Development of a data-driven model for the evaluation and improvement of process robustness in the design of deep-drawing tools

开发数据驱动模型，用于评估和提高拉深工具设计中的工艺稳健性

• 批准号: 520204466
• 负责人: Professor Dr.-Ing. Noomane Ben Khalifa
• 金额: --
• 依托单位: Institut für Produktionstechnik und -systeme (IPTS)
• 依托单位国家: 德国
• 项目类别: Priority Programmes
• 资助国家: 德国
• 项目状态: 未结题
• 来源: https://gepris.dfg.de/gepris/projekt/520204466?language=en
• 关键词: Development; data; driven; model; evaluation

In industrial deep drawing processes, stochastic fluctuations and disturbances of the manufacturing conditions occur, which can cause uncontrolled deterioration of the product properties. The immunity to these negative influences is referred to as robustness. Robustness in deep drawing can be assessed by sensors integrated into the press line. This generates extensive amounts of data that have potential to be used for machine learning modelling and for analysing complex interactions. The field of explainable AI, which serves to explain such data-driven models is becoming increasingly relevant. As such, the aim of the research project is to describe the effects of stochastic fluctuations and disturbances on product quality in an explainable way using data-driven models. The scientific approach is based on the fact that the flange length of the first forming stage can be used as a significant quality criterion. As a metrological solution, a camera system will be used for non-contact measurement of the flange length. The research project is divided into two stages. The first stage is concerned with developing the modelling approach based on a cross die geometry. For this purpose, training data will be generated by experimental and numerical investigations. In the second stage, this modelling approach will be applied to industrial geometries using process data from series production. At the end of the research project, a unified explanatory model will be derived. It is hypothesized that the transformation of absolute process values into relative data will favour comparability between different geometries. The purpose of this model is to determine the process robustness in the design of deep drawing tools.

在工业深冲过程中，制造条件会发生随机波动和干扰，这可能会导致产品性能不受控制地恶化。对这些负面影响的免疫力被称为鲁棒性。通过集成到冲压生产线中的传感器，可以评估深冲过程中的稳定性。这产生了大量的数据，这些数据有可能用于机器学习建模和分析复杂的交互。可解释的人工智能领域，用于解释这种数据驱动的模型，正变得越来越重要。因此，该研究项目的目的是使用数据驱动模型以可解释的方式描述随机波动和干扰对产品质量的影响。科学的方法是基于这样一个事实，即第一个成型阶段的法兰长度可以作为一个重要的质量标准。作为一种非接触式解决方案，将使用相机系统对法兰长度进行非接触式测量。研究项目分为两个阶段。第一阶段是关于发展的建模方法的基础上，一个跨模具的几何形状。为此，将通过实验和数值研究生成训练数据。在第二阶段，这种建模方法将应用于工业几何形状，使用来自系列生产的过程数据。在研究项目结束时，将得出一个统一的解释模型。据推测，将绝对过程值转换为相对数据将有利于不同几何形状之间的可比性。该模型的目的是确定工艺的鲁棒性在设计的拉深工具。