Automated Design of Experiments supporting Feature-based Optimisation of Manufacturing Processes

Abstract

This paper presents a novel process design to enhance time and cost-efficient AI training in manufacturing. As an alternative to time and resource expensive trial-and-error loops, the data basis of the proposed design enables data-driven parameter selection, where the parameter range in which the optimal solution is likely to reside, can be explored in a reproducible and systematical manner. A full-factorial DOE is generated and implemented from within the CAM software. Necessary production artefacts, like NC code, bill-of-material or work plan, are supplied per experiment. The heterogeneous data of different product life cycle phases are collected and related to the according manufacturing feature (i.e., drilling, face-milling, etc.). From within the CAM software, the NC-Code is manipulated to enable the identification of features during production, using feature markers. Instantiated as Siemens NX CAM extension, the novel design was tested on a 5-axis milling and drilling process on aluminium parts. The automated data set generation with feature correlation between different live-cycle phases was verified. As a result, the design supports feature optimization strategies for decision support systems - either as input for CAD/CAM, PLM, ERP and MES.