The digital representation of physical assets and process steps by digital twins is key to address pressuring
challenges like adaptive manufacturing or customised production. Recent breakthroughs in the field of
digital twins and Edge-based AI already enable digital optimization of individual process steps. However,
high-value goods typically include multiple step process chains including a broad range from generative and
additive processes over several steps of material removal up to assembly. Therefore, a digital twin over the
holistic process chain is necessary. While even the set-up of representative twins for a single step is already
challenging, a concept for monitoring of the interaction and overall quality control of holistic process chains
does not exist yet. The paper introduces a machine-learning method based on probabilistic Bayesian
networks to monitor the »digital twin quality« of coupled digital twins which includes several sub-instances
of digital twins. The approach identifies the contribution of each instance to the overall prediction quality.
Furthermore, it is possible to give a range-estimation for the prediction accuracy of the individual subinstances. It is therefore possible to identify the most influential sub-instances of digital twins as well as their
individual prediction quality. With the help of this information, the quality of the digital twin can be
improved by considering individual sub-instances in a targeted manner. Finally, a preview emphasises the
potential benefits of the quantum computing technology when dealing with parallel computation of largescale inference models.
|