Project Details
Projekt Print View

Derivation of cause-effect relationships for die design on the basis of data-driven process modeling for fine blanking

Subject Area Primary Shaping and Reshaping Technology, Additive Manufacturing
Term since 2023
Project identifier Deutsche Forschungsgemeinschaft (DFG) - Project number 520460745
 
The highly productive precision shearing process fineblanking enables the economical production of sheared parts with increased requirements on the quality of the shearing surface with regard to the contact area and the squareness of the surface, such as for dimensionally and formally accurate tooth flanks. Fineblanked parts are used in products from medical technology to electromobility. An optimal sheared part design as well as coordination of the tool parameters depends on the application and is usually based on a combination of empirical as well as formalised process knowledge (e.g. in the form of FE simulations, reference value tables). Despite this process knowledge, deviations between the expected and real sheared part quality can be observed. This so-called “process noise” cannot be modelled economically with calculation-intensive methods such as FE simulations. A systematic differentiation within the summary of different phenomena as process noise is necessary in order to make it explainable and usable by using suitable data-based, forming representations and learning methods. Influencing variables that cannot be explained or cannot be explained adequately include (i) unknown physical effects; (ii) physical effects whose existence is known but which are nevertheless difficult to model; (iii) causal relationships along the process chain; (iv) disturbance variables that can be determined in principle; (v) residual stochasticity that may not be explainable.The overall objective of the application is to make the process noise contained in the process signals partially explainable and predictable. To this end, the domains of empirical and formalised process knowledge, physics-oriented AI and data-centred AI are interlinked in this project in the form of a so-called "Data-to-Knowledge" pipeline. With FE simulations, existing process signals and those to be collected experimentally, as well as shearing surface characteristics, interdependencies of the variables die clearance, chamfer geometry and shearing surface quality, which are to be understood as active surfaces, are mapped digitally. With these digital images, supervised learning models are trained and interpreted and examined using Explainable AI (XAI) techniques. In addition, complex correlations and patterns contained in the data are captured by unsupervised learning and condensed into efficient representations (Representation Learning). These representations and the knowledge extracted by means of XAI about the cause-effect relationships of the system parameters are examined by means of causal tests (Causal Inference) and hypotheses with statistical guarantees are derived from them. These hypotheses are examined for their plausibility using empirical and formalised knowledge. The findings from the data-to-knowledge pipeline form the basis for an optimised design of the active surfaces in the second project phase.
DFG Programme Priority Programmes
 
 

Additional Information

Textvergrößerung und Kontrastanpassung