Abstract
As more and more critical parts are being fabricated via additive manufacturing (AM), there is an increasing need to ensure that the geometric and material properties of the parts conform to certain requirements. Currently this need is addressed by expensive, time consuming, and environmentally wasteful destructive testing of physical replica components. However, with more and more connected sensors being installed in AM machines, new possibilities are emerging for creating digital replicas, or digital twins, of the physical parts, inline. Such digital twins can be used as a basis for qualifying parts digitally, removing the need for excess physical parts. In this talk we will discuss approaches for creating digital twins from inline sensor data, focussing on the problem of extracting the geometry from volumetric images. To do this we transfer techniques developed in the medical imaging domain, to the setting of AM. A digital twin is never an exact representation of the physical object or process, as there are unknowns inherent in both the dynamics of the process and in the capture of the sensor data. We therefore also investigate the use of techniques such as Kalman filters and Markov random fields to capture these uncertainties in the digital twin.
