Abstract—This paper presents new techniques for locating and diagnosing faults on electric wires. There is a practical need to evaluate a cable’s health using the reflectometry method. The proposed fault location algorithm is based on the knowledge of the cable topology. The principle of the approach is to estimate the lengths of the segment in the cable by identifying them with measured pulses by the reflectometry techniques. Since the cable topology is a radial tree network, several possibilities of cable segments could be obtained with one measurement. A statistical algorithm has been developed to rank their possibilities by integrating the available information on the network. Finally, an example has demonstrated its efficiency for diagnosis needs. Index Terms—fault diagnosis, reflectometry, topology, statistic I. INTRODUCTION Cables diagnosis is more and more investigated in order to satisfy the reliability requirements on electrical connections. “Wired network mapping” may be a direct way to diagnose or locate faults. Time Domain Reflectometry has been used in the diagnosis of cable networks [3]. On the other hand, recent researches [1] have been carried out by iteratively simulating the wire network model until the estimated network impulse response matches the initial measurement. In this paper, instead of checking if a likely network is the tested one by iterative simulation, we list all likely solutions of a network and classify them by statistical analysis. In the second section of this paper, the method is introduced to identify the lengths of all sections of a network by using the reflectometry response. In the third section, a CAN network example is given to demonstrate the efficiency of this method. II. METHODOLOGY In this paper, the problematic is to identify the sections of a cable and their lengths from a time domain reflectometry measurement at one end with all the others in open-circuit. A. Assessment of the lengths of the segments The procedure is realized by iterations to fill the tree topology of the cable, as shown in Fig.1. It is assumed that one segment responds to two successors in maximum, and the topologies studied in this paper are composed of segments connected by welding points. Fig.1. A radial tree network Two types of reflected pulses have to be studied: --"Direct Pulses": pulses which travel from the measurement end to a welding point or an open-circuit end and return directly to the measurement end. The sign of these pulses are respectively negative and positive. --"Multi-Reflected Pulses": pulses which travel via more than two welding points or open-circuit ends. These pulses may be negative or positive. As all of the non-measurement ends are in open-circuit, they are all identifiable with a positive “Direct Pulse”. It is relatively easy to locate them by measuring the time difference between the injected signal and the “Direct Pulses”. B. Statistical approach It should be noted that a measurement may correspond to several length combinations for a topology. Therefore, a classification of the evaluated topologies in their representativeness of the measured one would be helpful in practical uses. The method presented in this paper defines two criteria in order to differentiate the lengths estimated in the assessment procedure. 1. If only one measurement is available: as the estimation of the lengths is based on the first several pulses, the first criterion consists in verifying the matching between the estimated lengths and pulses which have not been taken into account in the previous evaluation. 2. If several measurements at different ends are available, A new method of evaluating wired networks topology for fault diagnosis applications Ye Zhu, Marc Olivas, Fabrice Auzanneau CEA LIST, Embedded Systems Reliability Laboratory, Point Courrier 94 Gif-sur-Yvette, F-91191 France; E-mail:marc.olivas@cea.fr ; Phone: + 33 1 69 08 48 83 L01 L11 L12 L21 L22 L23 L24 … … … 978-1-4244-5347-4/09/$26.00 ©2009 IEEE 549