As rail transit systems advance toward greater automation and intelligence, cloud computing technology, with its remarkable scalability and robust data processing capabilities, has been steadily expanding its footprint in this domain. However, the adoption of cloud computing also introduces new safety challenges for train control systems. Traditional safety computers in train control systems rely on heterogeneous redundancy with symmetry to enhance safety. Nevertheless, the software in cloud computing environments, even if heterogeneous, may share the same source code, thereby triggering the risk of common-cause failures in the software. To address these issues, this study proposes a multi-level monitoring architecture system tailored to the characteristics of cloud-based safety computing platforms. This architecture innovatively integrates the three-level monitoring architecture pattern from the automotive field, the secure channel pattern, and the distributed safety mechanism architecture. It monitors the levels of common-cause software failures that cannot be eliminated through heterogeneity. The introduction of multi-level active monitoring for risk control has reduced the impact of common-cause software failures on system security. By constructing a formal security model, quantitative evaluations are conducted separately on the single-channel L2 and L3, the dual-channel L4 without degradation or monitoring, and the dual-channel L4 monitoring architecture with complete functions. This verifies the effectiveness of the proposed monitoring architecture in reducing the risk of common-cause software failures in the virtualization layer. This study provides a robust theoretical foundation and technical support for the security-oriented design and development of the next-generation intelligent rail transit systems.

As a safety critical system, train control system matters in passengers life and properties. It is important that how to guarantee the safety of train control system. Safety testing is an effective method to detect the safety holes and bugs in the system. However, most safety testing for train control system is manually executed based on expertise, which leads to a huge testing workload. Besides, manual generation will easily cause the problem of missing test cases. In this paper, a model-based safety test method is introduced, hoping to solve this problem. We select a core function of onboard system in Chinese Train Control System level 3 (CTCS-3) as representative to study the method. This function was analyzed by Fault Tree Analysis (FTA) firstly, and a set of timed automata network model of this function is built using the tools of UPPAAL, the bottom events are used to turn to fault models, injected into the whole system model. Then COVER, the real-time test case generation tool, is used to generate the safety test cases from the system model (included fault models) automatically, and states transition criteria is customized based on preferences to achieve userdefined test, the test accuracy and efficiency is improved..