Fault Localization Of Intermittent Connections For CAN Networks

Supported by National Natural Science Foundation of China No.51005205, this dissertation presents a data driven method for fault localization, on the basis of fault confinement mechanism and stochastic characteristics of CAN networks. As a common but hard troubleshooting phe-nomenon, the intermittent connections (IC) may interrupt network communication, which may result in degraded quality of service of the network and system-level failure in severe cases. There-fore, how to diagnose and locate IC in automation applications has been an important issue. The proposed method concentrates on discrete error events that are concurrently captured from differ-ent layers of network, and generalized zero-inflated Poisson (GZIP) models are applied to model the error events. Then fault localization is launched by comparing the estimated confident intervals of GZIP parameters with zero point.This dissertation is organized as follows.In Chapter1, the research background of fault diagnosis and localizing is described in detail, as well as the state of the art in this field. This chapter concludes with a summary of research objectives.In Chapter2, the error events caused by IC are analyzed, and a test-bed is designed to monitor the status of CAN bus and capture data the moment an error frame occurs. By combining node ID and pattern recognizing, useful information can be interpreted, including the time an error emerges, the interrupted packet, the node who originates the error frame, and the length of the error frame.In Chapter3, error events are classified into different types according to error confinement and node interactions of CAN networks. Characteristics of GZIP model are analyzed and the method to optimized the good-fitness of practical distribution of error data is also developed. Based on the stochastic characteristics of faulty nodes, confident intervals are calculated for fault localization. The effectiveness of proposed method has been verified using experiments.In Chapter4, the performance of fault localization is studied from several aspects, such as correctness, stability of fault localization. The study shows, the structure of model can be optimized to achieve better performance of fault localization. Furthermore, GZIP model based method works well despite different types and strength of IC.In Chapter5, the research work is summarized, as well as its scientific contributions. Future work is also presented.In addition, Appendix A proposes a novel data driven remaining useful life assessment method, by which, the mean time to bus-off state of the network nodes can be obtained based on GZIP model. Experimental results show that the proposed method can estimate TEC (Transmission error counter) and predict the remaining life of nodes.