This paper proposes an architecture for a simulation-based Driver Monitoring System (DMS) evaluation framework using a virtual driver model. Traditional DMS evaluation methods require the recruitment of human participants, which leads to high costs and extended development cycles. By replacing human drivers with virtual drivers generated from real-world driver data and augmented through AI, the proposed system enables repeated performance evaluations of DMS in a virtual environment. This approach allows DMS performance to be tested and optimized throughout the design process, significantly reducing the need for human participants and shortening the development time. The proposed system architecture comprises three main components: a virtual driver-based simulation software, DMS performance evaluation software, and the target DMS itself. The simulation software includes virtual drivers with cognitive, behavioral, and visual models, as well as a virtual in-cabin environment and camera sensor model. The performance evaluation software integrates evaluation protocols based on global standards such as Euro NCAP and UNECE, providing detailed feedback to improve the DMS design.