Evidence reconstruction for information fault-tolerance in fusion diagnosis
融合诊断中信息容错性的证据重构方法

In the fusion fault diagnosis based on evidence theory, the abnormal information may cause missed diagnosis of faults. To solve the fault-tolerance problem, we propose an evidence reconstruction method. On the basis of the existing frame of fault diagnosis discernment, the variances of multisensor signals are employed to calculate the signal relative reliability, from which new basic probability assignment functions are formed accordingly. The fused evidences are redistributed and the effect of abnormal information is reduced. Principles for selecting the adjustable parameter in this method are theoretically analyzed. Experiments have been carried out on a hardware experimental platform for vibration isolator. The vibration divergence fault is diagnosed by the fusion diagnosis system. Disconnection of the major acceleration signal is considered as the fault-tolerance problem. Finally, result analysis of the comparative experiment shows that the vibration divergence fault can be diagnosed in time in the occurrence of one-channel abnormal signal. The fault-tolerance of the fusion diagnosis system is improved.