Classification-decision Based Fault Diagnosis Using Evidence Reasoning

As the important branch of Dempster-Shafer(DS),Evidence Reasoning(ER)rule and its evidence combination method proposed recently define the reliability and weight of evidence and explicitly distinguish the difference between them,which is important to the acquiring and performance evaluation of evidence.Moreover,comparing with the traditional Dempster's rule,the ER rule given based on the orthogonal sum and the bounded sum reinterprets the extension of Bayes inference on the power set of discernment frame.This paper researches the classification-decision method based fault diagnosis using evidence reasoning,the main work includes:(1)Fault diagnosis for motor rotor system by using evidence reasoning.An approach about normalization of likelihood function is applied to acquire the diagnosis evidences of fault feature from the casting results of the intervals of fault samples.Calculating evidence reliability according to the inherent error of sensor and capabilities of interval of fault samples in recognizing various of fault modes.A bi-objective optimization model based on Euclidean distance is presented to train and acquire the optimal evidence weights.The ER rule is finally applied to combine multiple pieces of diagnosis evidence that considering reliability and weight factors,and then fault decision-making can be executed on the basis of the combined result.At last,a fault diagnosis experiment of motor rotor system is conducted,and the experiment result shows good diagnosis performance of the proposed method.(2)Fault diagnosis for track irregularities based on evidence reasoning.The abnormal vibration of train caused by track irregularities can lead to poor ride quality and even derailment.Track irregularities should be carefully monitored and diagnosed using appropriate condition monitoring methods to maintain good ride quality and trainstability.Therefore,this paper presents an ER rule-based inference model to estimate the vertical displacement of irregularity using acceleration data measured from in-service train.The performance of this approach is examined in two representative experiments and compared with that of classical neural network-based methodology.The finding shows the superiority of the proposed approach.(3)General classifier using evidence reasoning.According to the research about equipment fault diagnosis by using ER,it can be concluded that fault diagnosis is a multi-attribute classification decision making problem in essence.Therefore,this paper further presents a method for the design of general ER rule-based classifier,so that the ER rule can be popularized to solve classification problem in the generic sense.In detail,the evidence for each attribute can be acquired by statistical analysis on training samples with known classes;the reliability of evidence can be estimated by analyzing the classifying ability of each attribute,and initial classifier can be constructed on the basis of initial parameters;an optimization model using sequential linear programming(SLP)is proposed to obtain the optimal weight and referential values of each attribute;the ER rule is used to combine the pieces of evidence provided by all attributes of a sample and then make a classification decision according to the fused results.Finally,experiential results on five popular benchmark databases taken from University of California Irvine(UCI)machine learning database show high classification accuracy and universality that is competitive with other six classical and mainstream classifiers.