Zur Seitenansicht

Titelaufnahme

Links
Download
Klassifikation
Abstract
  • Demographic change and the resulting shortage of skilled labor increase the need for automation in industrial production systems. Since decisions influencing automation suitability are fixed mainly in early product development phases, reliable assessment methods are required even when product and process information is incomplete. Established Design for Automation and Automated Assembly (DFA/DFAA) methods provide valuable expert knowledge but are often too information-intensive for early-stage applications. This paper presents a hybrid intelligence approach that combines expert-based DFA/DFAA evaluation methods with machine learning–based surrogate models to support early assessment of automation suitability. First, a systematic literature review is conducted to identify key product-related criteria influencing automated assembly. Based on these criteria, a synthetic dataset of standardized product descriptions is generated and labeled using a validated DFA evaluation algorithm derived from established methods. Subsequently, a graph isomorphism network is trained to predict the Automation Suitability Degree directly from graph-based product representations. Validation results demonstrate high predictive accuracy and strong correlation with expert-based evaluations, indicating that the model reliably captures relative suitability trends even for partially specified products. While absolute errors increase for previously unseen product variants, the approach shows strong potential as a scalable decision-support tool.
Statistik
  • Das PDF-Dokument wurde 8 mal heruntergeladen.
Lizenz-/Rechtehinweis