Data-driven process modeling in stamping technology

在数字化和小型化的背景下，对复杂电子元件的需求越来越大。它们通常是在冲压和弯曲过程中产生的，其特点是许多成形操作和相关的相互作用。复杂的相互作用不能完全用模拟来表示，因此隐含知识仍然是必不可少的。因此，开发和使用新的模型是必要的，以便能够产生更多的过程理解。特别是，数据驱动建模在这方面提供了很有前途的潜力，可以从现有数据中提取额外的知识，从而形式化隐含的知识。因此，在本项目中，将基于不同的数值和真实数据来模拟冲压弯曲过程。这里特别感兴趣的是定义的质量标准，它可以是几何性质的，也可以是物理性质的。通过有针对性地引入和测量扰动变量，使全面了解该过程成为可能。为此，开发了允许影响该过程的模块。所提出的建模方法的一个特点是使用了基于集合的可达性分析。也就是说，问题的表述方式是保证可达状态在某个集合内。对于成形工艺，这意味着保证，例如，符合某些公差。同时，可实现状态集合的大小和分布允许对过程稳定性进行评估。与随机方法相比的另一个优势是可以确定清晰的因果链。因此，这种方法有望为增加对过程的理解提供新的可能性。然而，这需要开发新的算法来解决这个问题。在建立模型、检查一致性和正式验证之后，可以进行模型的验证。为此，在定义的工艺参数的基础上再次制造部件，并与模型的预测进行比较。其次是基于该模型的流程优化。使用优化工艺的后续生产最终允许在工艺之间进行比较，并对开发的模型的预测能力进行全面评估。

In the context of digitalization and miniaturization, the demand for complex electronic components is increasing. These are often produced in stamping and bending processes, which are characterized by many forming operations and associated interactions. The complex interactions cannot be fully represented by simulations, so implicit knowledge remains essential. Therefore, the development and use of novel models is necessary to be able to generate additional process understanding. Especially data-driven modeling offers promising potentials in this respect to extract additional knowledge from existing data and thus to formalize implicit knowledge. Therefore, in this project a stamping-bending process will be modeled based on different numerical and real data. Of particular interest here are the defined quality criteria, which can be of both geometric and physical nature. Through the targeted introduction and measurement of disturbance variables, a comprehensive view of the process is made possible. For this purpose, modules are developed that allow to influence the process. A special feature of the proposed modeling approach is the use of a set-based reachability analysis. That is, the problem is formulated such that reachable states are guaranteed to be within a certain set. For forming processes, this means guarantees, e.g., regarding the compliance with certain tolerances. At the same time, the size of the set and distribution of achievable states allows an evaluation of the process stability. Another advantage in contrast to stochastic methods is that clear cause-effect chains can be identified. Thus, this approach promises new possibilities to increase the process understanding. However, this requires the development of new algorithms that allow the problem to be solved. After the model has been built, checked for conformance and formally verified, the validation of the model can take place. For this purpose, components are manufactured again on the basis of defined process parameters and compared with the predictions of the model. This is followed by process optimization based on the model. Subsequent production using the optimized process finally allows a comparison between the processes and a comprehensive evaluation of the predictive capability of the developed model.