Abstrahierte Modellierung des nichtlinearen Verhaltens von Fügestellen in Baugruppen aus faserverstärkten Verbundmaterialien

Abstract

Efficient analysis and design of bolted joints in structural assemblies are crucial for ensuring the safety and performance of many large-scale systems. This area has attracted significant attention within the engineering research community for decades. While extensive literature exists on theoretical and numerical approaches for evaluating the mechanical behavior of bolted joints, many of these studies are limited to simplified scenarios. They often fail to capture the full nonlinear mechanical response under large strains or to accurately predict failure phenomena. Given this, it is essential to develop robust models capable of accurately predicting the overall behavior of bolted joints under realistic conditions. A fully nonlinear finite element (FE) model that effectively captures the intricate interactions and damage processes within bolted joints could serve as a viable solution. However, the high computational cost associated with such detailed FE simulations remains a significant challenge. A promising direction involves leveraging recent advancements in machine learning (ML) techniques. By combining ML with robust nonlinear FE analysis, it is possible to develop computationally efficient strategies for simulating the nonlinear behavior of bolted joints. In this context, the present work introduces a novel modeling approach aimed at efficiently simulating the nonlinear behavior of bolted joints. The approach incorporates an elasto-plastic material formulation capable of handling large deformations and is implemented within a commercial FE software using a user-defined element. To further reduce computational expense, a submodeling strategy is adopted, wherein submodels of individual wedges in the bolted region are created and assembled to represent the entire joint region. As a proof of concept, the framework is tested on bolted joints with metallic substrates. Finally, a feed-forward neural network (FFN) algorithm is embedded within the user-defined element, which significantly reduces computation time while maintaining accuracy. The proposed methodology aims to enable faster and more reliable predictive tools for the safe and efficient design of complex engineering structures.