The study aims to address the issues of low efficiency, high cost, and safety hazards associated with manual inspection of building facilities. To address these issues, the study proposes a virtual inspection system based on robotic process automation (RPA) technology. First, a dynamic digital twin building model and an update algorithm are constructed based on BIM data, forming a real-time mapping relationship with the on-site facility operation status. Next, a multi-modal deep learning-based device defect detection network is designed to achieve automatic high-precision defect detection. Finally, the heterogeneous system integration and the application of intelligent scheduling algorithms are realized based on the software robotics technology architecture, and an integrated software robotics inspection system is developed. Preliminary verification was carried out in experimental scenarios. The results show that the virtual inspection robot system based on the software robotics technology architecture has: the facility defect detection accuracy improved to over 97%, surpassing single-modal detection methods; the operation efficiency is increased by an average of more than 62.3% compared to traditional methods; the maintenance manpower demand is reduced to 60%-80%, exemplifying its substantial engineering application value.