Modeling of civil engineering structures with particular attention to incomplete and uncertain measurement data by using explainable machine learning (MoCES)

土木工程结构的状况是随着它的老化而日益迅速地退化。预防衰老的行动越早采取越有效。为了延长复杂结构的可用性，在比今天常见的更早的阶段需要更多的信息。为了走向预测性维护，需要对收集、融合和评估所有几何、材料、应力和老化数据的方法进行基础研究。数字化，关于数字双胞胎的产生，在这种背景下正具有全新的意义。它能够对运行和维护所需的所有数据进行组合和实时评估。我们建议的主要目标是研究和开发用于复杂建筑结构自动建模的新方法和新流程。其目的是融合各种各样的数据流，并考虑到它们的不确定性和不完整性。建模将在机器学习方法的基础上实现，将通过一个解释组件进行扩展，以重现对象建模和重建意义上的结论。对象重建和建模的重点是形成对象监控的基础-在这里，我们完全解决了DFG优先计划“Hundert plus”的目标。“Hundert plus”的全球目标是有条不紊地开发一种真实、物理物体(建筑物)的自适应、智能和数字表示(数字孪生)。最后，该模型将与整个使用寿命内的建筑物监测的测量数据相联系，并集中为预测性、数字化的建筑物管理提供压缩信息。本文提出的用于监测任务的自动语义对象重建和建模在今天的文献中或在实际应用中并不存在，因此将为快速有效地监测大型土木结构做出重要的贡献。

The condition of a civil engineering structure is characterised by increasingly rapid degradation as it ages. A preventive action against aging is more successful the earlier it is taken. To prolong the usability of complex structures, much more information is required at a much earlier stage than is common today. To move toward predictive maintenance, fundamental research is needed on the methods of collecting, fusing, and evaluating all geometry, material, stress, and aging data. Digitisation, regarding the generation of a digital twin, is taking on a completely new significance in this context. It enables the combination and real-time evaluation of all data required for operation and maintenance. The main goal of our proposal is to research and develop new methods and processes for the automated modeling of complex building structures. The aim is to fuse a wide variety of data streams and to take into account their uncertainty and incompleteness. The modeling, which will be realised on the basis of machine learning methods, will be extended by an explanatory component so that conclusions in the sense of object modeling and reconstruction are reproducible.The object reconstruction and modeling has the focus to form the basis for object monitoring - here we fully address the goal of the DFG Priority Programme "Hundert plus". The global aim of "Hundert plus" is the methodical development of an adaptive, intelligent, and digital representation (digital twin) of real, physical objects (buildings). In the end the model will be linked with measurement data from building monitoring over the entire service life and centrally provides compressed information for predictive, digital building management.An automated semantic object reconstruction and modeling for monitoring tasks as proposed does not exist in the literature or in practical use today and will therefore provide a significant and important contribution to the rapid and efficient monitoring of large-scale civil structures.