In-situ metrology, inverse analysis and first-principle modelling for the physics- and data-based prediction of highly non-linear material behaviour and failure in manufacturing technology

剪裁等断续制造过程使用机械机构来打破工件的材料凝聚力。在这个过程中，材料发生了高度的非线性变形和复杂的热粘塑性效应，以及材料损伤和失效的机理，这是基于物理状态变量和材料微观结构之间的相互作用而产生的，这些因素决定了制造过程的质量。由于缺乏对温度、应变和应力场等状态变量以及切削速度、切削间隙和刃口几何形状等工艺变量之间的相互依赖关系的了解，导致所有现有的建模方法都不能以可推广的预测方式使用，并且在局部狭窄的工艺窗口之外失去有效性。虽然过程中典型的高度非线性的材料行为以及由此产生的广泛的状态变量严重限制了现有局部表征和建模方法的有效性，但申请者恰恰将这一过程属性解释为关于过程和材料行为的全面信息的机会和来源。具体地说，该研究项目将把机器学习的新概率方法结合起来，用于逆材料参数和材料模型识别，基于高保真有限元方法的第一原理建模方法，以及以新方式进行的高分辨率现场测量方法，以便能够对材料行为进行准确和全面的预测，即在整个过程参数空间有效的预测，原则上也可以转移到其他过程。虽然以前的研究方法没有始终如一地将现代数值建模和分析方法与创新的实验测量和评估技术相结合，而是主要相互独立地考虑它们，但只有拟议的基于物理和基于数据的组合方法才允许充分开发关于预测性和概括性材料和过程建模的潜力。

Severing manufacturing processes such as shear cutting use mechanical mechanisms to break up the material cohesion of work pieces. In this process, highly non-linear deformations and complex thermo-visco-plastic material effects as well as mechanisms of material damage and material failure occur, which are based on the interaction between physical state variables and the material microstructure, which decisively determine the manufacturing process quality. The lack of understanding of the interdependencies between state variables such as temperature, strain and stress field as well as process variables such as cutting speed, cutting clearance and cutting edge geometry has the consequence that all existing modelling approaches cannot be used in a generalizable predictive way and lose their validity outside a local, narrow process window. While the highly non-linear material behaviour typical for the process and the resulting broad spectrum of state variables severely limits the validity of existing local characterisation and modelling approaches, it is precisely this process property that is interpreted by the applicants as an opportunity and source of comprehensive information on process and material behaviour. Specifically, the research project will combine novel probabilistic methods of machine learning for inverse material parameter and material model identification, first-principle modelling approaches based on high-fidelity finite element methods as well as high-resolution in-situ measurement methods in a novel way in order to enable accurate and global predictions of material behaviour, i.e. predictions that are valid in the entire process parameter space and in principle also transferable to other processes. While previous research approaches do not consistently combine modern numerical modelling and analysis methods and innovative experimental measurement and evaluation techniques, but rather consider them predominantly in isolation from each other, only the proposed combined physics- and data-based approach allows to fully exploit potentials with regard to predictive and generalisable material and process modelling.