Automated OCEL transformation for real-time conformance in complex manufacturing

Abstract Object-Centric Process Mining (OCPM) provides a sophisticated framework for analyzing multi-object interactions in complex manufacturing; however, its practical adoption is limited by the labor-intensive transformation of heterogeneous data into Object-Centric Event Logs (OCELs). This research addresses the bottleneck through an automated transformation pipeline that converts raw industrial event streams into conformance-ready OCELs. Unlike traditional static mapping, the architecture integrates a multi-criteria decision-making (MCDM) layer using CRITIC and PROMETHEE II to systematically select classifiers and process discovery algorithms under competing performance criteria. To minimize manual overhead, an uncertainty-aware active learning mechanism invokes expert validation only when entropy thresholds are exceeded, reducing human intervention to just 1.2% of events. Validation was conducted on a primary timber production dataset (54,976 events) and reinforced through cross-sector testing on metallic tube manufacturing and high-density industrial sensor streams. The framework achieved a 97.5% role-assignment accuracy and maintained an end-to-end latency of 1.27 ms, making it suitable for real-time deployment. By integrating object-centric process features with operational indicators, the pipeline improved real-time conformance prediction accuracy from 75.6% to 92.4% ( $$AUC = 0.934$$ ). These findings establish a robust, scalable foundation for automated compliance monitoring in high-performance production environments.