Frequency Selection for Reflectometry-Based Soft Fault Detection Using Principal Component Analysis

Abstract

This paper introduces an efficient approach to select the best frequency for soft fault detection in wired networks. In the literature, reflectometry method has been well investigated to deal with the problem of soft fault diagnosis (i.e. chafing, bending radius, pinching, etc.). Soft faults are characterized by a small impedance variation resulting in a low amplitude signature on the corresponding reflectograms. Accordingly, the detection of those faults depends strongly on the test signal frequency. Although the increase of test signal frequency enhances the soft fault "spatial" resolution, it provides signal attenuation and dispersion in electrical wired networks. In this context, the proposed method combines reflectometry-based data and Principal Component Analysis (PCA) algorithm to overcome this problem. To do so, the Time Domain Reflectometry (TDR) responses of 3D based-models of faulty coaxial cable RG316 and shielding damages have been simulated at different frequencies. Based on the obtained reflectograms, a PCA model is developed and used to detect the existing soft faults. This latter permits to determine the best frequency of the test signal to fit the target soft fault.

A New Insight on the “S” Shape Pattern of Soft Faults in Time-Domain Reflectometry

“…For soft ones, the problem is much more complicated since they are usually characterized by small, spatially localized impedance changes [ 7 ]. The resulting TDR signature is often very weak, hidden in the noise [ 8 , 9 , 10 , 11 , 12 ]. The defect is all the more difficult to detect when dealing with a complex electrical harness since its weak signature is mixed with many other ones [ 13 , 14 ].…”

Hybrid Intelligent Fault Diagnosis Model Based on Improved MPCA-V for Sensors in a Laboratory-Scale Wastewater Treatment Process

“…Since it is often difficult and also unavailable to obtain extensive failure data from industrial processes, some benchmarks have been put forward for the theoretical study of fault diagnosis, such as the Tennessee Eastman process. 11 Being different from the methods of BPNN and SVM, as a conventional and recognized multivariate statistical technique, independent component analysis (ICA), 16,17 PCA, 18,19 and Kernel PCA (KPCA) 20 have been widely employed not only for dimensional reduction but also for the modeling and monitoring of continuous processes based on the measurements from sensors. Among them, ICA weakens the amplitude and direction information of data variance and emphasizes the independence of implicit information between data, 16 which is not conducive to fault diagnosis of the SBR process with a complex coupling relationship between variables.…”