Advanced Forward Methods for Complex Wire

Fault Modeling
Eric J. Lundquist, James R. Nagel, Shang Wu, Brian Jones, and Cynthia Furse
Electrical and Computer Engineering
University of Utah
Salt Lake City, UT 84112

Abstract—This paper presents novel implementation of for-

ward modeling methods for simulating reflectometry responses of
faults in the shields of coaxial cable or other shielded lines. First,
cross-sectional modeling was used to determine the characteristic
impedance of various wire sections. These values were then
incorporated into longitudinal models to simulate the overall
reflectometry response. The finite-difference method (FDM) is a
cross-sectional modeling method that was used to simulate cross-
sectional characteristic impedance. Results using this method are
accurate within 1% of analytical solutions, and can be simulated
very quickly using in-house codes. The finite-integral technique
(FIT) was also used to model chafes on wires with TEM and
higher order modes. Because FIT is computationally slow, a curve
fitting technique was used to predict the chafe profile within
0.01% of the simulated values. Modified transmission matrix
(MTM) and S-parameter methods were used to provide responses Fig. 1. Flow chart of forward modeling types.
with accuracies within 0.3% of the measured profiles.
I. I NTRODUCTION AND BACKGROUND
Location and diagnosis of faults in aging electrical wiring far less studied than hard faults, and the forward modeling
can enable their timely repair, thus preventing costly and methods described in this paper present novel implementation
potentially hazardous failures. This research focuses on one of these methods specialized to the chafe problem. Chafing
of the most challenging problems in electrical fault location— insulation results in a very small change in the wire impedance,
finding small chafes in the shields of shielded wires (coax, and because the reflection depends on the magnitude of the
twisted shielded pair, etc.). These small faults produce such impedance discontinuity, this results in a very small reflec-
small reflection signatures that in many cases they are un- tion that may be lost in the noise of the environment and
detectable against the background noise in aircraft and space- measurements. Therefore, the problem has previously been
craft [1]. The goal of this work is to use the models developed considered more difficult to solve relative to the impact of
here to design new detection schemes and predict when small hard faults. Fortunately, chafes are much more detectable in
fault detection may be possible. shielded wiring, where noise levels are significantly lower
Hard faults (opens/shorts) have been well studied [2]. These and impedance levels remain more consistent along the wire
faults are easier to find, and traditional reflectometry measure- length, and where controlled impedances are less affected
ments are effective for locating them. However, partial faults, by surrounding structures, environment, and vibration [3].
or chafes, are more difficult to identify and model because Models and analysis of shielded cables, where the external
the fault signatures are small and the electrical system usually environment has little or no impact on the cable, enable
does not show any noticeable symptom until the fault is severe. location of much smaller faults than previously detectable.
Chafes are typically the result of abrasion or vibration against In modeling these wiring systems, two types of forward
other wire or structural members, and they often worsen over models are used together: cross-sectional and longitudinal.
time. Like human health, early detection is key to curing the Cross-sectional models are used to determine characteristic
problem and preventing catastrophic disasters. impedance of wire sections, and these impedances are then
The objective of this wire modeling is to simulate the implemented longitudinally where an overall system response
reflectometry response of the wiring system. Reflectometry is can be obtained using reflectometry. These two modeling
a method of determining wire characteristics from reflections processes are illustrated in Figure 1.
of high frequency electrical signals resulting from impedance This paper provides detailed models of shielded faults and
discontinuities. Finding the small anomalies of frayed wire a method to integrate fault models (including measured data)
before they become hard faults is of significant interest, in a unified forward model that describes the effects of the
yet a challenging problem. These types of faults have been fault in a realistic electrical system. Unique aspects of this
model include its modularity (ability to efficiently integrate
data from multiple simulations and measurements), detailed
fault models (including frequency dependence of the faults),
and the ability to model small faults with great precision
while still incorporating them into a full system model (which
normally has lower precision for more efficient computation).
II. C ROSS -S ECTIONAL M ODELING M ETHODS
A. Finite-Difference Method (FDM)
The finite-difference method [4] is a numerical tool for
solving the generalized Poisson equation,
ρ Fig. 2. left: Simulated damaged dielectric, where red represents the inner
∇ · (r ∇V ) = − , (1) dielectric and cyan represents the inner and outer conductors. The box
0 surrounding the center conductor represents a contour of integration for
calculating q. right: Voltage potential function V and electric field vectors
where V is the unknown voltage potential function, ρ is the E computed using FDM.
charge density function, r is the relative permittivity function,
and 0 is the permittivity of free-space. The basic principle
of FDM is to approximate the Poisson equation by replacing FDM model, and so may be simply fixed at a normalized value
derivatives by finite-differences and to sample the continuous of V0 = 1.0 V.
functions along a discrete, finite grid. The net result is a linear With C now a known value, we are ready to express the
system of equations that may be solved to find the potential characteristic impedance ZF using
fields and characteristic impedance of the wire. r
The fault impedance ZF is found by using techniques out- L
ZF = , (4)
lined in [5], [6] and [7]. First, a 2D cross-section of damaged C
cable is modeled by defining the appropriate dielectric function where L is the inductance per unit length. Although L is
r and boundary conditions for V . An example model is shown an unknown value, it is possible to sidestep this parameter
in Figure 2, which depicts a pie-chafe cutaway of φ = 36◦ by noting that the velocity of propagation up for a TEM
on a slice of RG-58 C/U coaxial cable. Such a model is then transmission line satisfies
solved via FDM to find potentials V , after which a set of 1
up = √ . (5)
electric field samples may be found by applying E = −∇V . LC
A simple method for this procedure is also outlined in [4],
In the absence of any dielectric insulation between the inner
which again uses finite-differences as an approximation to the
and outer conductors, it is a known fact that up = c0 =
gradient operation. Finally, the total charge q per unit length
2.996 × 108 m/s, which is the speed of light in a vacuum [8].
along the inner conductor is found by evaluating Gauss’s law.
Furthermore, because L is independent of any non-magnetic
In integral form, this is written as
I materials within the system, it is possible to write
0 r (x, y)E(x, y) · dn = q , (2) 1
c0 = √ , (6)
S
LC0
where dn is the differential normal vector that points outward where C0 is the capacitance per unit length in the absence of
along the closed surface S, which in 2D is simply a closed any insulation. With this information in mind, we may now
contour. Because E and r are both discretely sampled along a express ZF in a modified form as
r
rectangular grid, evaluation of q may be readily accomplished
r
L LC0 1
through the use of a finite summation along the appropriate ZF = = = √ . (7)
C CC0 c0 CC0
contour. Any choice for S is acceptable, provided that it
completely encloses the inner conductor of the model and fits As we can see, ZF is now independent of any inductance term
within the shield. It is therefore common to simply define S L and the only new information we need is C0 . Fortunately,
as a rectangle, as shown by the box surrounding the center this value may be readily computed by simulating an identical
conductor in Figure 2. system as before, but without any embedded insulation. For
Once the Gaussian contour has been integrated to find q, comparison, the characteristic impedance of an ideal coaxial
the next step is to calculate the capacitance C per unit length cable can then be calculated analytically [8], and demonstrates
of the FDM model. We therefore begin by noting that less than 1% error against the FDM simulations. We therefore
have a simple, accurate method for computing ZF through the
q
C= , (3) use of FDM.
V0 The cable can be modeled using this method by program-
where V0 is the excitation voltage of the inner conductor. This ming the cable dimensions, along with the fault size, in a
value is arbitrarily defined as a boundary condition within the 2D grid as shown in Figure 2. The box surrounding the
center conductor will then be used to calculate the contour
integral from the voltage potential matrix, eventually yielding
the characteristic impedance of the chafe through the process
described in this section. Results for this method compared to
other methods are found in Table I, where RG-58 was used as
the wire model. Measurement of undamaged cable is possible
simply by matching the cable end while connected to a
reflectometry scope. The load on the wire end can be adjusted
until the reflections become zero, thus matching the cable and
determining its characteristic impedance. Measurement of the
characteristic impedance of a single small chafe is much more
Fig. 4. Electric field (left) and magnetic field (right) in coaxial cable with
difficult, and thus methods of validated computational forward 60◦ cutaway. The chafe is 5 cm long and the frequency is 5 GHz.
modeling become rather useful. The analytical solutions for
determining characteristic impedance of coaxial cable have
been well studied and can be found in [8].
B. Finite-Integral Technique (FIT)
TABLE I
C OMPARISON OF CHARACTERISTIC IMPEDANCE RESULTS FOR The finite-integral technique (FIT) is a method that numer-
UNDAMAGED RG-58. ically solves electromagnetic field problems in the spatial and
Method Z0
frequency domains [9]. FDM provides the impedance of the
FDM 51.4Ω chafe on a 2D cross-section for a TEM wire. However, when
FIT 50.0Ω the shield is damaged as shown in Figure 4, the field lines bend
Analytical 50.9Ω and the models are no longer strictly TEM. Thus, FIT can be
Measured 50.0Ω
used to find the cross-sectional 2D characteristic impedance
of a 3D chafe, including the higher order modes developed
when the fields exit the chafe. Although this makes FIT more
computationally expensive, it also allows one to model the
effects of more complex faults at specific, higher frequen-
cies. FIT is therefore a popular method used in commercial
software packages such as Computer Simulation Technology
(CST) [10].
Figure 4 shows the output of an FIT simulation for a 60◦
cutaway in RG-58 C/U coaxial cable at 5 GHz. Because FIT
can be computationally expensive at high resolution, we can
utilize a polynomial curve fitting algorithm to minimize the
points required. This is demonstrated by the impedance profile
shown in Figure 5.
Fig. 3. Undamaged (left) and damaged (right) coaxial modeling dimensions.
Like most iteration-based algorithms, FIT can be compu-
tationally expensive at high resolution or where the point of
TABLE II interest is small and the wire is long. While FDM can be run
C OAXIAL CABLE DIMENSIONS (a − d IN MM ) AND DIELECTRIC in a matter of seconds, FIT simulations can take several hours.
CONSTANTS OF INSULATION r AND JACKET j .
Efficiency and precision can be critical in modeling, yet they
Type a b c d r j
are often mutually exclusive. Instead of running the numerical
RG58 0.41 1.47 1.75 2.48 2.25 (PE) 3.18 (PVC) modeling for every single chafe possibility, we can utilize a
polynomial curve fitting algorithm to minimize the detailed
simulations required.
TABLE III
The quasi-TEM mode simulation using FIT combines the
C ROSS - SECTIONAL IMPEDANCE FOR PIE CHAFE ON RG-58 CABLE effect of both TEM and higher order modes. This method was
CALCULATED USING FDM. used to calculate the characteristic impedance of chafed cable.
Polynomial curve fitting was used to represent the impedance
Fault Width w RG-58 ZF
0 mm 51.4 Ω profile of the faulty coax, as in Figure 5.
0.7 mm 51.9 Ω This method yielded the characteristic impedance of chafe
1.5 mm 53.1 Ω
2.2 mm 55.2 Ω
by programming the dimensions of the cable and the chafe
2.9 mm 62.0 Ω in CST software, which then computes the cross-sectional
impedance. Results for this method are found in Figure 5,
and a comparison this method is found in Table I.
 Fig. 6. Reflections from chafe simulated in FDTD. Here, a sinusoidally
modulated Gaussian pulse is reflected from the impedance discontinuity
Fig. 5. Coaxial RG-58 C/U cutaway angle vs. characteristic impedance at 5 located at 100 m on a 200 m wire.
GHz.

can be expected from such faults. A discretized approach to

III. L ONGITUDINAL M ODELING M ETHODS adjusting the RLGC parameters can be implemented in a cell-
Once the characteristic impedance of the chafe is obtained by-cell manner, where the resulting characteristic impedance
using FDM, FIT, or perhaps measurements, a longitudinal gradually changes across the length of the fault. FDTD can
modeling method can be used to simulate the overall forward then be used to obtain reflectometry responses for chafes and
response. These methods simulate the time-domain reflec- other impedance discontinuities by simulating the reflected
tometry (TDR) response. Although the step-function TDR is voltages and taking the corresponding impulse signal into
presented, these methods can also be applied to pulse-shaped account. These voltages can be obtained at any point along the
TDR, spectral time-domain reflectometry (STDR), or spread- line and source signals can be modified within the software.
spectrum time-domain reflectometry (SSTDR) [11]. The only An example of the transmitted and reflected signals is shown
difference in the simulation process is the multiplication of in Figure 6. Calculations can be done in either the time or
the different source signal in the frequency domain before frequency domain with any type of input signal.
performing the inverse Fourier transform.
B. Modified Transmission Matrix (MTM) Method
A. Finite-Difference Time-Domain (FDTD) Method The transmission matrix method, which evaluates ABCD
The finite-difference time-domain (FDTD) method is a matrices [13], is used to evaluate linear networks and is
computational electrodynamics modeling technique that solves commonly used in microwave engineering [14]. A modified
the differential form of the telegrapher’s equations in the time version of this method was used to simulate the reflectometry
domain. These equations are defined as response of cascaded transmission lines, as shown in the
simple TDR setup displayed at the top of Figure 7. This is
∂V (z, t) ∂I(z, t) called the modified transmission matrix (MTM) method. A
− = R(z)I(z, t) + L(z) , (8) TDR tester, with characteristic impedance ZS , is connected
∂z ∂t
∂I(z, t) ∂V (z, t) to the transmission line as a signal source. A load, with
− = G(z)V (z, t) + C(z) , (9) characteristic impedance ZL , is at the end of wire. The
∂z ∂t
transmission line has a characteristic impedance ZT , phase
where R (resistance), L (inductance), G (shunt conductance),
constant β, and length l. The equivalent circuit is depicted at
and C (capacitance) are the wire or fault parameters, which
the bottom of Figure 7.
can be defined either for the wire as a whole or changed
The source (M1 ), lossless transmission line (M2 ) and the
per cell in the simulation. The voltages and currents at any
load (M3 ) can be represented in ABCD matrices as:
point along the line are simulated, including reflected voltages  
from impedance discontinuities caused by wire chafing. These 1 ZS
M1 = (10)
reflections, along with the input signal, can then be used to 0 1
calculate the reflectometry response [12].
 
cosβl jZT sinβl
One advantage of the FDTD method is that it can be M2 = (11)
jYT sinβl cosβl
used for simulation of faults containing graded (gradual)    
1 0 1 0
impedance changes along the line. Many faults contain graded M3 = = (12)
1/ZL 1 YL 1
changes, so these simulations provide a more realistic method
of determining the types of reflections and signal changes that The consolidated matrix becomes:
 Fig. 9. A multi-section setup with a reactive load.

Fig. 7. TDR setup and equivalent circuit.

Fig. 8. n-section cascaded transmission line.

M = M 1 M2 M 3 (13)
 
(1 + YL ZS )cosβl + j(ZT YL + ZS YT )sinβl ξ
= Fig. 10. Simulated and measured results of the multi-section setup with a
ξ ξ
(14) reactive load shown in Figure 9.
where ξ represents the unnecessary parameters that can be
discarded in order to conserve computational expense. For a
lossy transmission line, M2 can be written as where S(ω) is the source signal in frequency domain and
  H(ω) is the transfer function of the TDR. Figure 9 shows
coshγl ZT sinhγl a multi-section setup with a reactive load. The result is shown
M2 = , (15)
YT sinhγl coshγl in Figure 10, showing good agreement between simulated
where the complex propagation constant γ = α + jβ and the and measured values. Although the step-function TDR is
attenuation constant α is nonzero. presented, this method can also be applied to pulse-shaped
To use the MTM method for TDR, we need to consider TDR, STDR or SSTDR. The only difference in the simulation
the wave propagation of the forward and reflective paths. Fig- process is the multiplication of the different source signal in
ure 8 shows an n-section configuration. TDR data is typically the frequency domain before performing the inverse Fourier
acquired between source (M1 ) and the first section of the wire transform.
(M2 ). The TDR transfer function is essentially the relationship Integration of FIT and MTM: The MTM and FIT methods
or ratio of voltages V1 (incident) and V2 (reflected). were combined to predict the TDR signature for a chafed
The forward path V1 and reverse path V2 are represented RG58 coax, as shown in Figure 11. A Campbell Scientific
by: TDR100 is used as the test source. A shield cutaway of 120◦ ,
" n # 5 cm in length, located at 6.5 ft on a 12 ft RG-58 coaxial
Y cable was synthesized. This stepped voltage TDR produced a
V1 = Mx · Vn (16)
small reflected pulse shown in Figure 12. This is because the
"x=1
n
#A chafe creates two overlapping reflections—one at the start of
the chafe and another, nearly equal and opposite, at the end
Y
V2 = Mx · Vn (17)
x=2 A
of the chafe.
Instead of discretizing the wire into numerous FDTD grids,
where MA denotes the element A of the ABCD matrix M .
the MTM method simply represents the entire structure with
The transfer function is calculated as,
three sets of ABCD matrices, which represent the section
" n # " n #
V2 Y Y before the chafe, the chafe itself and the section after. In
H(ω) = = Mx / Mx , (18) order to have a functional and realistic forward model, the
V1 x=2 x=1A A frequency-dependent characteristic impedance of the chafed
and the time domain response is then simply the inverse wire section is obtained using this proposed method.
Fourier transform, The TDR result of the chafed scenario described previously
is presented in Figure 12. The measured and simulated results
Γ(t) = F −1 {S(ω)H(ω)} (19) agreed excellently at the location of interest. If this simulation
 Fig. 11. Shield damage at 6.5 ft on 12 ft RG-58.

Fig. 14. Chafe dimensions and location used in the S-parameter model.

efficiently.
C. S-Parameter Method
Another longitudinal method is the S-parameter ap-
proach [15]. This method differs from the MTM method
in that matched load conditions are used to determine the
matrix parameters, rather than open/short conditions. The S-
parameters are then defined by transmitted/reflected waves and
reflection coefficients.
In order to simulate the response of the wire system
(the forward model), a system of S-parameter equations was
Fig. 12. Simulated and measured results for the 5 cm, 120◦ shield cutaway derived for the damaged wire case. This case included one
shown in Figure 11. chafed section of length z2 , located at a distance z1 along a
wire of total length zT . This is shown in Figure 14.
The transfer function H(ω) was derived from S-parameter
was done entirely using 3D FIT, it could take several hours to theory using a highly detailed model [15]. The forward voltage
complete depending on the resolution setup. With the defined VM (ω) can then be obtained by multiplying the frequency
fault profile and the assistance of the frequency-domain MTM response VS (ω) of the input (source) signal with the transfer
method, the proposed method took less than a second to function H(ω) of the system. Time-dependent voltage vM (t)
perform the same task. Additionally, with the defined fault is obtained by using the inverse Fourier transform. The fol-
profile, one can easily plot the prediction of 5 cm chafes of lowing equations outline the steps taken to obtain vM (t), the
various angle cutaways on an RG-58 at 6.5 ft. This is shown simulated time-domain reflectometry (TDR) response:
in Figure 13.
With the integration of cross-sectional and longitudinal
VS (ω) = F {vS (t)} (20)
models, the modeling of chafed wires can be made more
VM (ω) = H(ω) · VS (ω) (21)
−1
vM (t) = F {VM (ω)} (22)
The resulting system response from (22) can then be plotted as
shown in Figure 15. This method was validated to be accurate
within 0.3% of measured impedance profiles.
TABLE IV
ACCURACY OF REFLECTION COEFFICIENT |Γ| OBTAINED FROM EACH
METHOD .

Method |Γ|
FDTD 0.01
MTM 0.01
S-parameters 0.003

IV. C ONCLUSION
This paper presents novel implementation of forward meth-
Fig. 13. Prediction of fault signatures on chafed RG-58, with a 5 cm chafe ods used for simulating faults in the shields of coaxial cable
located at 6.5 ft on a 12 ft cable. and other shielded lines. First, cross-sectional modeling was
 putational resources. The S-parameter method was also used
to simulate the reflectometry response in the frequency and
time domains in a matrix approach, with accuracy of about
0.3% of measured reflectometry profiles.
These new methods have proved highly useful for simula-
tion and analysis of complex systems. Results can be obtained
by using detailed models of the faults and a method to integrate
multiple fault models (which can include measured data) in a
unified forward model that describes effects of the fault and its
surrounding system. Models of shielded cables can be used,
where the external environment has little or no impact on
the cable, and thus potentially enable the diagnosis of much
smaller faults than have previously been detectable. This can
lead to accurate identification, location, and diagnosis of faults
with high precision, providing real solutions for greater safety
and reparability in aerospace wiring systems.

Fig. 15. Forward voltage response of system with chafe located at 6 m on R EFERENCES
a 7 m line. The tiny reflection resulting from the chafe is circled. [1] C. Furse, P. Smith, M. Safavi, and C. Lo, “Feasibility of spread spectrum
sensors for location of arcs on live wires,” Sensors Journal, IEEE, vol. 5,
pp. 1445 – 1450, Dec. 2005.
[2] Y. C. Chung, N. Amarnath, and C. Furse, “Capacitance and induc-
used to determine the characteristic impedance of various tance sensor circuits for detecting the lengths of open- and short-
wire sections, and then longitudinal methods were used to circuited wires,” Instrumentation and Measurement, IEEE Transactions
on, vol. 58, pp. 2495 –2502, August 2009.
simulate the reflectometry response using these characteristic [3] L. Griffiths, R. Parakh, C. Furse, and B. Baker, “The invisible fray: a
impedance values. critical analysis of the use of reflectometry for fray location,” Sensors
The finite-difference method (FDM) is a cross-sectional Journal, IEEE, vol. 6, pp. 697–706, June 2006.
[4] J. R. Nagel, “Solving the generalized poisson equation using the finite-
modeling method that was used to simulate cross-sectional difference method (FDM),” Jan. 2011. Feature Article, IEEE APS
characteristic impedance. Results using this method are ac- Homepage.
curate within 1% of the analytical solution for characteristic [5] M. F. Iskander, Electromagnetic Fields and Waves. Waveland Pr. Inc.,
2000.
impedance of a coaxial cable and can be simulated rather [6] M. N. O. Sadiku, Numerical techniques in electromagnetics. CRC Press,
quickly using an efficient algorithm which was develeoped. 2001.
[7] M. E. Kowalski, “A simple and efficient computational approach to
Because chafed wire carries a mix of TEM and higher order chafed cable time-domain reflectometry signature prediction,” Stinger
modes, the finite-integral technique (FIT) can be used to model Ghaffarian Technologies (SGT), Inc., NASA Ames Research Center,
chafes between these modes. Because FIT is computationally Moffett Field, CA 94035.
[8] F. Ulaby, Fundamentals of Applied Electromagnetics. Prentice Hall,
slow, by simulating limited number of points, a curve fitting 1999.
technique can be used to predict the chafe profile within 0.01% [9] S. Rao and G. Gothard, “Recent techniques in electromagnetic modeling
of the simulation values. A polynomial equation is generated and analysis,” in Electromagnetic Interference and Compatibility, 1995.,
International Conference on, pp. 131 –137, December 1995.
to represent the chafe fault profile that can be used in the [10] “Computer Simulation Technology, Inc.” http://www.cst.com.
inversion scheme to predict the nature of the fault. The coaxial [11] P. Smith, C. Furse, and J. Gunther, “Analysis of spread spectrum time
cable example is presented, but the same concept can be used domain reflectometry for wire fault location,” Sensors Journal, IEEE,
vol. 5, pp. 1469 – 1478, dec. 2005.
in other wire types. [12] J. R. Andrews, “Deconvolution of system impulse responses and time
The modified transmission matrix (MTM) and S-parameter domain waveforms,” Picosecond Pulse Labs, November 2004. Applica-
methods provide quick yet realistic solutions to transmission tion Note AN-18.
[13] S. Wu, V. Telasula, and C. Furse, “Forward modeling and noise analysis
modeling. MTM simplifies the transmission line by repre- with ABCD method,” 2011.
senting each line section with a single ABCD matrix. Thus, [14] D. M. Pozar, Microwave Engineering. John Wiley and Sons, 2005.
modeling a cascaded transmission line system is easily done [15] S. Schuet, D. Timucin, and K. Wheeler, “A model-based probabilistic
inversion framework for wire fault detection using TDR,” I2MTC 2010
by connecting the modulized blocks. The ABCD matrices Special Issue of IEEE Transactions on Instrumentation and Measure-
not only represent the characteristic impedances of trans- ment, May 2011.
mission lines, they also characterize the reactive components
such as capacitors and inductors. These frequency-dependent
components are typically not simulated in the time domain
methods due to the limited capabilities. Unlike its time domain
counterparts which divide wires into numerous small sections
or meshes, the performance of the MTM and S-parameter
methods does not depend on the resolution of the transmission
length. Therefore, the modeling process requires fewer com-