Answers are grounded in this paper and cite their sources.
Evaluates automated cycle-time measurement in mixed-model assembly, highlighting inefficiencies and variability.
This study presents a non-intrusive, AI-enabled video analytics system to automate cycle-time measurement in mixed-model manual assembly lines, where traditional manual time studies remain labour-intensive, error-prone, and difficult to scale across diverse product variants. The proposed method integrates object detection and segmentation models to identify tools, parts, and product size directly from real-time video streams, enabling the system to classify operator actions and determine process time. To ensure reliable performance in high-variability environments, an AI-enabled multi-stage processing pipeline is introduced that converts frame-level detections into a binary object-presence matrix, applies Lean-based action logic to distinguish Value-Added Time (VAT) from Non-Value-Added Time (NVAT), and employs a modified clustering strategy to filter noise, manage operator-driven workflow deviations, and segment complete process cycles. The system was deployed and evaluated in a high-mix, low-volume industrial environment where the proposed system achieved an average product-count error of 5%. Moreover, the system generated detailed VAT/NVAT breakdowns and cycle-time distributions across different product sizes. The findings demonstrate how product complexity, operator behavior, and workflow interruptions influence cycle-time variability and reveal hidden inefficiencies that are difficult to detect using manual observation.
No takes yet. Share an insight, caveat, or question.
Sanchez et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: