A scheme for the Health Index and residual life of

cables based on measurement and monitoring of

diagnostic quantities
Gian Carlo Montanari1,2
Paolo Seri1 R.E. Hebner2
1
Department of Electrical, Electronic and information
Engineering “Guglielmo Marconi” 2
Center for Electromechanics
University of Bologna University of Texas
Italy USA
Abstract— Defining a methodology providing an objective health, or condition, of the electrical apparatus under
assessment of the health conditions of electrical apparatus is a investigation is still unavailable. The estimation of the residual
must for an effective asset management approach, which can life is even more tricky because it will depend on the diagnostic
provide the most appropriate Return of Investment (ROI) property investigated and its relation to the aging rate of
keeping the asset at the desired reliability level. At present, apparatus sub-components.
diagnostic and monitoring techniques are available to evaluate
electrical apparatus conditions, but too often the output is made For example, if the measured property is partial discharges
by patterns, values and subjective indications which are not able (PD) in a MV cable, its deviation from the base values (for
to provide a real clue to help asset and maintenance manager unaged cable) may lead to significantly different remaining life
decisions. This paper proposes an algorithm for the estimation of estimation. In electrical apparatus ageing is generally a
a health index, HI, for polymeric cable systems based on off-line multifactor process involving a coupled response to electrical,
and on-line measurement of diagnostic markers. The health index thermal, and mechanical stresses. The failure that results from
is a number ranging between 0 (worst condition, indicating that process nearly always presents itself as electrical failure.
imminent failure) and 1 (unaged cable, potentially able to provide and For polymeric cables, the mechanical forces are generally
the design life), which can allow a functional estimation of a minor factor and there are various diagnostic properties which
residual life to be obtained. An example of HI and residual life can be measured to assess cable aging/reliability associated
calculation for a cable system is provided, which shows feasibility
with the main aging factors, electrical and thermal stresses [1].
and potentiality of the algorithm.
Some of those can be measured only off line, other on line,
Index Terms— Cable, Health Index, Residual Life, sporadically or through continuous monitoring. Some
measurements require damaging the cable, such as breakdown
I. INTRODUCTION voltage and water tree counts, but the latter can be considered
A fundamental question from asset and maintenance when doing cable maintenance due to a previous failure. This
managers of electrical facilities is how to establish the condition paper defines an algorithm which can, based on diagnostic
of electrical apparatus and residual life compared to the design property measurement, develop an health index (𝐻𝐼) for new
life. Ideally, one would want a well-established physics-of- and aged cables, and demonstrates how to derive a model for
failure model with measurable parameters that provides reliable residual life estimation based on the 𝐻𝐼 value. Various
insight as to the expected life of the cable. The measured properties are considered, in order to provide a broad
parameters would accurately account for all relevant situations background, which is mostly appropriate for MV polymeric
in manufacturing and previous use. Unfortunately, a cables. However, the approach could be also applied to oil-
sufficiently reliable and comprehensive model has yet to be paper and to HV cables, choosing the appropriate properties.
developed.
II. DEFINITION OF HEALTH INDEX AND CALCULATION OF
But even in the absence of the ideal situation, asset RESIDUAL LIFE
managers have to make the most reliable assessments possible. The health index (𝐻𝐼) is meant to be a simple indicator of
Answers are, in general, based on measurements of diagnostic the health conditions of an electrical apparatus. It is based not
properties and interpretation of the test data. More recently, only on the measurement of diagnostic properties, but also on
condition monitoring devices are being installed on some types various information about age, environmental conditions,
of electrical apparatus (particularly MV generators, HV GIS, operating conditions and previous maintenance of the apparatus
HV transformers and HV cables), but results require often the [2]-[4]. Focusing in the insulation system, which is often the
interpretation of experts and an objective evaluation of the

978-1-5386-7703-2/18/$31.00 ©2018 IEEE

Authorized licensed use limited to: Nanyang Technological University. Downloaded on December 03,2021 at 01:44:10 UTC from IEEE Xplore. Restrictions apply.
weakest component of an electrical apparatus, and considering 𝐴 𝐴
a diagnostic marker 𝑋 (which could be also a vector of values 𝐿 = − (6)
𝐾(𝑆) 𝐾(𝑆)
of different diagnostic quantities), the health index can be
defined as: where 𝐿 is residual life. Taking 𝐴 and 𝐴 in relative values,
so that 𝐴 = 1 and 𝐴 = 1 − 𝐷𝐻𝐼(𝑡 ), eq. (6) provide an
𝐻𝐼 = 1 − Pr(𝐹|𝑋) (1) estimation of residual life, considering that
where Pr(𝐹|𝑋) is the conditional probability of failure given a 𝐴
level of the diagnostic marker 𝑋. Therefore, the larger the 𝐾(𝑆) = (7)
𝐿
probability of failure, the lower 𝐻𝐼. If diagnostic properties are
permanently or intermittently measured as a function of time and life 𝐿 can be expressed for electrothermal stress, by eq. (4).
under operation, 𝑡, a dynamic health index (𝐷𝐻𝐼) can be
defined, according to (1), where the diagnostic property is
function of the time under operation when measurements are
done, 𝑋(𝑡) [4],[5]. The aging function, A, could be expressed
as a function of 𝑋 [6]-[8]:
𝐴 = 𝐹(𝑋) = 𝐾(𝑆)𝑡 (2)
where 𝐾(𝑆) is aging rate under stress 𝑆. The diagnostic marker
𝑋 would be defined having initial value 𝑋 at 𝑡 = 0 and ageing
𝐴 = 0. The limit value for 𝑋 , which indicates end of useful
life (not necessarily breakdown, but high risk for the insulation
to not be able to withstand operating or transient stresses) is
reached when 𝐴 = 𝐴 :
𝐴 = 𝐹(𝑋 ) = 𝐾(𝑆)𝐿 (3)
being 𝐿 the life of the insulation system. At the design stage, 𝐿 Fig. 1. Representation of time behavior of aging function, 𝐹(𝑋), Aging limit,
is the life at a chosen failure probability, according to 𝐹(𝑋 ), health index at time 𝑡 , 𝐴 , and design, 𝐿 , and residual life, 𝐿 (eqns.
specification provided by the manufacturer/client of the (2), (3), (5), (6)).
insulation system. If the design stress estimation is made
correctly, based on the knowledge of the operating stresses and III. ALGORITHM FOR CABLE HEALTH INDEX ESTIMATION
the use of life models which can be extrapolated to the design The first step to structure the 𝐻𝐼 or 𝐷𝐻𝐼 algorithm is to divide
life and stress, the electrical insulation is expected to last the equipment under test (EUT) in different subcomponents 𝑖,
according to the specifications. An example of life model which may have different technologies and materials. For each
typically used to design cable insulation is [6],[8]: subcomponent, a number of 𝑗 diagnostic markers (𝑋 , ) are
𝐿 = 𝐶𝐸 (4) considered and associated to a score, 𝑆 , . Typically, 𝑆 , = 1 ÷
4, but any other range is acceptable.
which is the so called inverse power model, where 𝐸 is
electrical stress at the operating temperature (lower or equal to In the case of polymeric cables, diagnostic markers measured
the thermal class of the insulation [9]) and C a parameter, off line could consist of partial discharge inception, PDIV,
function of temperature. Equation (4) provides a straight line in voltage dissipation factor, tan(𝛿), and water tree content
log log plot, which is extrapolated to design life and field based (endowing two sub-properties which may have different impact
on accelerated life testing [10]. The insulation system should on cable reliability, that is, length and density). Also,
reach 𝐴 in correspondence of the design life, 𝑡 = 𝐿 , but only thermography observations can be done at the terminations and
if no maintenance was required. Depending on the value of 𝐴 joints (when accessible) to look for the presence of hot spots.
or the 𝐷𝐻𝐼 in correspondence of 𝑡 = 𝐿 , life extension plans Water trees, however, could be measured only cutting a
can be developed, increasing the Return of Investment (ROI) specimen from cable, thus, in general, after a failure occurred.
for the asset. Based on the above definitions, and having A property of fundamental importance, partial discharges, PD,
achieved a 𝐻𝐼 value for a cable, an objective estimation of can be measured on-line, possibly through permanent cable
residual life, compared to 𝐿 , could be achieved, at least monitoring. Also PD, however, must consider sub-properties,
formally, as follows. If measurements of a diagnostic marker based on the type of defect which generated discharges. Indeed,
property, or a series of diagnostic markers, 𝑋 , is done at time internal, surface and corona discharges are all able to
under operation 𝑡 , and 𝐻𝐼 or 𝐷𝐻𝐼 are derived, the value of understood as PD pulses and be collected in PD patterns, but
Aging limit for 𝑡 = 𝑡 can be written as (equations (2) and (3)): while the first is generated in defects internal to insulation and
is highly harmful for insulation aging (being able to break
𝐴 = 1 − 𝐻𝐼 = 𝐾(𝑆)𝑡 (5) polymeric bonds and, thus, degrade material very rapidly), the
The residual aging before reaching the limit which corresponds last, i.e. corona, can be just a disturbance coming from outside
to the design life is 𝐴 − 𝐴 (see Fig. 1 where equations (2), (3) the cable, and/or indicate other phenomena potentially harmful
and (5) are represented under the assumption that 𝐾(𝑆) is not for cable life, such as loose grounding bolts and so on. Surface
function of time (S is constant)), thus from (3) and (5): discharges can occur in accessories, e.g. at the interface
between cable insulation and joint insulation/semicon, and have

Authorized licensed use limited to: Nanyang Technological University. Downloaded on December 03,2021 at 01:44:10 UTC from IEEE Xplore. Restrictions apply.
generally harmfulness which is intermediate between internal
and corona discharges [11]. Based on the above considerations,
a weight has to be attributed to each diagnostic marker or sub-
marker, depending on its harmfulness in terms of aging and
breakdown processes. The more the specific diagnostic
property is associated with an aging mechanism which has fast
degradation rate, the more its weight will be large. For this
purpose, weighting factors, 𝑊 , , are applied to each diagnostic
marker and subsystem (ranging, e.g., between 1 and 5). A
weighted average, 𝑊 , , can be used to account for score
and weight of all markers considered for the specific sub-
component of the EUT, both the markers and the sub-markers
(the latter first, then the former):
∑ 𝑆 , ∙ 𝑊,
𝑊 , = (8)
∑ 𝑊,
where 𝑀 is the number of diagnostic markers monitored for the
subcomponent 𝑖. A further step is to take into account all
information coming from visual inspection. It can be carried
out, in general, on accessories, in order to check major/minor
problems and the likelihood of the information provided by a
diagnostic marker. These inspections will result in an Adjusting
Factor, 𝐴𝐹 , which is summed to the 𝑊 , bringing to the
partial score, 𝑃𝑆𝐶 of the subcomponent 𝑖 under test. If data on
the Failure Rate contribution of each sub-component to the total
Failure Rate of the equipment, 𝐹𝑅 , is available (based e.g. on
IEEE publications, papers, insurance reports, or experience in
the asset under consideration), this is another coefficient to be
taken into account in the 𝐻𝐼 estimation. Eq. (9) summarizes
these considerations into the algorithm for 𝐻𝐼 or 𝐷𝐻𝐼
estimation:

𝐹𝑅 ∙ (𝑃𝑆𝐶 − 𝑆𝐶 )
𝐻𝐼 = 1 − (9)
𝑆𝐶 , − 𝑆𝐶 ,

where 𝑁 is the number of subcomponents of the apparatus

under monitoring, 𝑆𝐶 , and 𝑆𝐶 , are the maximum and 2A
minimum score values for the diagnostic indicators of the
subcomponent 𝑖 (e.g. 4 and 1, respectively, where 1 is good
conditions and 4 is bad conditions (action), or alternatively
1=green, 4=red, according a color scheme [2],[5][12]). Finally,
information on failure rate data, equipment loading,
environmental conditions and Age can be taken into account to
amend by an Adjusting Factor, 𝐴𝐹 , the value of 𝐻𝐼 obtained
from eq. (9).
𝐻𝐼 = 𝐻𝐼 + 𝐴𝐹 (10)
If Age is low, and loading is limited on the average, so that the
mean operation temperature is significantly lower than the 2B
thermal class, the final value of 𝐻𝐼 can be decreased Fig. 2. Structure for 𝐻𝐼 calculation for cable systems. Three sub-components
accordingly. As an example, it is worthwhile remembering that, are chosen, i.e. cable, joints and terminations (Fig. 2A), each of one having a
as a rule of thumb, a reduction of 10°C of operating temperature series of diagnostic marker and, in the case of PD and water tree count,
will correspond to an increment of 50% of thermal life [13]. marker sub-classes. Fig. 2B is the continuation of Fig. 2A bringing to the
Note that Age is a concept drastically different from Aging. final 𝐻𝐼 calculation.
While Age, indeed, is operation time from the apparatus
commissioning, Aging is, as shown above, related to Thus, 𝐻𝐼 is fundamentally related to Aging, not Age, but taking
degradation of functional properties. into account also Age can help in finding out an objective figure
for 𝐻𝐼. Also, it is noteworthy that most of the above
coefficients, that is, 𝑆 , , 𝑊 , , 𝐴𝐹 , 𝑃𝑆𝐶 and 𝐴𝐹 are

Authorized licensed use limited to: Nanyang Technological University. Downloaded on December 03,2021 at 01:44:10 UTC from IEEE Xplore. Restrictions apply.
function of time under operation, thus they contribute to the must be underlined that the diagnostic markers could be the
value of 𝐷𝐻𝐼. Figure 2 shows a scheme of the 𝐻𝐼 algorithm for same for each sub-component, i.e. for cable and accessories,
cable systems. Three sub-components are chosen (cable, joints, with the limitation that the value of the dissipation factor,
terminations), each of one having a series of diagnostic markers tan(𝛿), measured on the cable system is largely affected by
and, in the case of PD and water tree count, three and two cable rather than by joints or terminations.
marker sub-classes, respectively.
V. HI CALCULATION EXAMPLE
IV. DIAGNOSTIC MARKERS FOR CABLES AND THEIR The 𝐻𝐼 calculated according to eq. (10), with the above scores
SCORE/WEIGHT and weight and further correction factor, is a number ranging
It is not easy, and it can be subjective, to provide score and between 0 and 1, which can be associated with a simple traffic
weight for the diagnostic properties that can be used to estimate light information, such as red when 0 < 𝐻𝐼 < 0.25, yellow
cable 𝐻𝐼. Considering the above-mentioned properties, some when 0.25 ≤ 𝐻𝐼 < 0.75 and green when 0.75 ≤ 𝐻𝐼 < 1.
indications could be as follows. This would be of immediate understanding for the asset
manager and it can be implemented easily in any SCADA
The score, 𝑆 , , can range between 1 and 4, while weight management software. Then, knowing the measurement time in
between 1 and 5. For PDIV, the property can be defined as relation to Age, equations (6), (7) and (4) can be used to
𝑋, = where 𝑉 is nominal phase voltage, and it can be estimate the residual like (at a given failure probability). As an
measured by off-line measurements using voltage generators as example, let us consider a MV cable system having zero PD in
resonant, VLF or OWTS systems. Weight 𝑊 , could be set the cable, but 250 mV of PD in one joint, dissipation factor
equal to 5, because the presence of PD during operation is very which is two times the initial one (thus score =2), thermography
harmful for cable system reliability. Property ranges and showing a hot spot 10 °C higher than the thermal class in a
relevant 𝑆 , values could be: 𝑋 , ≤ 1, 𝑆 , = 4; 𝑋 , = 1, 𝑆 , = termination (thus causing accelerated thermal aging, so that the
weight could be =3 and the score = 3). Failure statistics for that
3; 1 < 𝑋 , ≤ 1.2, 𝑆 , = 2; 𝑋 , > 1.2, 𝑆 , = 1. type of installation (cable material, brand, accessory material,
Considering PD amplitude, measured off line at nominal technology) indicates 𝐹𝑅 values of 20%, 45% and 35% for
voltage or on line during cable system operation, again the cable, joints and terminations. From eq. (9), 𝐻𝐼 before
weight must be the maximum, i.e. 5, only if the detection application of the last adjusting factor, 𝐴𝐹 , is 0.83.
system or the expert doing the measurement is able to confirm Considering that the thermal history is benign, but the cable has
that PD are generated by a defect internal to insulation (internal 25 years of Age, a value for 𝐴𝐹 could be -0.1, so that the final
discharges). In this case, one can consider 𝑆 , = 1 when PD 𝐻𝐼 becomes 𝐻𝐼 = 0.73. This value may fall is into the
amplitude is negligible (e.g. amplitude smaller than 10 mV), yellow-light category, which indicates attention from the
𝑆 , = 2 for amplitudes between 10 and 100 mV, 𝑆 , = 3 for maintenance manager side. The residual life, considering
PD in the range 100 to 1000 mV and 𝑆 , = 4 for values above design life of 30 years (failure probability 1%) and suitable life
model coefficients suitable for XLPE operating at 90°C, results
1000 mV. For surface discharges, the weight should be lower
21.9 years, which is not a precise value, of course, but a sort of
than for internal PD, e.g., 3, and 𝑆 , = 1 for no or negligible quantitative indication able to support asset manager decision
PD, 𝑆 , = 2 for, e.g., PD amplitude up to 100 mV, 𝑆 , = 3 for about cable replacement vs maintenance policies. This result
PD from 100 to 1000 mV, and 𝑆 , = 4 for values >1000 mV. indicates that the considered cable system is aging at slower rate
Note that to attribute such 𝑆 , PD location has to be of that considered for the design, for example to due to better
established, because amplitude attenuates considerably as a insulation behavior under stress or lower-than-design operating
function of the distance between PD site and measurement stresses.
device. Corona discharges must have low weight, because they
can be of unknown nature and not coming from the cable under VI. CONCLUSIONS
evaluation, or they could come from some problems as As stated in the introduction, this is not a deterministic approach
corroded or loose ground connection, thus 𝑊 , = 1. 𝑆 , could to estimating residual life because the basic information does
be in general taken equal to 2, because it is unlikely that corona not exist to support such an approach. The approach proposed
source can be identified and located. in this paper is a small step toward a Bayesian rather than a
deterministic approach to life prediction. In the proposed
Water trees can be split in two sub-classes, that is 𝑋 , = approach, information is weighted according to its expected
𝑤𝑎𝑡𝑒𝑟 𝑡𝑟𝑒𝑒 𝑑𝑒𝑛𝑠𝑖𝑡𝑦 and 𝑋 , = . The weight influence on aging. If the approach is widely adopted, it should
could be 5 and 3, respectively, as a consequence of the fact that provide information on the uncertainty of the weighting factors
long trees are more harmful for cable reliability than high that could support the application of more rigorous Bayesian
density of small trees. The values of 𝑆 , could be 1 for zero statistics.
trees, 𝑆 , = 2 for 𝑋 , ≤ 10 or 𝑋 , ≤ 0.1, 𝑆 , = 3 for Based on this approach, asset and maintenance manager’s need
10 < 𝑋 , ≤ 10 or 0.1 < 𝑋 , ≤ 0.5, 𝑆 , = 4 for to have simple and clear output from diagnostic measurements
𝑋 , > 10 or 𝑋 , > 0.5. performed on cable systems can be addressed successfully.
This can support appropriate maintenance plans in order to
Let us note that the above values are purely an indication, based maximize the ROI, keeping the electric asset reliability at the
on subjective evaluation and experience. Changing such values desired level.
does not affect at all the approach described here. Finally, it

Authorized licensed use limited to: Nanyang Technological University. Downloaded on December 03,2021 at 01:44:10 UTC from IEEE Xplore. Restrictions apply.
Regarding the example provided in the paper, it must be [5] G.C. Montanari, P. Seri, “A partial discharge-based health index for
underlined that the choices of score and weight can be modified rotating machine condition evaluation”, to be published in IEEE
Electrical Insulation Mag., February 2018.
to give more push to action after critical property worsening. If, [6] L. Simoni, Fundamentals of Endurance of Electrical Insulating
e.g., there is a case where PD amplitude reaches score 4 and, Materials, CLUEB, Bologna, Italy, 1983.
hence, an action should be planned in short times even if the [7] G.C. Montanari, G. Mazzanti., “From thermodynamic to
global 𝐻𝐼 is not indicating red light yet, then the weight of PD phenomenological multi-stress models for insulating materials without
amplitude can be increased, e.g. to 10. In addition, health or with evidence of threshold”, J. Phys. D: Appl. Phys., Vol. 27, pp.
1691-1702, 1994.
indexes for each subcomponent can be taken as a reference for [8] G.C. Montanari, G. Mazzanti, L. Simoni, "Progress in electrothermal life
the maintenance decision. Eventually, a final correction to modeling of electrical insulation during the last decades", IEEE Trans.
global 𝐻𝐼 (relevant to the whole cable system) could be on Dielectrics and Electrical Insulation, Vol. 9, n. 5, pp. 730-745,
provided if one of the diagnostic properties associated to the October 2002.
most harmful degradation mechanisms has become critical (e.g. [9] IEC 60216, Electrical Insulating Materials – Thermal Endurance
Properties – Part 5. Determination of Relative Thermal Endurance Index
the 𝐻𝐼 could enter into an harmonic weighted mean with (RTE) of an insulating material, 2007.
the critical property(ies) score(s)). [10] IEC 61251, Electrical Insulating Materials - A.C. Voltage Endurance
Evaluation, 2014.
REFERENCES [11] A. Cavallini, M. Conti, A. Contin, G. C. Montanari, "Advanced PD
[1] IEC 60505, Evaluation and Qualification of Electrical Insulation inference in on-field measurements. Part 2: Identification of defects in
Systems, 2010. solid insulation ", IEEE Trans. on Dielectrics and Electrical Insulation,
[2] A. Jahromi, R. Piercy, S. Cress, W. Fan, "An approach to power Vol. 10, n. 3, pp. 528-538, June 2003.
transformer asset management using health index," IEEE Electrical [12] G.C. Montanari, P. Morshuis, L. Paschini, “A Smart Grid Approach to
Insulation Magazine, vol.25, no.2, pp.20,34, March-April 2009. Condition Based Maintenance of Renewable Energy Assets”, Saudi
[3] F. O. Fernández, A. Ortiz, F. Delgado, I. Fernández, A. Santisteban , A. Arabia Smart Grid, pp. 1-5, Jeddha, Saudi Arabia, December 2014.
Cavallini “Transformer health indices calculation considering hot-spot [13] V. M. Montsinger, “Loading transformers by temperature”, AIEE
temperature and load index”, IEEE Electrical Insulation Magazine, Vol. Trans., Vol. 67, pp. 113-122, 1948.
33, no.2, pp.35-43, April 2017.
[4] G.C. Montanari, “Condition monitoring and dynamic health index in
electrical grids”, IEEE CMD, pp. 286-289, Xian, China, September
2016.

Authorized licensed use limited to: Nanyang Technological University. Downloaded on December 03,2021 at 01:44:10 UTC from IEEE Xplore. Restrictions apply.