Transfer Learning Approaches In The Domain Of Radial-Axial Ring Rolling For Machine Learning Applications

Aug 3, 2026

Upload by Skyler

Growing data availability is driving data-centric machine learning in the forming industry to increase resource efficiency and optimize processes. Prior work shows that machine learning can classify the roundness of rings produced by radial-axial ring rolling, yet accuracy is constrained by limited and imperfect data. This paper evaluates transfer learning strategies to leverage industrial logs and smaller datasets, and further studies improvements via an additional synthetic dataset generated by GANs.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages12

Reading Time30 min

AccessFree to download

1. Introduction

2. Theoretical background and related work

2.1 Theoretical background

2.1.1 Radial-axial ring rolling

2.1.2 Machine learning for time series predictions

CONFERENCE ON PRODUCTION SYSTEMS AND LOGISTICS CPSL 2023-2 5th Conference on Production Systems and Logistics Transfer Learning Approaches In The Domain Of Radial-Axial Ring Rolling For Machine Learning Applications Johannes Seitz 1 , Tobias Moser 1 , Simon Fahle2 , Christopher Prinz 1 , Bernd Kuhlenkötter1 1Chair of Production Systems, Ruhr-University Bochum, Bochum, Germany 2 Westnetz GmbH, Dortmund, Germany Abstract Due to increased data accessibility, data-centric approaches, such as machine learning, are getting more represented in the forming industry to improve resource efficiency and to optimise processes. Prior research shows, that a classification of the roundness of shaped rings, using machine learning algorithms, is applicable to radial-axial ring rolling. The accuracy of these predictions nowadays is still limited by the amount and quality of the data. Therefore, this paper will focus on how to make the best use of the limited amount of data, using transfer learning approaches. Since acquiring data for homogenised databases is time, energy and resource consuming, logged data gathered by the industry is often used in research. This paper takes both, industrial data from thyssenkrupp rothe erde Germany GmbH and a smaller dataset of an inhouse research plant, into account. Additionally, a synthetic dataset, created by generative adversarial networks, is considered. To accomplish an improvement of machine learning predictions using accessible data, three transfer learning approaches are investigated in order to extend existing models: (I) transferring from a radial-axial ring rolling mill to a different mill containing less available data with a ratio of 20:1,(II) learning from unlabelled data using an autoencoder and (III) training on synthetic data. The obtained improvements are further evaluated. Based on these results, future possible investigations are elaborated, in particular the consideration of transfer learning from the less complex cold ring rolling process. Keywords Radial-Axial Ring Rolling; Machine Learning; LSTM; Predictive Quality; Transfer Learning 1. Introduction To cope with environmental and economical requirements, techniques to increase resource efficiency in the forming industry, are applied, such as material and process efficiency [1] . Especially the reduction of parts which either need to be reworked or are scrap, is of interest. In the hot forming process radial-axial ring rolling (RARR), there are still challenges to cope with this condition, since the establishment of astable RARR process according to GUO ET YANG is difficult and depends on many influencing factors as for instance the ring growth behaviour [2] . Hence, several approaches have been conducted to obtain these improvements, including the use of machine learning (ML) algorithms. In RARR FAHLE ET AL. proved the applicability of ML in order to predict quality, through a classification task of the roundness of formed rings after each ring roll procedure [3] . Since ML algorithms are sensitive to the data quality and data amount, predictions are still limited. The conducted survey by FAHLE ET AL. demonstrates this issue: although all considered producing companies are storing data, just 50 % of these companies are analysing and using this data [4] . Additionally, in terms of using ML for quality predictions, it is necessary to DOI: [https://doi.org/10.15488/15](https://doi.org/10.15488/15)􀀖0􀀖 ISSN: 2701-6277 447 connect process data with its target data. In terms of predictive quality in RARR, the target data is the corresponding measured quality parameter (label) . To generate data for scientific applications, this procedure is consuming time and process energy. This is especially the case, if a dataset for ML applications is needed, since the performance is depending on the data amount. Based on the difficulties concerning data gathering and availability, this paper contributes three approaches t

FAQs

Why is transfer learning needed for quality prediction in radial-axial ring rolling?

Machine learning performance is limited by the amount and quality of available labeled data. Transfer learning enables improving prediction models using accessible, limited datasets.

What data sources are considered to build and evaluate the models?

The study uses industrial logged data, a smaller in-house research dataset, and also a synthetic dataset created by generative adversarial networks (GANs).

What three transfer learning approaches are investigated?

The paper studies: (I) transferring from one radial-axial ring rolling mill to another mill with less data, (II) learning from unlabelled data via an autoencoder, and (III) training using synthetic data.

Aug 3, 2026

Upload by Skyler