Condition Monitoring And Health Evaluation Method Of Train Communication Network

With the increase of usage life,there’s an increasing demand for maintenance and repair of the high-speed electric multiple units(EMU).High maintenance costs have gradually become one of the key factors to the sustainable development of the high-speed railways.As one of the significant subsystems of EMU,the train communication network(TCN)is used to transmit important data such as control command,diagnostic message and communication management.However,due to the lack of effective monitoring way of TCN status,it is impossible to evaluate the health status of the network.Besides,the recent maintenance mode of planned maintenance and posterior maintenance may lead to the situation of delayed maintenance time,excessive and high maintenance costs.Therefore,the condition monitoring and health evaluation of the TCN has become one of the hot research issues of the EMU Prognostics Health Management(PHM)system.Taking TCN as research target,a data-driven method for condition monitoring and health evaluation of train communication network is proposed,which provide a theoretical basis for condition maintenance.The main research contents of this dissertation are as follows:(1)Based on the analysis of communication protocol and common faults in the Multifunction Vehicle Bus(MVB),the condition data is collected and the network condition features are extracted from the time domain,frequency domain,and statistical information.The most relevant features are selected to build a characteristic set using the chi-square method.(2)To detect abnormal in time and avoid network performance degradation,a condition monitoring method of TCN based on ensemble learning is proposed.To improve the generalization ability of the algorithm,heterogeneous ensemble method is formed by i Forest(Isolation Forest),SOM(Self-organizing feature Map)and SVDD(support vector domain description)under the way of weight vote,which can monitor network condition accurately.(3)To evaluate the condition of network and repair network before performance degradation,a health evaluation method for TCN based on LOF is proposed.The percentile health score of each device is obtained based on LOF and pauta criterion.Considering the different degrees of influence of each device node on the network,the entropy weight method is used to assign different weights to the devices.After weighted,the percentile health score of the whole network is acquired,realizing the quantization of network condition.In order to accomplish the visualization of network maintenance,health condition is banded according to the threshold,which can provide a theoretical basis for maintenance decision-making.(4)An experimental platform was constructed and multiple sets of fault injection experiments were conducted such as series,parallel,unterminated,crosstalk,and open circuit.The experimental results showed the validity of the proposed method.For the convenience of engineers,a visualized software was designed and related modules of software were tested.