AI-Driven Bottleneck Detection for Labor-Intensive SME Manufacturing

Bottleneck detection and dynamic scheduling remain critical challenges in labor-intensive manufacturing systems, particularly for small and medium-sized enterprises, where manual processes and unpredictable delays frequently disrupt production flow. This paper proposes a lightweight, AI-driven framework tailored for SMEs for real-time delay prediction and task reallocation, built on the Configurable Virtual Workstation System architecture. The system integrates physical workstations with RFID-based logging, edge processing using microcontrollers, and centralized coordination logic. Two supervised models—Random Forest and Logistic Regression—were trained using 500 task records from a 3-station prototype simulating cutting, assembly, and inspection operations to predict potential delays. Logistic Regression achieved a recall of 0.69 and an F1-score of 0.56 for detecting delays, outperforming Random Forest, with F1-score of 0.36. Task reallocation based on real-time prediction helped reduce queue build-up and mitigate emerging bottlenecks. While promising, the system’s limitations include reliance on synthetic data and testing within a controlled laboratory setup. This architecture offers a low-cost and scalable solution for bottleneck detection, that is simple and deployable on edge devices.