Vous êtes sur la page 1sur 7

VA TECH Transmission & Distribution

DESIGN PROCESS FOR RELIABLE AND COST EFFECTIVE


TRANSFORMERS
M. P. SARAVOLAC
VA TECH T&D (VA TECH Peebles Transformers Ltd., UK)

Summary
Modern transformer manufacturers operate in a very
competitive environment, under a continuous pressure
to optimise their products without compromising quality
and reliability. The mature transformer technology
offers a limited scope for significant improvement. As a
result, the emphasis is placed on the design and
manufacturing processes, in order to generate market
differentiators.
An example of the design optimisation process, capable
of refining the design at the early proposal stage, to the
extent when all detailed verification calculations can be
executed, is briefly presented in the paper. This
algorithm can ensure that the true; optimum, verified
and manufacturable design is found at the proposal
stage. Running the detailed verification calculations
from the early stage in the process increases the
probability of producing a reliable design.
A specific approach to optimisation of the transformer
dielectric design is also shown, utilising an example of
the inter-winding insulation configuration.
The dielectric system optimisation may be accompanied
with the increased risk of generating excessive stresses
within the area of interest. In order to effectively control
this risk, an approach was adopted which requires clear
understanding of dielectric phenomena associated with
particular configurations and overvoltage transient
conditions on one hand and knowledge of the failure
probability and its mechanisms associated with the
same type of insulation structure and applied load on
the other.
Existence of the sophisticated design tools capable of
accurately determining the distribution of electric
stresses in the time domain and at a high level of spatial
resolution is critical in this respect.

It is shown in the paper that once the knowledge of the


breakdown statistics associated with particular
configurations and test conditions is known, it is
possible to integrate the on-line risk management in
the design process, thus ensuring a high reliability of
the generated design.

Keywords
Design Optimisation Dielectric Design Design
Process Failure Probability Risk Management

1. Introduction
Power transformers are the key assets in power systems.
Failure of transformers in service, which can sometimes
be catastrophic, can have enormous impact on cost,
security and reliability of supply, as well as on other
neighbouring equipment in the substation, the
environment and safety of personnel.
Modern electricity companies require cost effective and
reliable transformers. In order to remain competitive,
modern transformer manufacturers face a very difficult
challenge; they need to optimise their products for the
minimum initial or the total capitalised cost, without
compromising the quality, so that the reliability and the
outage-free service length is unaffected, if not
improved.
The present conventional transformer technology is
known to be mature, with limited scope for significant
improvements. In order to generate market
differentiators, transformer manufacturers have to input
significant effort in improving their core processes such
as design and manufacture.

VA TECH Transmission & Distribution


2. Design optimisation proposal stage
The proposal design requires the derivation of an
optimum design in a minimum design time. It creates a
base for the detailed production design. If this design is
not detailed enough it is not possible to run all
verification calculations at this point, and it may happen
that during the contract design stage some verification
calculations require design modifications that may move
the design away from the optimum point and can
cause additional costs. It is possible that there is another
optimisation point, where all verification calculations
are satisfied but may not be picked up in the coarser
optimisation process.
In this respect it is beneficial that design can be detailed
to the extent at which the key verification calculations
can be carried out during the optimisation process. This
process can ensure that a true optimum point is found,
and the design is technically verified and also easy to
manufacture. The proposal design time is limited by the
tender deadline date and it is not practical to detail all
possible combinations of design parameters. In addition
to this, the requirement for reduced invitation to closing
date time periods is sometimes compounded by the
requirement for increased schedule data.
The above indicates a need for an effective optimisation
algorithm that will require a limited number of designs
(design parameter combinations) in order to arrive to an
optimum point. Since the number of combinations is
reduced, more detailed verification calculations are
possible. Execution of detailed verification calculations
at multiple points during the proposal design process
minimises the probability of a need for design
modifications during the contract design stage, thus
ensuring that the optimum and reliable design will be
found. This will also eliminate a need for using
unnecessarily high safety margins during the proposal
design stage, which could also affect the optimum
solution.
In general, the objective of any design optimisation is to
find the optimum set of design variables, (x1,,xn), that
will provide the extreme limit (minimum or maximum)
of the chosen design characteristic, called the object
function, f(x1,,xn ) [1].
The design variables for transformer optimisation
usually include, but are not limited to: winding space
factors, current densities, flux density, core dimensions,
etc., while the object function is usually the cost, either
initial or the total capitalised cost.
In every practical design there are constraints on design
variables either individual, e.g. current density, flux
density, stress level, etc. (i=1,n, n: number of design
variables);

xi min xi xi max

(1)

or in a form of functions of design variables,

Lmin g ( x1 ,..., x n ) Lmax

(2)

These constraints, also called state variables, can


usually include performance guarantees such as
reactance, losses or temperature rises, dimensions (e.g.
insulation clearances), transport weight or local specific
manufacturing limitations, etc.
The transformer optimisation object function can be
written in a form:

f ( x1 ,..., x n ) = TIC + C CL + PC

(3)

where:
TIC - Transformer Initial Cost
CCL - Capitalised Cost of Losses
PC - Penalty Costs (for not meeting performance
requirements).
There are two different commonly used approaches to
the solution of this general design optimisation problem.
The first method, termed the multiple design method, is
based on generating a large number of alternative
designs, covering combinations of the selected key
design variables.
The design engineer provides input as close as possible
to the perceived optimum point keeping the parameters
within the pre-defined constraints.
The optimisation algorithm [1] is searching for the
minimum of the object function in several steps, each
time refining the pre-set increments of the design
variables and performing thousands of different designs
in accordance with the input data. Constraints for design
variables are also defined and the penalty cost functions
apply for values of design variables outside these
boundaries.
The design generation programs and performance
verification subroutines integrated in this process are
simplified (coarse designs based on winding space
factors) due to the requirement to generate designs for
literally all combinations of the key design variables.
The target is to find the optimum design i.e. to find the
minimum of the object function that meets all
performance requirements and design constraints.
The second approach utilises the non-linear or
geometric programming method, where an object

VA TECH Transmission & Distribution


Fig.1 An example of the plot of the capitalised cost,
function of the LV and HV current densities, in m.

function is minimised, subject to a defined set of


inequality
constraints.
A
classical
geometric
programming problem requires clear definition of the
cost and constraint functions in terms of the key design
variables [2].

f ( x1 ,..., x n ) = a0 + a j1 x j + a j 2 x 2j

This method requires the development of the


mathematical model for each specific transformer
type/configuration in advance, before commencing the
optimisation process, which is its main disadvantage.

The constraint function of design variables is assumed


to be of the same shape:
n

g ( x1 ,..., x n ) = b0 + b j1 x j + b j 2 x 2j

Another disadvantage is in the number of coefficients in


polynomial approximations of the object and constraint
functions, typically over 1000. Such an algorithm is
lacking flexibility and cannot easily be combined with
more general performance verification or cost
estimation programs.

(5)

j =1

g ( x1 ,..., xn ) Lc
where Lc is the constraint limit and the optimisation
problem can be defined as:

F ( x1 ,..., x n ) = f ( x1 ,..., x n ) + l g ( x1 ,..., x n )

The main advantage of this method is that it provides


the optimum design in one attempt. In addition to an
optimum
set
of
design
parameters,
the
design/optimisation program based on this method also
provides "sensitivity factors" i.e. quantities which
represent the change in cost with respect to a change in
a given parameter.

(6)

where F(x1,,xn) is an auxiliary function and, l is a


Lagrangian multiplier for a particular constraint [3].
The method is based on generating a pre-defined
number of trial designs, M. This number is higher than
(2n+1) where n is the number of design variables. In
this way an over-determined system of equations is
generated with more equations than unknowns. In
matrix form it can be written as:

The third approach, featuring a requirement for


generating a limited number of proposal designs during
the optimisation process, is derived from the general
geometric optimisation method [1]. This alternative
constrained optimisation method enables execution of
the more detailed verification calculations for every
individual step of the process. It offers a more flexible
optimisation algorithm that can be combined with a
variety of application specific design synthesis
algorithms.

{ f } + l {g} = [X ]{a} + l [X ]{b}

(7)

where {f},{g}, {a} and {b} are vectors of dimension M


[X] is a matrix of dimension (2n+1) x M.
The above system of equations can be solved in the least
squares sense for coefficients of vectors |a |and |b|.

The form of the polynomial object function (parabola) is


selected to represent as closely as possible the shape of
the cost function which is known from experience to
have shallow minima as shown in Figure 1.

The auxiliary function can have an extreme only when


all partial differentials for all design variables are equal
to zero. This is used to form a system of linear equations
which is used to express design variables in the
following format:

Cost
x 106 []

xi = w(l , ai1 , ai 2 , bi1 , bi 2 ) i = 1,...n

1.765

(8)

After substitution of xi in equations (5) and solution for


l, the design variables can be expressed in terms of the
coefficients of the object and constraint functions:

1.76

4
1.755

(4)

j =1

3.5
s it
en
td
n
rre
Cu
2]
LV
mm
[A/

4
3.5
HV C
u
3
[A/ m 2rre nt de n
m]
sity

2.5
2.5

xi = v(ai1 , ai 2 , bi1 , bi 2 ) i = 1,...n

(9)

Design variables determined in this way are used to


generate a new design, which is added to the initial list
of trial designs and the process of deriving the
polynomial coefficients is repeated. The process is
3

VA TECH Transmission & Distribution

3. Dielectric design optimisation


One of the areas that can offer a significant potential for
optimisation is the dielectric system design.
Optimisation of the dielectric system design is of vital
importance especially for the manufacturers of the High
Voltage transformers. Optimisation of the insulation
clearances in the main inter-winding and end insulation
configurations may result in an overall transformer size
reduction. This in turn can result in a reduction in
material content and associated labour cost.
Optimisation of the dielectric system can be considered
as another constraint optimisation problem, when the
minimum of the object function (insulation distance) is
sought while the critical design parameters, such as the
maximum local level of electric stress, are kept within
the pre-defined margins. These constraints are defined
by the manufacturer and depend on the admissible level
of probability of initiation of the dielectric failure (e.g.
PD inception) forming part of the adopted design
verification criteria.

It is possible now to achieve a very fine spatial


resolution by modelling transformer windings as very
fine networks, adopting individual disc or turns as
geometric entities and using the well-proven lumped
parameter models for transient analysis. With such
sophisticated tools it is possible to locate the point of the
maximum stress, taking into account the actual
insulation structure and winding design. It is also
possible to derive the variation of this stress in time
during the application of the overvoltage transient. An
example of the stress transient is shown in Figure 2 for
the case when the maximum stress is located on the
edge of the conductor on the inside surface of the
impulsed HV winding of the GSU Transformer, facing
the main high to low voltage winding gap.
Stress transients on selected section
30
Electric stress [kV/mm]

stopped when the cost and the design variables of the


last design and the previous designs are within predefined tolerances.

Emax

25
20
15
10

T50

tEmax

This is a complex task, which requires consideration of


simultaneous effects of various test and system overvoltage transients on the transformer insulation. The
objective is to arrive at an optimum configuration
ensuring at the same time that all stresses appearing in
the insulation structure have adequate dielectric strength
and are within the pre-defined safety margins. In order
to solve this optimisation problem and find the optimum
design without compromising the quality and reliability
of the product, transformer manufacturers require
accurate knowledge of the following critical aspects:

Location, amplitude and waveform associated


with the maximum local electric stress,
Probability of dielectric failure associated with
the particular level of stress and its waveform,
Failure mechanisms.

An example of optimising the main (radial) insulation


between transformer windings, subjected to an impulse
voltage, has been selected to briefly illustrate the
approach within the limited scope of this paper.
3.1 Location and amplitude of the maximum stress

10

15

20

25

30

35

40

45

50

55

60

65

Time [ms]
Axial Stress

Radial Stress

Resultant Stress

50% Emax

Figure 2.
The value of the maximum electric stress, Emax , is
critical for evaluation of dielectric strength. In this
respect, the accuracy of the calculation is very
important. For the insulation configuration under
consideration, of particular interest is the electric stress
on the surface of conductors that face the main radial
clearance between two windings of interest. The
transient analysis combined with the knowledge of the
particular winding design can provide voltage transients
at all nodes/conductors in the area of interest.
With this information the task of calculating the stress
transients and determining the location and incidence of
the maximum stress is relatively simple. For the known
set of the boundary conditions the local maximum stress
can be determined accurately by performing finite
element analysis (FEA). Figure 3 shows an field plot for
such configuration.

The problem of predicting the distribution of the voltage


transients within the transformer windings is well under
control by modern transformer manufacturers who use
ever more accurate and sophisticated design and
analytical tools.

VA TECH Transmission & Distribution


3.2 Acceptance criteria and probability of failure
1

Contours of insulated conductors


Disc winding section

Detail shown in Figure 4.

Conductor edge protection

Figure 3. Field plot of the area containing the maximum


local stress.
1

Area: 0.9*Emax E Emax

ANSYS 5.6
NOV 21 2001
15:13:50
NODAL SOLUTION
STEP=1
SUB =1
TIME=1
EFSUM
(AVG)
RSYS=0
PowerGraphics
EFACET=1
AVRES=Mat
SMN =3.757
SMX =27.004
3.757
6.34
8.923
11.506
14.089
16.672
19.255
21.838
24.421
27.004

The knowledge of the actual stress waveform enables


the design engineer to derive the amplitude and the
incidence time of the peak electric stress, and the time
interval during which the amplitude of the transient is
over 50% of its peak value. This allows a derivation of
the equivalent peak value for the standard test
waveform.
With this knowledge the design engineer can evaluate
the dielectric strength at the particular point taking into
account the waveform specific design verification
criteria. These criteria are based on the knowledge of
the admissible level of stress and associated dielectric
breakdown probability. This information is based on
historical and experimentally derived data and is
specific for particular insulation configurations and
applied voltage waveforms. An example of the curve
showing the dependence of the cumulative breakdown
probability curve on the electric stress level is given in
Figure 5.
Cumulative Breakdown Probability Distribution
Configuration Type: #
m = 9.862, Emean BD = 63.79
T50 = 50 ms

Probability

Insulation Barriers

0.011
0.01
0.009
0.008
0.007
0.006
0.005
0.004
0.003
0.002
0.001
0
0

10

20

30

40

50

Electric Stress [kV/mm]

Figure 4. Maximum local stress at the edge of


conductor. Magnified detail as shown in the Figure 3.
In order to fully utilise the capability of the FEA for
deriving the stress transients for the whole duration of
the applied test transients, the analysis would need to be
executed for every moment of interest during the
transient period. This is not practical. An alternative
approach needs to be adopted, whereby the parametric
FE models of the typical configurations are used to
derive databases containing stress values for practical
combinations of the key design parameters. In addition
to the values of the peak local stresses, these databases
also contain information required to estimate the
volume of oil under maximum stress. This information
is then integrated in the transient analyses programs to
be accessed and extracted on-line during the calculation
of stress transients.

Figure 5.
The Weibull distribution model is known to provide a
good correlation with the experimental data [4]. Using
this model the breakdown probability can be expressed
as a function of the ratio of the calculated maximum
local stress and a mean breakdown value experimentally
derived for a particular configuration of interest.

p = 1- e

-a (

Emax m
)
EmeanBD

(12)

where:
p probability of dielectric breakdown,
Emax - maximum local stress,
Emean BD mean breakdown voltage experimentally
derived for the particular configuration and applied
waveform,
a,m scale and shape parameters, respectively,
configuration and test voltage waveform specific.

VA TECH Transmission & Distribution


The acceptable level of stress is defined in conjunction
with the above probability distribution model. One of
the widely used verification criteria for evaluation of the
calculated peak stress value is the Volume Under Stress
criterion [5]. The evaluation of the calculated peak
value is done by means of the curve (derived for the
acceptable probability of failure), which provides the
admissible value based on the volume of oil under stress
level of 90% to 100% of the Emax.

E adm ( p adm ) = cv V90z -100%

(13)

where;
Eadm(p) admissible value of the stress for admissible
breakdown probability corresponding to volume under
stress,
Cv and z empirical constants, depending on applied
voltage waveform,
V90-100% volume of oil under 90 to 100% of the
maximum calculated local stress value,
padm admissible breakdown probability, e.g. 1 in 1000.
The verification criterion can now be defined as:

E max E adm ( p adm )

This knowledge is necessary for effective identification,


assessment and avoidance or mitigation of potential
risks. In general, risk management can be defined as the
process of anticipating risks, determining appropriate
plans to effectively avoid or lessen the potential risks,
recognising the occurrence of risks and taking timely
action to counter their effect [6].
Once the potential risks are identified, simple and
effective techniques can be used as a guide to the risk
management action and required level of resources.
One example of these techniques requires the
construction of a Risk Map or Probability/Impact Grid
when the impact and probability of risk factors have
been evaluated as Low, Medium or High, using the
application specific set of measures. After determining
the likelihood and probable consequences of potential
risk events it is possible to prioritise and define action
plans to control the risk. These actions can range from
those of eliminating risks with potential medium or high
impact, to those of mitigating the effect of those risks
with potentially low probability of occurrence and
impact.

(14)
RISK MAP PROBABILITY vs IMPACT GRID

The above criterion defines the boundaries of the


optimisation process. The optimisation process is
contained within the limits defined by this criterion for
all overvoltage transients of interest.

The design optimisation approach shown in this paper is


capable of producing competitive designs for the
specified performance requirements. The optimisation
objectives are achieved without compromising quality
or performance requirements of the product.

IMPACT

4. Risk management and design optimisation

Elimination

High

Medium
Mitigation

Low
Low

Medium

High

PROBABILITY

In order to achieve this, the knowledge of the process


capability is critical including the knowledge of the
inherent sources and nature of risks and their probability
and consequences. The scope of the potential risk areas
can cover a wide range of aspects, from commercial,
financial, technical to health and safety.
For example, if the proposal design optimisation
algorithm is not convergent or stable enough it may
result in selecting a non competitive base design, or if
there is insufficient knowledge of the mechanical,
thermal or dielectric withstand capability of a particular
configuration, uncontrolled increase of localised stresses
may result in excessive ageing or damage or even a
failure.

Figure 6.

5. Conclusions
Modern power transformer manufacturers require
effective design processes capable of achieving the
optimum design without compromising the quality and
reliability of the product.
A design process that can enable execution of the
detailed verification calculations from the beginning of
the proposal design stage provides an effective answer
to this requirement.

VA TECH Transmission & Distribution


Optimisation of dielectric systems can result in
significant benefits, but at the same time it may be
associated with an increased risk of generating
excessive stresses within the optimised configuration.

7. Bibliography
[1] M.P.Saravolac; Use of Advanced Software
Techniques in Transformer Design; (IEE (P7)
Colloquium on Design technology, an integrated
approach to the design of T&D plant, 17th June 1998,
Grey College, Durham University).

This risk can be controlled throughout the design


process if a clear understanding exists of the failure
probability and mechanisms associated with the
calculated local stresses.

[2] J.H.McWhirter; Optimum Design by Geometric


Programming, (IEEE Summer Meeting, Portland, Ore.,
July 18-23, 1971).

To achieve the above an effective design process is


required that can provide accurate calculation of the
stress transient at the location of the maximum stress. A
full on-line support of the Finite Element Analysis can
be provided during the calculation of the stress
transient, based on using data generated by multiple
executions of the parametric FE models for
configurations of interest.

[3] Applied Dynamic Programming; R.E.Bellman &


S.E. Dreyfus; (Princeton Press, 1962).
[4] S. Yakov on behalf of CIGRE WG 01/TF 02 of SC
15; Statistical Analysis of Dielectric Test Results,
(CIGRE Brochure 66 of 1991).

On-line risk management can be integrated in the design


process if the knowledge of the breakdown statistics
associated with particular configurations and test
conditions is known.

[5] S.Palmer, W.A.Sharpley; Electric Strength of


Transformer Insulation, (Proc. IEE, Vol.116, No. 12,
December 1969).
[6] John Garside, Risk Management; (John Garside
Associates, November 1997).

Once the risk factors are identified and the probability


of their occurrence is known, simple and effective
techniques, such as Probability Impact Grid can be used
for risk management.

6. Acknowledgements
The author would like to thank Messrs. W. Seitlinger,
T. Bickley and J. Fyvie of VA TECH Transformers, for
their useful comments and suggestions.

2002
For further information please contact the author at contact@vatech-td.at
The original document is available at http://www.vatechtd.com
7

Vous aimerez peut-être aussi