Realignment of railway curves Proc. Instn Civ.

Engrs, Transp.,
1997, 123,
D. O’Dwyer, BE, MEngSc, CEng, MIEI, AFPWI Nov., 234 – 240

Paper 11424

■ This paper describes the correct procedure lying within a bed of ballast, see Fig. 1. The Written discussion
for calculating the lateral displacements ballast suports the sleepers, restricts lateral closes 15 April 1998
necessar y to realign a railway cur ve. The and longitudinal movement of the track, and
procedure is based on the traditional transfers the vertical, lateral and longitudinal
assumptions, uses versine measure, and is loads to the subgrade.
easy to apply. The paper shows that for a 2. Railway track is subject to vertical, longi-
set of design versines to be achievable they tudinal, and lateral forces. These forces and
must satisfy two conditions. First, the sum the flexible nature of ballasted track cause
of the design versines must equal the sum track alignment to deteriorate with traffic.1,2 In
of existing versines, and second, the cen- cur ves, the magnitude of the lateral loads
troid of the new versine diagram must depends on the bogie design, and coefficient
coincide with the centroid of the old ver- of friction between the wheels and rails and
sine diagram. The mathematical basis of the cant deficiency or cant excess.3 Equilib-
the method is derived and the problems rium cant is a function of vehicle speed v,
with traditional procedures are high- cur ve radius R and running gauge Gr and is
lighted. The versine corrections which are given by the equation
necessar y in transition cur ves are given.
Methods are discussed for finding optimal Gr v 2 Grv2
Eeq 5 > (1)
valid solutions to realignment problems.
The use of linear programming in cur ve R Îg 1 Rv
2
4

2
Rg

realignment and a number of linear pro-

gramming formulations are described. 3. Track which carries a mixture of pas-
senger and freight traffic, which travel at dif-
ferent speeds, cannot be canted to avoid
Notation lateral forces. Free cur ving rarely occurs even
C chord length where equilibrium cant has been applied3 In
D cant deficiency tangent track lateral forces occur through
Dmax maximum allowable cant hunting, lateral wind loads and through
deficiency dynamic effects caused by poor alignment.
Eeq equilibrium cant 4. Track alignment must be corrected per-
Ei cant at station i iodically because poor alignment causes pas-
Emax maximum allowable cant senger discomfort and increases wear on both
Gr running gauge 5 track track and rolling stock. Fig. 2 shows the cur-
gauge 1 rail-head width vature diagrams for a circular cur ve with two
g acceleration due to clothoidal transitions as set out initially, and
gravity 5 9·81 m/s2 after its geometr y has deteriorated. When the
k(s) cur vature as a function of position track alignment, or the corresponding ride
on the cur ve index, exceeds the allowable tolerance limits,
L length of cur ve the cur ve will need to be realigned.
Pi slew, pull, or throw at station i 5. There are occasions when railway
Pmini, Pmaxi maximum allowable slews at sta- cur ves need to be realigned for reasons other
tion i than alignment deterioration caused by traffic.
Pmax maximum slew at any station i The cur ve may need to be altered to increase
R radius of cur vature line speed, the existing transitions may be
s distance from the start of the imperfect, or the cur ve may need to be altered
cur ve to increase line-side clearances.
Vi versine measurement at station i 6. Most routine cur ve realignment is car-
u(s) direction of cur ve as a function of ried out using tamping machines. Tamping
position on the cur ve Dermot O’Dwyer,
machines can realign track automatically,
Lecturer in the
v speed: m/s using a transfer function to smooth the cur ve, Department of
or can be programmed or controlled man- Civil, Structural
ually.1,4 When working automatically, tampers and
Introduction perform best when the slews required are Environmental
Most railway track is not rigidly anchored to a small. It is usual to control the tamper man- Engineering,
concrete pavement. Traditional track com- ually when major realignments are required or Trinity College
prises an assemblage of rails and sleepers where the slews to be applied are greater than Dublin
234

Downloaded by [ TUFTS UNIVERSITY] on [26/09/16]. Copyright © ICE Publishing, all rights reserved.
 REALIGNMENT OF
RAILWAY CURVES

can be achieved in a single lift. This involves

sur veying the cur ve manually and calculating
the lateral displacements necessar y to realign
the cur ve.5

The relationship between cur vature,

versines and equilibrium cant Fig. 1. Ballasted track
7. Railway cur ves are set out initially by
standard sur veying techniques using theodo-
lite and tape or by measuring offsets from a
baseline. However, once a cur ve is in place it
is more efficient to sur vey its geometr y by Fig. 2. Diagram of
recording a set of versines.5, 6 Typically a 20 m track curvature
chord length is used with versines being meas- initially and after
ured ever y 10 m (half the chord length), see deterioration with
Fig. 3. The sur vey results are tabulated as traffic
shown in Table 1.
8. Figure 4 shows that the relationship
between the radius of an arc and its versine V
(the perpendicular offset from the mid-point of
the chord to the arc) is given by the equation

C2 V
R 5 1 (2)
8V 2

where C is the chord length. The second term, Fig. 3. Alignment

V/2, is ver y small as compared with the first survey by measuring
term and can be omitted. Therefore the rela- versines
tionship between the cur vature, (k 5 1/R), and
versine can be expressed as

1 8V
k 5 > (3) Table 1. Versine measurements
R C2
Station Distance: m Versine: m
9. The relationship between versine and number
cur vature is approximately linear. Similarly
equation (1) shows that for a given speed the 0 0 0·000
relationship between equilibrium cant Eeq and 1 10 0·005
2 20 0·010
cur vature is also approximately linear.
3 30 0·015
Therefore 4 40 0·020
5 50 0·020
1 Grv2 8Grv2
Eeq > > V (4) 6 60 0·020
R g gC2 7 70 0·015
8 80 0·005
hence equilibrium cant, cur vature and versine 9 90 0·000
are linearly related. For any given cur ve the
diagrams of equilibrium cant, cur vature and
versine all have the same shape. Fig. 5 shows
the equilibrium cant, cur vature and versine
diagrams for a circular cur ve with clothoidal
transitions. V
C /2 C /2

Transition cur ves

10. Transition cur ves are added between
tangent track and circular cur ves to avoid sud-
den changes in lateral acceleration (lateral R–V
shock). The most common form of transition R
cur ve is the clothoidal spiral. The defining
property of a clothoidal transition cur ve is that
its rate of change of cur vature is constant. Fig. 4. Relationship
Thus, the principal advantage of the clothoidal between versine and
transition is that its rates of change of cant chord length
235

Downloaded by [ TUFTS UNIVERSITY] on [26/09/16]. Copyright © ICE Publishing, all rights reserved.
 O’DWYER

and cant deficiency are linear. Hence clothoi-

dal transitions are easy to cant.
11. Versines and superelevation can be cal-
culated for more elaborate forms of transition
cur ve; however, the rate at which the cant is
applied in such cur ves is nonlinear. For similar
transition lengths more complicated, S-shaped, (a)
transition cur ves reduce the maximum rate of
change of shock but have higher maximum
shock rates than simple clothoids,7 see Fig. 6.
Such complication is unnecessar y for all but
ver y high-speed lines.
(b)
Fig. 5. Diagrams for:
(a) versine; (b)
Realignment equilibrium cant; and
12. Realigning a cur ve involves altering its (c) curvature, for a
local cur vature and hence its versine readings. circular curve with
This is achieved by displacing the track lat- (c) clothoidal transitions
erally; that is, the track is pushed in ‘here’ and
pulled out ‘there’ in order to obtain the
desired cur ve shape. The difficulty lies in cal-
culating the magnitude and direction, inwards
or outwards, of the slew (lateral displacement)
necessar y at each station along the cur ve to
achieve the required shape. The traditional
methods for choosing the required slews are
based on the relationship between lateral dis-
placements and the corresponding changes in
versine. Knight expressed this relationship as Fig. 6. Curvature
follows6 diagrams for clothoidal
If a joint (or station) of a cur ve is shifted outwards and S-shaped
( 1 ) or inwards ( 2 ) a certain distance (the transitions
throw), the versine at that station will be altered
by an amount equal to the throw, and in the same
direction, while the versines at the stations on
either side will each be altered by an amount
equal to half the throw, and in the opposite
direction.
du
13. The changes in the versines resulting #kNew(s)ds 5 #kOld(s)ds 5 #Start
End
ds
from a series of slews can be expressed math- ds
ematically as 5 uEnd 2 uStart (6)
Versine and curvature are linearly related,
1 1
V Old
i 2 Pi 2 1 1 P i 2 Pi 1 1 5 V New
i (5) see equation (2), hence equation (6) is sat-
2 2 isfied if the areas under the existing versine
diagram and the proposed versine diagram
14. Traditional methods are based on the are the same. This is achieved if the sum of
designer choosing a set of desirable versines the proposed versines equals the sum of the
and then tr ying to find a series of slews which existing versines, thus
come close to achieving them.5, 6 However, any
OV OV
n n
proposed set of new versines must satisfy two New
5 Old
(7)
i i
constraints, as follows. i50 i50

(a) The angle subtended by the cur ve must (b) The track at the end of the cur ve must not
remain the same, that is the track at the be displaced laterally relative to the tan-
end of the cur ve must be tangential to the gent track, that is the end of the cur ve
second straight. This constraint is satis- must line up with the second straight.
fied if the areas under the cur vature dia- This is equivalent to requiring that the rel-
grams for the existing and realigned ative displacement of the existing and the
cur ves are the same. The area under both proposed cur ves should be zero at their
cur ves must equal the change in angle common intersection with the second
between the two tangents straight. Mathematically the condition is
236

Downloaded by [ TUFTS UNIVERSITY] on [26/09/16]. Copyright © ICE Publishing, all rights reserved.
 REALIGNMENT OF
RAILWAY CURVES

duNew Therefore only n 2 1 of the n 1 1 new versines

#(L 2 s)(uNew(s) 2 uOld(s))ds 5 #(L 2 s) ds are independent or, equivalently, the n 1 1 new
ds versines must satisfy two additional con-
duOld straints. If the values of the new versines are
2 ##(L 2 s) ds 5 0 (8) picked at random there will not be a solution.
ds 17. The first constraint can be found by
summing the left-hand and right-hand sides of
In terms of versine readings this condition
equations (10(0))—(10(n)). All the P terms on
is satisfied if
the left-hand side cancel and equation (7)

O (V
n results.
New
i 2 V Old
i )(n 2 i) 5 0 (9) 18. The second constraint is obtained by
i50
multiplying both sides of equations
(10(0)—(10(n)) by i and then summing both
Equation (9) is equivalent to requiring
sides. The P terms cancel and equation (9) is
that the centroids of the old and new cur-
obtained.
vature diagrams, and hence the centroids
of the versine diagrams, coincide. Equa-
tions (7) and (9) assume that all the ver-
Circular cur ve with clothoidal
sine readings are for the same chord
transitions
length.
Consider realigning a standard cur ve compris-
15. These conditions, equations (7) and ing a circular section plus two clothoidal tran-
(9), are satisfied by any combination of real sitions. Sur veying the cur ve by measuring the
lateral displacements, that is if the rail is not existing versines gives the area under the
cut then it must still turn through the same existing versine cur ve and locates the existing
angle and cannot be displaced relative to the versine diagram’s centroid. Fig. 7 shows the
second tangent. Any set of slews will generate shape of versine diagram for a perfectly
a cur ve which satisfies the two conditions but aligned circular cur ve with clothoidal transi-
not all sets of design versines have a valid tions. The shape of the versine diagram, and
solution. Only sets of versines which satisfy hence the value of the versine at any point
the two constraints have real solutions. This along the cur ve, is fully described by
fact was disguised by the traditional methods specifying
of tabulating and expressing the problem.5
(a) C1 the chainage at which the first transi-
16. The constraint equations (7) and (9)
tion starts
can be generated directly from the relation-
(b) C2 the chainage at which the first transi-
ship between versines and slews given by
tion ends and the circular cur ve starts
equation (6). At each of the n 1 1 stations
(c) C3 the chainage at which the circular
(points on the cur ve at which a versine is
cur ve ends and the second transition
measured) equation (6) is satisfied, therefore
begins
this gives a series of equations
(d) C4 the chainage at which the second tran-
1 sition ends
0 2 Pl
V old 5 V New
0 (10(0)) (e) Vmax the maximum versine.
2
20. The cur vature diagram can also be
1 defined using different variables such as the
V old
1 Pl 2 P2 5V New
(10(1))
2 transition lengths, the length of the circular
section and the rates of change of cur vature in
: :
the transitions. In all cases five variables are
1 1 required to define the versine diagram. The
i 2 Pi 2 l 1 Pi 2 Pi 1 l
V old 5 V New
i (10(i))
2 2 versine at any point along the cur ve is a func-
tion of these five variables. These variables
: :
cannot be chosen arbitrarily: the variables
1 must be chosen such that the resulting ver-
V nold2 l 1 Pn 2 l 2 Pn 2 2 5 V nNew
2l (10(n 2 1))
2 sines satisfy the two constraints (7) and (9).

1
n 2 Pn 2 l
V old 5 V New
n (10(n))
2

The first and last stations must have zero

slews (P0, Pn 5 0) since any lateral displace-
ment of these stations would affect track out- Vmax
Fig. 7. Versine
side the section of track to be aligned. If the diagram for a circular
new versine values are specified then there curve with two
are n 1 1 equations but only n 2 1 variables. C1 C2 C3 C4 clothoidal transitions
237

Downloaded by [ TUFTS UNIVERSITY] on [26/09/16]. Copyright © ICE Publishing, all rights reserved.
 O’DWYER

Given five variables and two constraints, only

three of the variables are independent. There-
fore specifying any set of three variables
defines the cur ve uniquely. One important
obser vation which results is that since there
are only three independent variables it is not B
possible to specify the start and end points of A
both transitions and still obtain a cur ve which
satisfies the constraint equations. The stan- a na Fig. 8. Versine
(n<1) corrections
dard attempts to define the start and end
points of both transitions and the radius of the
circular cur ve and then calculate the slews are Simultaneous linear equations
bound to produce inexact solutions. 23. Once design versines, which satisfy
21. While the designer can choose the val- equations (7) and (9), have been selected and
ues of the three independent variables for this corrected, as per equations (11)—(14), finding
form of cur ve, other considerations may dic- the slews required to achieve these versines is
tate the values which some of these variables elementar y. Equations (10(0))—(10(n)) give
must take. The sum of the maximum allowable the relationship between the existing versines,
cant plus maximum allowable cant deficiency the design versines and the slews. The unique
will limit the maximum permissible versine. values for the n 2 1 non-zero slews can be
Limits on the allowable rate of gain of cant will obtained by using Gaussian elimination8 to
impose limits on the minimum transition solve any n 2 1 of the n 1 1 equations. Alter-
lengths, and switches and crossings or plat- natively the slews at the first and last stations,
forms may dictate where the cur ve begins or P0 and Pn, can be included and all n 1 1 equa-
ends. If more complex cur ves are allowed then tions solved. If the versine values satisfy the
the number of independent variables and two constraints then the solution should yield
range of choice increases but so does the P0 and Pn equal to zero. This procedure offers
complexity. a check on the validity of the chosen versine
values.

Versine correction
22. At most stations on a cur ve the meas- Optimization approaches
ured versines and theoretical versines agree; 24. It is usually desirable to find a solution
however, the versines measured at the stations which is not only acceptable but which is also
at either side of the start or end of a transition optimal in some sense. Different cur ves will
cur ve need to be adjusted, see Fig. 8. These require different types of optimum solution.
adjustments do not affect equation (7) and The variety of possible scenarios makes sug-
have a ver y small effect on equation (9). The gesting a single optimum criterion question-
adjustment required is a function of the chord able.9 For example, guidance which suggests
length used to measure versine and the posi- that the sum of the slews should be equal to
tion of the tangent point relative to the sur vey zero9 should be used with caution. Consider a
stations. By approximating the clothoidal section of tangent track where a single sur vey
cur ve with a cubic parabola,5 the versines at station is displaced, see Fig. 9. Realigning this
the stations before and after the transition track by moving the displaced station to its
point are calculated as original position would correct the alignment
but the resultant sum of the slews will not be
n3a3 n3DV zero.
VA 5 5 (11) 25. If the designer can formulate an objec-
12RL 6
tive function which quantifies the relative fit-
ness of various valid solutions then the optimal

F G DV6
solution can be found. Although the number of
VB 5 (n 1 1)3 2 2n3 (12) unknown slews may be large, the number of
independent variables required to define a set
A similar adjustment must be made where the of design versines is small. The value of the
transition and circular cur ves meet. objective function for any cur ve is some func-
tion of the independent variables. If the objec-
n 32DV tive function is continuous then gradient
VC 5 Vmax 2 (13)
6

Fig. 9. Tangent track

F
VD 5 Vmax 2 (n2 1 1)3 2 2n32 G6 DV
(14) with one misaligned
station
238

Downloaded by [ TUFTS UNIVERSITY] on [26/09/16]. Copyright © ICE Publishing, all rights reserved.
 REALIGNMENT OF
RAILWAY CURVES

methods may be applicable. If the objective

function includes certain types of penalty func-
tion (e.g. a large penalty might be imposed on
a solution if the realignment required the pur-
chase of additional land), then the objective
function may not be continuous. If the objec-
tive function is not continuous then the use of
genetic algorithms10 or, if the search space is
small, a grid search, may be practicable.

Linear programming formulation minimize the sum of the slews necessar y to Fig. 10. Track section
26. If the realignment problem is subject to achieve a given line speed. The objective func- in which the allowable
a large number of constraints where a non- tion for this formulation is the sum of the slews are constrained
standard cur ve is unavoidable then linear pro- absolute values of the slews at each station by structures, cuttings,
gramming11 can be used to optimize the cur ve. embankments and
O )P )
n
The procedures for this form of design are, if (18)
property boundaries
i
anything, easier than the standard approach. ii 5 01
However, the form of the solution is so unre-
stricted that the cur ve which is generated 30. The problem can also be formulated to
must be vetted by an experienced track engi- minimize the largest single slew. This requires
neer to ensure that it is both suitable and safe. adding an additional set of constraints of the
27. Linear programming involves finding form.
the values of an often large number of varia-
2 Pmax # Pi # Pmax (19)
bles which maximize or minimize a linear
objective function subject to a set of linear while the objective function becomes minimiz-
constraints. Standard algorithms exist for find- ing the value of Pmax
ing the optimum values where they exist and 31. Unfortunately, finding the maximum
modern programs can deal with tens of thou- allowable line speed given an existing cur ve
sands of constraints.12 and a set of limits on slews at each station is a
28. Consider the fundamental equations nonlinear problem. However, it can be tackled
(10(0))—(10(n)) which give the relationship in stages by optimizing for increasing values of
between versines and slews. Linear program- velocity until no acceptable solution can be
ming can be used to find the optimum values achieved.
of the design versines and slews simultane-
ously. However, equations (10(0))—(10(n))
place no restriction on the shape of the result- Conclusions
ing cur ve and, therefore, additional constraints 32. This paper has presented the problem
must be added into the problem formulation. of realigning railway cur ves, and the tradi-
The maximum cur vature at any station can be tional realignment methods, in a mathematical
limited by adding constraints of the form format. Tradition cur ve realignment proce-
dures which attempted to find the slews
g(Emax 1 Dmax) g(Emax 1 Dmax) required to produce a specified set of versines
2 2
# Vi # (15)
Grv Grv2 cannot yield a perfect solution unless the pro-
posed versines satisfy two conditions. First,
The rate of change of cur vature can also be the sum of the new versines must equal the
limited. The constraints required in this case sum of the existing versines. Second, the cen-
will have the form troid of the new versine diagram must occur at
2 DVmax # (Vi 2 Vi 2 1) # DVmax (16) the same location as the centroid of the exist-
Additional constraints can be added on the val- ing versine diagram.
ues of the slews which can be applied at each 33. The versines of a standard cur ve com-
station, that is posed of two clothoidal transitions and a circu-
lar cur ve can be defined in terms of five
2 Pmini # Pi # Pmaxi (17) variables. Thus, in realigning a standard
This form of constraint demonstrates the cur ve, since there are two constraints and five
advantage of the linear programming formula- variables, only three variables are
tion. The limits on the slew values can repre- independent.
sent the edges of the track bed or the 34. The slews necessar y to achieve any set
distances to tunnel walls, bridge abutments or of design versines can be found by solving the
platforms, see Fig. 10. simultaneous linear equations
29. The objective function can be formu- (10(0))—(10(n)).
lated to solve a number of different types of 35. The optimum solution to any realign-
problem. The problem can be formulated to ment problem can be obtained by formulating
239

Downloaded by [ TUFTS UNIVERSITY] on [26/09/16]. Copyright © ICE Publishing, all rights reserved.
 O’DWYER

an objective function and using standard 4. MARRIOTT D .C. Improving the performance of
search techniques to find the values of the track lining and levelling machines. Proc. Instn
independent variables. Civ Engrs Transport, 1996, 117, Aug., 177–185.
5. British Railway Track. The Permanent Way Insti-
36. For particularly constrained problems,
tution, Echo Press, Loughborough, 1993, 6th
linear programming is a ver y powerful tech- edn.
nique. However, the solutions generated by 6. KNIGHT P. E. String-lining of railway cur ves. The
the linear programming approach require care- Railway Gazette, London, 1938.
ful vetting. The cur ves generated are non- 7. GUBAR J. Railway transition cur ve planning meth-
standard and the rate of change of cant may ods. Rail International, 1990, April, 31–43.
8. KREYSZIG E. Advanced Engineering Mathematics.
be nonlinear.
Wiley, 1983, 5th edn.
9. SESHAGIRI RAO M. Toptrack cur ve realignment
program. The Permanent Way Institution Journal,
1993, 111, Part 3, 220–225.
References 10. GOLDBERG D. E. Genetic Algorithms in Search
1. ESVELD C. Modern Railway Track. MRT produc- Optimisation and Machine Learning. Addison
tions, West Germany, 1989. Wesley, New York, 1989.
2. PROFILLIDIS V. A. Railway Engineering. Avebur y 11. TAHA H. A. Operations Research. Macmillan,
Technical, Ashgate Publishing, Aldershot, 1995. 1987, 4th edn.
3. FASTENRATH F. Railroad Track: Theory and Prac- 12. SCHRAGE. L. LINDO — An Optimization Modeling
tice. Frederick Ungar, New York, 1981. System. Boyd & Fraser, USA, 1991, 4th edn.

240

Downloaded by [ TUFTS UNIVERSITY] on [26/09/16]. Copyright © ICE Publishing, all rights reserved.