Heavy industry, characterized by the production of large and complex products, relies on fixed-position assembly methods that constrain the flexible adjustment of work processes. Hence, unexpected delays, resource shortages, and abnormal situations cannot be promptly addressed, leading to frequent discrepancies between on-site conditions and production plans, ultimately hindering on-time delivery. However, most resource allocation and schedule management still rely on the experience and judgment of on-site workers, limiting the ability to respond promptly to unforeseen issues. This study aims to utilize digital twin (DT) technology to accurately recognize issues in manufacturing environments and support system-driven decision-making based on real-time data, thereby improving production schedule adherence and delivery reliability. To this end, we propose a framework that integrates AI-based computer vision analysis with simulation-based verification techniques, enabling real-time monitoring and prediction of dynamic changes in the production environment. The framework enhances production flexibility, improves adherence to delivery schedules, and minimizes operational downtime, offering a scalable solution for improving production efficiency and responsiveness in complex manufacturing settings.