Bayesian Network Based Fault Diagnosis In CBTC System

The world railway transportation has been greatly developed in the past several decades. As the brain of railway transportation system, train control systems control the train speed, and guarantee the safe trains distance. With the development of communication and computer technology, train control systems are gradually developing from Track Based Train Control (TBTC) to Communication Based Train Control (CBTC).As a complex operation control system, the CBTC system carries out its controlling functions with integrated hardware and software. There are some specified characters of CBTC, such as polymorphism for failure modes, independence of components, deviation of function and timing sequence, etc. In addition, the system safety is affected by quality of software controlling, human factor and other indirect factor. All these factors make it harder to diagnose system fault.Compared with traditional fault diagnose technology, Bayesian Network has complete theory foundation, perfect inference algorithm, and strong study capability. It is one of the most effective theory models in inference and uncertain knowledge area, and has overwhelming advantages when used to diagnose the fault caused by equipment uncertainty and correlation.Beginning with the introduction of basic principal of CBTC system, this thesis describes the safety functions in CBTC systems. With a typical fault scenario-train speed surpass the safe speed, we model the fault scenario and the direct and indirect cause of the fault using Bayesian Network. The specific faulty scenario includes: exceedance of Safe Speed, incorrect information to driver, incorrect speed and distance supervision and protection, unsafe dynamic speed profile, and incorrect determination and Supervision of EOA (End of Authority).With the powerful Bayesian Network toobox in Matlab, this thesis analyzes the reasoning inference and causal inference for all the BN models. The inference results illustrate the major causes to the fault when the train speed surpasses the safe speed. Meanwhile, with the powerful calculation capability of BN tool, the variations of fault probability as well as the probability of the causes to faults are studied when the model parameters change. With the numerical result, we give some suggestion in the thesis as regards to decreasing the faulty occurrence probability. The Bayesian Network based fault diagnosis method present in this thesis is not limited to the fault scenario described in this thesis. It can also be used to diagnose other fault scenario in CBTC systems.