The primary applications of wire EDM are found in tool and die manufacturing, as well as in engine and medical technology. It is predominantly used for producing high-value components and often serves as the final critical manufacturing step. Consequently, process reliability and repeatability are of utmost importance and can be ensured through intelligent control and automation solutions. In this context, the growing digitalization of manufacturing processes, driven by Industry 4.0, highlights the need for data-driven approaches in wire EDM.
The aim of this work was to develop a data-driven model for evaluating the wire EDM process, primarily based on continuously recorded electrical process data to ensure the model’s transferability and general validity. Machine learning models were trained on this process data to enable real-time evaluation of the process based solely on electrical signals. The objective was to achieve this by developing a regression model to assess quality and a classification model to evaluate productivity. The scientific framework of this study is shaped by the data analysis methods and techniques employed, and the structure of the work is accordingly aligned with the development of data-driven models.
A system was first developed to enable the real-time recording of temporally and spatially resolved individual discharges within the continuous wire EDM process. Following systematic data processing, including data reduction and feature extraction, characteristic values were subsequently correlated with process productivity and product quality. Building on these initial process data insights, a regression model was created to predict product quality. For this purpose, a neural network was trained to estimate the component's curvature based on continuously recorded data, achieving high prediction accuracy and explaining a significant portion of the data variability. Process productivity was assessed through a classification model using a deep learning approach, where various neural network architectures were explored. The results demonstrated high accuracy, particularly noteworthy given that all evaluations were conducted using entirely unseen data. The findings were applied to develop a Digital Twin in an industrial context, capable of visualizing the real-time curvature of the workpiece on a dashboard using continuously processed data.
Autor | Küpper, Ugur |
---|---|
Gewicht | 0.3 kg |
Erscheinungsdatum | 27.01.2025 |
Eigene Bewertung schreiben
Technologie der Fertigungsverfahren
Data-driven Model for Process Evaluation in Wire EDM
ISBN: 978-3-98555-260-3
Lieferzeit: 2-3 Tage
54,00 €
inkl. 7% MwSt.
Kurzbeschreibung
Wire EDM is widely used for producing high-value components, requiring high process reliability and repeatability. This work focuses on developing data-driven models to evaluate wire EDM processes using real-time electrical signals. A regression model predicts product quality, while a classification model assesses productivity. The approach shows high accuracy and was applied to create a Digital Twin that visualizes workpiece curvature in real-time through an industrial dashboard.
Auf Lager