Enhanced Predictive Maintenance for Shipbuilding Hull Block Assembly Using Multi-Modal Data Fusion and Reinforcement Learning

**Abstract:** Shipbuilding digitalization necessitates predictive maintenance strategies to minimize downtime and maximize operational efficiency. This research proposes a novel system leveraging multi-modal data fusion and reinforcement learning (RL) to predict and prevent failures in hull block assembly processes. By integrating data streams from Computer Vision (CV) for defect detection, acoustic sensors for anomaly analysis, and Programmable Logic Controller (PLC) data for process monitoring, the system constructs a comprehensive operational profile. A deep reinforcement learning agent, trained on historical data and simulated operational scenarios, dynamically optimizes maintenance schedules, leading to a projected 35% reduction in unplanned downtime and a 15% improvement in overall equipment effectiveness (OEE). The commercial viability lies in decreased maintenance costs, improved product quality through proactive intervention, and enhanced utilization of existing shipbuilding resources.