Digital coupling of multiscale analyses in modelling and monitoring

已建基础设施的安全运行需要强大的预测模型来预测系统或结构特性。这些预测模型必须根据所考虑的现有结构的当前状态不断更新，特别是如果预期寿命应为几十年的话。这种程序只能考虑建模假设与实际情况和不确定参数或分散参数之间的偏差。这就要求，除了各自的结构监测系统外，还需要一个适当的数值模型，以足够的精度描述可能发生的损害情况。在本项目中，采用了多尺度方法来描述结构损伤。起始点是宏观尺度上完整结构的数值模型。根据监测数据确定的临界区采用混合多尺度模拟战略，在中尺度上以更高的分辨率进行模拟。通过应用无损检测方法获得与当前结构状态有关的信息，这些信息是在中尺度上进行局部模拟所必需的。为了实现监测系统和数值模型之间的信息流，必须建立数字耦合。因此，理论驱动的机器学习方法被应用到这一背景下。描述输入和输出数据之间关系的模型可以通过神经网络从实验数据中创建。然而，这些基于数据的模型并不一定与物理现象有关。因此，不能直接与所考虑的系统的结构行为放在一起。计划在项目内开发的方法不仅向元模型提供关于从测量数据获得的输入和输出参数的信息，而且还向元模型提供用于数值模型的基于物理的关系。通过这种方法，建立了数字孪晶的实验研究与数值结构模型之间的耦合。这种耦合不仅允许从测量数据中识别模型参数。它还允许反向信息流，以基于通过模型获得的模拟结果来控制移动无损检测设备。要开发的方法需要创建多尺度的数值模型、用于数值耦合的工具以及对不同尺度的结构进行实验分析。在这种情况下，这个信息链的组件首先用可以在实验室受控条件下测试的结构元素来验证。在第二步中，将对DFG优先方案2388的参考结构进行验证。

A safe operation of built infrastructure requires robust prognosis models for the prediction of system or structural properties. These prognosis models have continuously to be updated with respect to the current state of the considered existing structures especially if the expected lifetime should cover several decades. Deviations of the modelling assumptions from real situation and uncertain or scattering parameters can only be taken into consideration by such a procedure. This requires, beside a respective structural monitoring system, a suitable numerical model that describes possibly occurring damage scenarios with sufficient accuracy. In this project, a multiscale approach is applied to describe structural damage. Starting point is a numerical model of the complete structure on macroscale. Critical zones that were identified based on monitored data are modelled with a higher resolution on the mesoscale using a hybrid multiscale modelling strategy. The information related to the current structural state that is necessary for local modelling on the mesoscale is obtained by the application of non-destructive testing methods.For the information flow between monitoring system and numerical model, digital coupling has to be established. Therefore the method of theory-driven machine learning is applied in this context. Models describing relations between input and output data can be created from experimental data by means of neuronal networks. However, these data-based models are not necessarily related to physical phenomena. And therefore, can not directly be put into context with the structural behaviour of the considered system.The approach intended to be developed within the project, feeds the meta models not only with information about input and output parameters derived from measured data but also with physics-based relations that are used for the numerical models. In this way, the coupling between experimental investigations and the numerical structural model of a digital twin is established. This coupling does not only allow for the identification of model parameters from measured data. It allows also an information flow in reverse direction to control mobile non-destructive testing equipment based on simulation results obtained by means of the model.The methodology to be developed requires the creation of numerical multiscale models, tools for the numerical coupling as well as experimental analyses of the structure on different scales. In this context, the components of this information chain are first validated with structural elements that can be tested under controlled conditions in the laboratory. In a second step, a validation will be carried out on the reference structure of the DFG priority programme 2388.