Effects of cross-factory horizontal and cross-stage vertical data integration on predictive readiness for data-driven manufacturing optimization

In the domain of manufacturing, the pursuit of process optimization has undergone a notable transition towards a data-driven approach. However, it must be noted that the quantity of available data is often inadequate for generating precise predictions regarding the outcomes of manufacturing processes. Consequently, this paper presents an empirical investigation of how horizontal and vertical data integration, as part of a proposed conceptual optimization workflow, influence the predictive readiness of multi-stage regression models intended for downstream surrogate-based optimization in manufacturing. While data integration has been widely discussed as a means to overcome limited data availability, quantitative evidence on its effects within multi-stage regression settings remains scarce. To address this gap, a case study based on a publicly available multi-stage continuous-manufacturing dataset is conducted to quantify the impact of varying degrees of horizontal integration across similar machines and vertical integration across consecutive process stages. The results show that (i) horizontal data integration can improve model robustness to data drift by expanding the attribute value range, (ii) vertical integration of cross-stage process data has the potential to enhance prediction accuracy compared to siloed modeling, and (iii) more general models trained on heterogeneous environments do not necessarily yield better performance for environment-specific prediction tasks. Full physical validation of the proposed optimization workflow and real-world deployment are outside scope due to dataset limitations. The findings provide practical and scientific insights on when and how data integration supports data-driven manufacturing optimization.