GMM Clustering-Based Decision Trees Considering Fault Rate and Cluster Validity for Analog Circuit Fault Diagnosis

Junyou Shi,Zili Wang,Qingjie He

doi:10.1109/access.2019.2943380

Abstract

Traditional decision trees for fault diagnosis often use an ID3 construction algorithm. For promoting the accuracy and efficiency of decision trees, considering the cluster validity and fault rates, this paper proposes two improved trees, CV-DTs and FR-DTs. This paper mainly has two highlights. The first highlight is to propose a CV-DT which is constructed by an improved ID3 algorithm considering the cluster validity index. A new cluster validity index which can compare the cluster validities of different attributes is proposed to modify the information gain. This method selects the splitting attributes with higher classification credibility and increases the diagnostic accuracy. The second highlight is to propose an FR-DT which is constructed by an improved ID3 algorithm considering the fault rates. This algorithm not only considers the partitioning ability of each attribute, but also considers the isolation priority of faults with higher fault rates. This method decreases the average diagnostic steps and promotes the diagnostic efficiency. Through a simulation case and a real board case, these decision trees are proved to be effective diagnostic tools which have higher accuracies or efficiencies in analog circuit.

Highlights

Due to the basic characteristics of analog circuits such as non-linearity and tolerance of components, inefficient fault models, inadequate accessible nodes, and uncertainty in the measurements, advanced fault diagnoses in analog circuits have attracted a lot of research attention [1]–[7]
One of the information gain entropies is based on the number of clusters and considers the partitioning ability of each attribute
The other is based on the fault rates and considers the isolation priority of faults with a higher fault rate

Summary

INTRODUCTION

Due to the basic characteristics of analog circuits such as non-linearity and tolerance of components, inefficient fault models, inadequate accessible nodes, and uncertainty in the measurements, advanced fault diagnoses in analog circuits have attracted a lot of research attention [1]–[7]. For the purpose of promoting the diagnostic accuracy and efficiency of decision trees, this paper proposes the fault diagnosis using the GMM clustering based decision trees considering the fault rate and cluster validity separately. Focusing on the second drawback, we propose an improved ID3 algorithm considering the fault rate and construct a FR-DT This algorithm firstly determines the number of fault data based on the fault rate.

IMPLEMENTATION ROUTINE

CV-ID3 ALGORITHM

Findings

CONCLUSION