Active Reliability Evaluation Of CAN Nodes Using Partially Observed Information

With time going, cable fatigue, aging and damage is inevitable, which can cause intermittent connection problems that will lead to the communication lines suffering from short random disconnections. The fault impacts on both quality drawbacks and threat to personal safety. This study is based on the intermittent connection fault. In this study, we developed a methodology that estimates the reliability of the CAN nodes using the information of the observable node on the network.Firstly, the random walk model for the observable node bus-off time prediction is proposed. The method to complete the missing transmit error count(TEC) records is put forward. It is proved to be effective and reliable by experiment.Secondly, time windows are used to partition the distribution of the error frame in the time axis, and the markovian chain is built in each time window. Then, the model is established to estimate the TEC values of the normal nodes in different time. Furthermore, two ways to deal with unrecognizable error frames are proposed. The experiments turn out the reliability of the model respectively. Furthermore, the Bayesian networks are built and learned with the observable node’s TEC estimated values, TEC deviations and the variables in the time windows. Then, the TEC estimated values are revised with the Bayesian networks.The experiments analyze the revised effectiveness in the condition of different variables.Finally, the prediction of the bus-off time of the normal nodes is studied. The error frames are recognized. Then, the TEC values are estimated. After that, the estimated values of the normal node are revised with the Bayesian network trained by the observable node. Furthermore, the bus-off time is estimated with the random walk model. The experiments prove the effectiveness and reliability of the model.