Property-controlled forging: analysis and experimental validation

在等温锻造中，刀具被加热到工件温度，使得近净成形对加工性较低的材料成为可能。最近，等温锻造技术的进步使得在商用喷气发动机中批量生产金属间钛铝化物（TiAl）制成的涡轮叶片成为可能。然而，目前，由于对最终几何形状的高要求以及对铸造结构高度变形的要求，这种技术存在许多局限性。以目前的工艺和工装技术，涡轮叶片的材料利用率往往不到10%。因此，在第一阶段，研究了具有局部成形特性的等温扁刀锻造过程，并推导了合适的控制器。在第二阶段，我们专注于利用工件的局部加热在几个冲程中控制扁平工具锻造的性能。提出并分析了利用形变几何的Level-Set描述的合适控制策略。在第三阶段，目标几何形状的联合和鲁棒控制以及多冲程控制下的变形程度将是我们研究的重点。由M. Bambach领导的提案部分侧重于使用基于神经网络的替代模型预测材料的微观结构。这些传感器可以在接近生产的条件下用作实时软传感器。计划实现目标几何形状和变形程度的联合预测，以便为最终生产工艺步骤生产刀具。由M. Herty领导的提案部分研究了基于训练神经网络给出的可验证代理软传感器的鲁棒控制器的数学方面。利用贝叶斯定理设计具有误差保证的新型控制的公式将得到发展。成形、几何、微结构控制和数学控制理论等方面的研究问题在各个方面紧密联系在一起。

In isothermal forging, the tools are heated to the workpiece temperature, so that near-net-shaping becomes feasible for materials of low workability. Advances in isothermal forging have recently made it possible to use turbine blades made of intermetallic titanium aluminides (TiAl) in serial production in commercial jet engines. Currently, however, many limitations of this technology can be found due to the high demands on final geometry as well as the demand on the high degree of deformation of the cast structure. With the current process and tooling technology, the material utilization of turbine blades is often less than 10%. In phase 1, a property-controlled isothermal flat tool forging process with local forming was therefore investigated and a suitable controller has been derived. In the 2nd phase we focused on property-controlled flat tool forging in several strokes using local heating of the workpiece. Suitable control strategy using a Level-Set description of the deformed geometry has been proposed and analysed. In the third phase the joint and robust control of target geometry as well as deformation degree under controls of multiple strokes will be our focus. The part of the proposal led by M. Bambach focusses on the prediction of the material micro-structure using surrogate models based on neural networks. Those allow to be used as real-time soft-sensors under near-production conditions. It is planned to achieve the joint prediction of target geometry and deformation degrees, such that tools can be produce for the final production process step. The part of the proposal lead by M. Herty investigates the mathematical aspects of robust controllers based on verifiable surrogate soft-sensors given by the trained neural networks. A formulation using Bayes’s theorem for the design of novel controls with error guarantees will be developed. The research questions in forming, geometry, microstructure control and mathematical control theory are closely linked in various aspects.