Global developments towards sustainable precision manufacturing are driving increasing demands for cost reduction, shorter lead times, and tighter tolerance windows. In particular, the energy-efficient yet highly accurate manufacturing of large-scale precision workpieces, for example, for large-scale machinery as well as the wind power and gas turbine industries, constitutes a key challenge. Thermo-elastic deformations, caused by fluctuating ambient temperatures and local transient heating effects, directly affect workpiece quality and can be controlled only to a limited extent with existing approaches. Active climatization of manufacturing environments, machines, and components effectively reduces thermo-elastic influences, but is often uneconomical due to high energy and infrastructure costs. Passive approaches or simple compensation models, in turn, are insufficient to reliably correct these effects. As existing models are not suitable for direct sensor coupling, computationally too expensive, or insufficiently accurate, a gap persists between available measurement data and real-time-capable deformation predictions. The objective of this work is the development of a hybrid model and sensor-based framework for virtual climatization that enables a state-based assessment of thermally induced deformations of large-scale workpieces in non-controlled production environments. The core artifact is a thermal digital twin that couples spatially distributed temperature measurements with different modeling strategies to make deformations predictable in context-related real-time. The work follows the methodological framework of Design Science Research and combines different sensing concepts with direct and inverse modeling based on physically consistent methods. For experimental validation, an automated large-scale workpiece test rig with local heating capability is developed and used in conjunction with two complementary geometric measurement systems (a mobile coordinate measuring arm and a laser tracker). Virtual and real deformation and temperature data are used for physical and simulation-based validation of the developed measurement and modeling strategies. They enable both a direct analysis of thermo-elastic effects and the derivation of time-dependent three-dimensional deformation fields. As an extension, an approach for the inverse reconstruction of internal temperature and heat flux distributions using Physics-Informed Neural Networks (PINNs) is developed. The PINN-based approach serves the exploratory investigation of inverse modeling methods and extends the FEM-based deformation analyses under unknown boundary conditions. The results demonstrate successful applicability to inverse heat transfer problems, while simultaneously identifying current limitations regarding convergence, data quality, and real-time suitability. Overall, this work provides a validated procedural model for the hybrid integration of real and model-based information sources and thus establishes a foundation for predictive, model-supported assessment of thermally induced deformations of large-scale precision workpieces in an industrial context.
Dominik Emonts (Thu,) studied this question.