The concept of digital twin represents a cutting-edge approach that seamlessly integrates virtual and real-world environments, effectively addressing complexity and uncertainty issues encountered throughout the lifecycle of agricultural equipment. This innovation is poised to accelerate the transformation and modernization of the agricultural mechanization and equipment industry. However, the practical application of digital twin technology for agricultural machinery is still in its nascent stages, and typical case studies and practical solutions are yet to be developed. In light of the unique characteristics of digital twin and agricultural machinery, a cloud-fog-edge-terminal collaborative digital twin system architecture and operation mechanism was proposed, integrating the 5D model and mobile edge computing technology. Specially, a digital twin prototype system for a large corn harvester with grain direct harvesting capabilities was developed, focusing on the high broken grain rate during the threshing process. This system enabled functions such as model prediction, model update, real-time monitoring, and optimization decision-making. Field experiments were conducted, the results showed that the digital twin system effectively enhanced the adaptability of the virtual model, maintaining good predictive performance. Furthermore, the decision optimization method based on digital twin can reduce the broken grain rate by an average of 24.24% compared with manual harvesting mode, and by an average of 15.78% compared with feedback control mode. These findings confirmed that the prototype system can effectively improve the quality of corn grain harvesting. Overall, the proposed system architecture and implementation method were feasible and can provide a reference for further research and application of digital twin in the agricultural machinery industry.