On-Line Fault Diagnosis Device Of Train Communication Network

Train Communication Network(TCN),as the central nerve of data transmission and information interaction in trains,undertakes the important task of transmitting data such as control commands,status monitoring and fault information.Once a fault occurs,it will affect the safe and stable operation of the train.With the rapid development of machine learning algorithms,data-driven fault diagnosis can use a large amount of labeled state data to obtain the mapping relationship between fault features and failure modes,so as to accurately identify failure modes,and has been successfully applied in many industrial fields.However,due to the lack of reliable in-vehicle network health diagnosis devices,the network status data at the moment of failure cannot be obtained in time,and the fault diagnosis is difficult and the results are hysteretic.In addition,due to the limited hardware and software resources in the vehicle environment,new challenges are raised for data acquisition,data storage,and online deployment of diagnostic algorithms.Therefore,it is of great significance to carry out the research on the online fault diagnosis device of the train communication network and to diagnose the network fault timely and accurately,which is of great significance to improve the reliability of train operation and reduce the maintenance and repair cost.Based on the analysis of the communication mechanism and failure mechanism of the train communication network,the research designs and develops a network online fault diagnosis device.The machine learning algorithm is deployed online in the device to identify common network failure modes and realize online fault diagnosis.The main research contents of the thesis are as follows:Firstly,taking Multifunction Vehicle Bus(MVB)as the main research object,combined with relevant standards,the communication mechanism and fault mechanism were analyzed,and on this basis,an online fault diagnosis device for train communication network was designed.It mainly includes waveform acquisition and buffering module,data processing and transmission module,power supply module,and analyzes the function and main circuit of each module;the software part designs the network data acquisition and processing flow,and analyzes the training cost,interpretability and model in terms of deployment and other aspects,various machine learning algorithms are compared and analyzed.Because the decision tree has the advantages of strong interpretability and low requirements for software and hardware resources,the decision tree algorithm is finally used for online network fault diagnosis.Secondly,In order to realize the automatic acquisition of network A/B dual-channel data by the on-board device,the method of sending control voltage is programmed and controlled by embedded software,and the automation of voltage-triggered acquisition and buffer reset of the device is realized;by controlling the programmable clock divider in the high-speed AD sampling chip realizes the synchronous acquisition of MVB dualchannel physical waveform data,and splits and restores the dual-channel mixed data.In addition,in order to solve the problem of limited storage space of the online diagnosis device,the data compression method of Huffman coding is adopted to compress the data files before storing,which can effectively reduce the space occupancy rate of the data.Thirdly,a decision tree algorithm BO-CART model based on Bayesian optimization is proposed,which extracts the steady-state amplitude,maximum value,minimum value,slew rate,edge distortion and other characteristics from the network physical layer waveform.The optimization method is used to find the optimal parameters and establish a diagnostic model.The test experiments in the public data set verify the effectiveness of the optimization method.After the trained model is visualized,it is rewritten in C language,and then deployed to the diagnostic device to realize online fault diagnosis.Fourthly,the MVB network platform is built in the laboratory environment,and the simulation of common fault types is completed by artificially injecting faults,and the algorithm verification and performance test of the online fault diagnosis device are carried out.The experimental results show that the fault diagnosis accuracy of the device is Reaching 99.1%,the time for diagnosing 1000 pieces of data is 23 ms,which verifies the effectiveness of the device.