Predictive Machine Vision and AI-Driven Dynamic Line Balancing for Intelligent Pharmaceutical Fill–Finish Operations

Abstract Aseptic fill–finish lines operate under stringent good manufacturing practice (GMP) constraints and high throughput demands, where small disturbances can cascade into bottlenecks that erode overall equipment effectiveness (OEE). Conventional static balancing and manual adjustments often respond too late to prevent efficiency losses. This study introduces a predictive, machine vision–driven line balancing framework that integrates real-time visual analytics with plant control data to enable intelligent decision support in pharmaceutical manufacturing. The system fuses overhead video with PLC/MES event logs to generate synchronized queue and cycle-time features at 1 Hz. These signals feed a bottleneck detection and predictive classification model that anticipates emerging flow imbalances using interpretable indicators such as queue slope, cycle-time deviation, and variance trends. A constrained decision policy then recommends GMP-bounded interventions conveyor trims, pacing adjustments, or operator prompts under human-in-the-loop supervision. Evaluated over one year of operational data (732 shifts), the framework achieved a 14.5% increase in throughput, a 39% reduction in bottleneck duration, and an OEE proxy improvement from 0.41 to 0.61. The results demonstrate that intelligent, data-driven control using vision and predictive analytics can enhance resilience and efficiency without violating regulatory guardrails. The study advances the concept of interpretable AI for intelligent manufacturing, bridging predictive modeling with compliance-aware decision support in aseptic production environments.