientation and therefore to the quality and reliability of the production process. Considering both manual single-unit manufacturing and advanced mass-unit fabrication, requirements on the production quality may differ, but certainty on the achieved result is crucial. A digital twin model, deterministically derived from produced parts, can be transferred into a virtual simulation environment to check for potential deviations of fiber alignment, resulting from variations in source material or composite production. Transferring that deterministic information into a virtual simulation environment allows for an estimation of the part’s structural potential despite any possible deviations by carrying out numerical simulation predictions on that model. This step of quality assessment can help reduce scrap parts by relying on simulation data that may demonstrate the feasibility of parts despite the containment of deviations with an otherwise uncertain impact. Therefore, further steps toward digitalization of the composite production process chain, especially on the characterization of the production quality, are aspired. In this contribution, a vision system based on a Microsoft Azure Kinect RGB-D camera is introduced which is used to digitalize the composite preform configuration from machine vision data by evaluating the achieved local fiber orientation as result of the complex preform draping process by digital image processing. A digital workflow is introduced that enables to feed the captured real-world data back into a digital environment where numerical simulations with the “as-built” fiber orientation can be carried out. The obtained results are used for assessing production quality and composite performance in the presence of possible deviations. The system, which consists of a camera array of consumer grade, can acquire real-world data and then transfer the data into a virtual environment.

From Pixels to Precision: A Survey of Monocular Visual Odometry in Digital Twin Applications

“…This technique is crucial for pose estimation [ 22 ]. Likewise, for the digital twin modeling of composite structures, the Azure Kinect camera is utilized to capture both depth and texture information [ 23 ]. Drone inspection imagery is instrumental in forming operational digital twins for large structures, enabling the creation and updating of digital twin models based on high-quality drone-captured images [ 24 ].…”