Session: 14-06-01: Machine Learning and Deep Learning in Safety, Reliability, and Maintenance I

Paper Number: 145271

145271 - Data-Driven Functional Modeling of Corroded Bolted Joints: A Framework for Remanufacturing 

The circular economy emphasizes resource efficiency and sustainability through key aspects such as remanufacturing. Remanufacturing is the process of restoring used products to a like-new condition, extending their lifecycle. Ensuring the quality and reliability of remanufactured products depends on accurately predicting function-oriented variables of the remanufactured product. Bolts are the most commonly used machine element, and their remanufacturing can have a significant impact on the circular economy. However, remanufacturing bolted joints is a challenging task due to the uncertainties that arise from prior product usage. Corrosion often affects bolted joints, which complicates the tightening process and makes it hard to predict the function-oriented variables of refurbished bolted joints. Various methodologies are currently employed to address this challenge, including analytical models, numerical models like Finite Element Analyses (FEA), and empirical models. Each of these approaches comes with a distinct set of strengths and weaknesses. Analytical models offer fast computation but often yield inaccurate results due to the inherent nonlinearities of bolted joint systems. In contrast, FEAs are precise and capable of capturing these nonlinearities and providing detailed results, but they require substantial computational resources. The research gap is that the current models, namely analytical models and FEA, frequently fail to meet accuracy criteria. Empirical methods, including statistical design of experiments (DoE) and machine learning, provide an alternative approach by identifying functional relationships independently of system complexity.

Addressing this research gap, this paper presents a data-driven framework that aims to overcome the challenges associated with estimating function-oriented variables in corroded bolted joints during the remanufacturing process. Our approach utilizes the benefits of coupling empirical methods with artificial intelligence techniques to develop a predictive functional model. This model aims to accurately predict key variables such as load capacity and thread friction coefficient for the remanufactured bolted joints upon tightening and untightening. Additionally, it establishes a methodology for function-oriented evaluation.

This research methodology focuses on enhancing predictive models within the context of remanufacturing. To address existing limitations, we develop a functional predictive framework. Within this framework, we identify Influencing Variables (IVs), Descriptive Variables (DVs), and Target Variables (TVs). IVs are variables that affect sample conditions and can control the experiment, but are not recorded during the experiment. DVs involve all variables recorded during the experiment, which can be either qualitative or quantitative. TVs are the variables we aim to predict or control in the process, such as the load capacity and the friction coefficient of the newly bolted joint. We apply empirical models to collect and determine the relevant variables and their factor levels. Following this, we incorporate machine learning algorithms to boost predictive precision.

The proposed framework is structured around three main components. The first is the development of a custom test rig, designed to accurately evaluate and quantify the variables impacting corroded bolted joints, specifically the IVs and DVs. The second component involves identifying the key variables; IVs and DVs that affect the behavior of these joints, with a particular focus on the degree of corrosion. Using a Design of Experiments (DoE) approach, we systematically define the levels of these factors to generate a comprehensive test plan that captures the entire range of variables that are encountered during the experimental phase. The aim is to comprehend how these IVs influence the DVs and TVs, and consequently, the overall process behavior. The final element of the framework is the development of a predictive model, which is based on empirical data from various DoE tests and is refined using machine learning techniques to ensure its predictive accuracy and robustness.

By embracing a data-driven approach, our research seeks to fill the gaps left by existing modelling techniques, thereby enhancing the reliability of modelling techniques for corroded bolted joints. While it’s currently theoretical and may encounter AI-related and experimental challenges, this framework contributes to the circular economy, product quality, reliability, and sustainability. Our framework can eventually be applied beyond bolted joints, addressing uncertainty across various products.

Presenting Author: Nehal Afifi IPEK – Institute of Product Engineering at the Karlsruhe Institute of Technology (KIT)

Presenting Author Biography: The author completed her undergraduate and postgraduate studies at the German University in Cairo, including a Master's thesis focusing on precision assembly using sensitive robotics and deep reinforcement learning, which was completed in Germany at the Zentrum für Mechatronik und Automatisierungstechnik gemeinnützige (ZeMA). She is currently a research associate and PhD student at IPEK Institute for Product Engineering at Karlsruhe Institute of Technology (KIT). Her research interests lie in the interaction between artificial intelligence (AI) and product engineering. Currently, she's involved in several research projects investigating AI utilization in predictive maintenance and remanufacturing, and its impact on circular economy.

Authors:

Nehal Afifi IPEK – Institute of Product Engineering at the Karlsruhe Institute of Technology (KIT)
Jan-Philipp Kaiser Institute of Production Science (wbk) at Karlsruhe Institute of Technology (KIT)
Andreas Wettstein IPEK – Institute of Product Engineering at the Karlsruhe Institute of Technology (KIT)
Gisela Lanza Institute of Production Science (wbk) at Karlsruhe Institute of Technology (KIT)
Sven Matthiesen IPEK – Institute of Product Engineering at the Karlsruhe Institute of Technology (KIT)