6th World Congress of Structural and Multidisciplinary Optimization

30 May – 03 June 2005, Rio de Janeiro, Brazil

Local Shape Optimization for Airframe Life Extension

M. Heller, R. Braemar, W. Waldman, R. Evans, M. McDonald and R. Kaye

Defence Science and Technology Organisation, Melbourne, Australia

(manfred.heller@defence.gov.au)

1. Abstract
Life extension repairs of ageing metallic airframe components can provide significant economic benefits by avoiding the need for
component replacement, by potentially increasing the interval between periodic in-service inspections. Here we consider the benefits
of using precise optimal rework shapes, which minimize the peak local stresses as compared to initial non-optimal stress
concentrators, such as circular holes. The free-form optimal shapes are obtained using an iterative gradientless finite-element method,
which is based on an analogy with biological growth. This work is undertaken in an in situ reworking context, where only material
removal is allowed, and where typically some locally ‘damaged’ material is also removed. Initially, a review is given of important
lessons learned from the application of optimal rework shapes to locations in the F-111 aircraft. Enhancements to the automated 2D
and 3D local shape optimization method are then presented, including robustness and multi-peak analysis for perturbed or multiple
load conditions. Finally, numerical examples are provided to demonstrate the approach and to quantify the effect of the trade-off
between key parameters, such as manufacturing constraints, aspect ratio and crack detection thresholds, on fatigue-related predictions
for the optimal shapes. It is shown that large increases in fatigue lives and inspection intervals will typically be achieved. However,
under certain circumstances, stress-based shape optimization may yield negligible increases in inspection intervals.

2. Keywords: Shape optimization, gradientless method, fatigue life, airframes, stress concentration

3. Introduction
Typically during the working life of an airframe, a few key stress-concentrating locations can become fatigue critical, and an
effective repair option is needed. An appropriate repair can provide significant economic benefits, by avoiding the need for
component replacement, and/or typically increasing the interval between costly periodic in-service inspections. Also, in some cases, a
repair is the only option because component replacement is not viable. The repair approach adopted here is to determine precise
optimal rework shapes that minimize the peak local stresses, while at the same time removing the crack region at the initial non-
optimal stress concentrator (e.g. an initial circular hole).
The general approach adopted is an iterative gradientless finite element analysis, based on an analogy with biological growth.
A key aspect is that the optimal shapes are free-form (subject to any prevailing geometric constraints), in the sense that they are not
defined by analytic functions. Apart from minimizing peak stress, the design requirements for optimal rework shapes can be
significantly different to the case where a new component is being designed (which is more often addressed in the literature). For
example, in the rework context: (i) the optimal hole must be larger than the existing hole; (ii) different manufacturing constraints
operate, because the rework must typically be done in situ; (iii) the shapes need to be robust to errors in load direction uncertainty
and/or manufacturing misalignment, or geometric variability in the aircraft fleet; (iv) inspection issues may need to be taken into
account; and (v) potential weight saving is usually irrelevant. To date, the authors and their colleagues have been developing a
capability in local airframe shape optimization, and have considered and solved a number of generic and practical problems [1–6].
Further applications are currently being undertaken, and improvements are being made to the analysis capability, with a focus on
addressing practical application issues, which affect how the optimization process should be undertaken.
Hence, in the present paper the following work is presented. Initially, a review of important lessons learned from practical
applications of optimal rework shapes for the life extension of the wing pivot fitting (WPF) in the General Dynamics F-111 is given.
Secondly, new enhancements for a fully automated 2D and 3D local shape optimization method are given. Here we minimize the
magnitude of the multiple, constant-stress segments around the hole boundary, which typically consist of tensile and compressive
stress states. We now include robustness constraints, so that the peak stresses are independent of perturbations in the dominant
loading direction (within a prescribed range), and/or are optimal for distinct load cases. Thirdly, numerical examples are given to
demonstrate the usefulness of the approach, and to quantifying the effect of the trade-off between peak stress minimization, hole size,
manufacturing constraints, robustness constraints, non-destructive inspection limits, and certification basis. The numerical examples
focus on geometrically constrained holes, which are typical of airframe structures.

4. Previous F-111 practical application – lessons learned

4.1 Background
The critical stress concentrating features in the WPF are managed via safety-by-inspection due to the possibility of manufacture-
induced flaws in the high strength, but flaw intolerant, steel which is used. This has resulted in a significant ongoing maintenance
burden due to the regular inspections that are required. The key to easing the maintenance costs is through reducing the frequency of
the inspections, i.e. by increasing the inspection interval. Shape optimization via material removal only was selected as the repair
approach, based on ease of implementation, lower long-term maintenance costs, lowest risk, and the ability to increase flight
envelopes [6]. The finite element model used to do the final development of the optimal shapes (semi-automated) is shown in
Figure 1. For each wing, four optimal stiffener runouts (SROs) were determined, along with four optimal fuel flow vent holes
(FFVHs), with the locations shown in Figure 1b and 1c, respectively. As a sample, an optimal hole shape for FFVH #13 is shown in
Figure 2a as compared to blueprint and traditional rework geometries, providing peak stress reductions at the critical location of 53%
and 38%, respectively.

4.2 Positive lessons learned

Large stress reductions were predicted by using the optimized shapes, where only material removal was allowed. The stress
reductions have been demonstrated by full-scale static strain surveys, and fatigue tests have shown that the optimal shapes are highly
durable and can tolerate significant corrosion and mechanical damage. Overall, the inspection interval is demonstrated to be doubled.
A complex but robust manufacturing process was devised which involved an electrical discharge machining process to achieve: (i)
material removal with minimal heat affected zone; (ii) in situ measurement of the WPF geometry to account for individual
dimensional variations between WPFs; and (iii) a custom-built support arm to position the probes and machining instruments deep
inside the structure via restrictive access holes. To date a total of 96 stress concentrating features have been reworked with only a few
requiring additional engineering assessment.

4.3 Technical issues and difficulties

Some positional tolerances could not be met due to larger-than-expected variations in the geometry between various WPFs. A more
comprehensive survey of the variations in WPF geometry in the fleet may have lead to a more robust and flexible manufacturing
process, however this would have been difficult to achieve without affecting the availability of aircraft. Even though the optimal
shapes were designed to be robust to deviations, and that a relatively flexible manufacturing process was devised, (e.g. in situ
measurements), it is evident that some additional effort to further increase these aspects may have mitigated some of the additional
engineering assessments that were required. While the optimal shapes were developed to minimize the amount of material removed,
a number of size-related issues remained, including: (i) the electrical discharge process was time consuming, with a single rework
taking around 8 hours to machine; and (ii) due to the magnetization process, it was difficult to use the preferred non-destructive
inspection (NDI) method (magnetic rubber inspection) if the optimal shape was large. Hence, a greater understanding of the
interaction of NDI limits, hole size and inspection intervals is important, and this is the focus of this paper.

FFVH 13
SRO 5
SRO 4 FFVH 14

SRO 3

FFVH 12

FFVH 11
SRO 2

(a) (b) (c)

Figure 1. F-111 wing pivot fitting 3D finite element model showing: (a) a section view of the upper plate region, (b) detailed view of
stiffener runout features selected for optimization, and (c) FFVH features selected for optimization F-111 WPF upper plate [6].

20
120° 90° 60°
30°
150°
Vertical position, y (mm)

10

180°
0 0°

-10
330°
210°
300°
-20 Up 270° blueprint geometry
240° traditional rework
Inboard optimal rework
-30
-30 -20 -10 0 10 20 30 40

Horizontal position, x (mm)

(a) (b) (c)

Figure 2. Typical optimal reworks for F-111 WPF: (a) hole shapes, (b) manufactured holes, and (c) manufactured stiffener runout.

5. Shape optimization method

The simple stress optimization approach used here for subsequent further development is an iterative gradientless approach, yielding
free-form shapes (fully automated). It is used here since it is simple to implement, readily amenable to extension, and has been shown
to achieve very good solutions for stress minimization for realistic geometrically constrained problems, which is often not a focus of
other investigators. Our approach is based on the aim of achieving a minimum peak stress by directly attempting to make the stresses
uniform along the longest segment of a stress-concentrating boundary. This idea is well known, and has been successfully used in
various analytical [7, 8, 9], numerical [1, 5, 10, 11], and experimental [12] studies. The different numerical algorithms have various
advantages and disadvantages in their implementation. The present approach is to add material to local regions of high stress and/or
remove material from local regions of low stress on the stress concentrator boundary. Our prior work has typically focused on
significantly minimizing a single dominant stress peak (tensile or compressive) on boundary segments of stress concentrators [1, 3,
5]. More recently we have extended the approach to multiple peaks [13, 14], since the most fatigue critical location on a hole
boundary may not be the one with the biggest active tensile peak. For example, different boundary segments can have different
residual stresses or probabilities of detection of cracks.

5.1 Multi-peak method

Consider a hole in a remotely loaded large 2D plate, as shown in Figure 3a, where 0 ≤ µ ≤ 1 is the non-dimensional arc length around
the hole boundary. Here µ = u/U, where U is the total arc length around the hole boundary. There are k nodes distributed around the
hole boundary, and the local tangential stress at the i-th node is σi. The optimization is subject to a minimum radius of curvature
constraint ρ ≥ ρmin, and an optional geometric bounding constraint consisting of a polygon with m sides. The former is activated to
ensure that sharp corners, which are commonly found in ideal optimal shapes [2, 9], do not develop. The typical multi-peak
distribution of tangential stress around the boundary of a circular hole in a uniaxially loaded plate is shown in Figure 3b. Here there
are a total of four distinct stress peaks around the closed boundary (i.e. σp1, σp2, σp3 and σp4), which in general are not equal to one
another, together with four zero crossings. The zero crossings are used to identify the set of n subregions of positive and negative
stress, each subregion lying between a consecutive pair of zero crossings. Consistent with the work of Vigdergauz and Cherkayev
[9], as well as prior unautomated multi-peak shape optimization work by some of the authors [6], we postulate that a constrained
optimal shape will have n zones of peak uniform stress. For some idealized problems there will be only one peak stress region (e.g. a
biaxially loaded plate with an optimal elliptical hole). During each iteration, the stresses are computed at all the nodes around the
hole boundary. Then, the i-th boundary node, which is located in the q-th stress subregion, is moved in the direction of the local
outward normal to the boundary by a distance d iq , given by:
⎛ σ iq − σ thq ⎞ , where
d iq = ⎜ ⎟ sc σ thq = max σ iq , q = 1, 2, …, n (1)
⎜ σ th ⎟
q
⎝ ⎠
where σ iq is the tangential stress at node i in the q-th stress subregion, σ thq is the stress threshold corresponding to the peak positive
or negative stress occurring in the q-th stress subregion (updated each iteration), s is a constant step-size scaling factor (e.g. 0.04),
and c is a characteristic length. Hence, at each iteration we are only removing material from low stress regions. The amount of
removal at each location is linearly proportional to the difference compared to σ thq . This process is repeated until the stresses in each
subregion converge to a reduced constant value, within some prescribed tolerance, over the maximum possible length of arc. The
quality of the solution is assessed at each iteration by computing the length of each of these zones, µ q , from which
n
µT = ∑ µ q ( µT ≤ 1) (2)
q =1

can be determined. Here µT represents the normalized total combined length of the constant stress zones around the entire boundary
of the hole. When the optimal solution has been obtained, it is found that µT is a maximum for the problem under consideration. The
method is well proven to converge to known ideal solutions, as well as more complex constrained cases [14]. To avoid local mesh
distortion, after each iteration, while the node movements define a new shape, the nodes are moved tangentially around the boundary
to maintain the initial relative circumferential node spacing. Furthermore, radially away from the hole boundary, the relative spacing
of elements is maintained. All analyses were linear elastic, with Young’s modulus E = 71 GPa, and Poisson’s ratio v = 0.33. For 2D
cases eight-noded quadrilateral elements were used, while eight-noded brick elements were used for 3D cases.
4
σp1 Initial shape
σp3 σp1
Optimal shape
PLATE Boundary of hole 3
µ1 µ3 µ1
Tangential stress, σ

Γ Nodes
2
σ*p3
ni σ*p1 σ*p1
Zero
Hoop stress σi 1 crossings
di
ti
ρ u 0
i+1 σ*p2 σ*p4
i
i–1
k-1 k 1 2 -1
µ2 µ4
Polygonal constraint
σp2 σp4
boundary (optional)
-2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
µ
(a) (b)
Figure 3. Description of multi-peak method: (a) notation for geometrically constrained shape optimization of a hole in a remotely
loaded plate, and (b) typical idealized multi-peak stress distribution around initial circular hole and corresponding optimized hole.

5.2. Extension to in-built robustness

The aim here is to determine a notch shape that renders the peak stress constant and minimal for a range of load orientations, along a
boundary segment. It is useful to initially explain this for only one subregion for the 2D case. Consider Figure 4a, which shows an
arbitrary hole in a plate, subjected to a remote uniaxial load that can be applied over a range of angular orientations. The angular
orientation, α, of that load is discretised into an appropriate number of individual load cases l = 1...m, where l is the load case
identifier and m is the total number of load cases. For an arbitrary initial hole shape, the stress distribution σl,i around the hole, as well
as the magnitude and position of the peak stress, will usually be different for each load angle, as illustrated in Figure 4b. The locus of
the maximum peak stress distribution σ imax, q = max σ lq=1Km,i for all load cases is also shown in Figure 4b. To render the peaks minimal
and constant, we can calculate the nodal movements in a similar fashion to the basic method in Eq.(1), except here the stress term σi
is replaced with the locus of the maximum stress term σ imax, q . Hence, for the q-th subregion, at each iteration the boundary node
movements for the robust stress method are given by:
⎛ σ max, q − σ q ⎞
d iq = ⎜⎜ i q th ⎟⎟ sc where σ th = max( σ i ) , q = 1, 2, …, n (3)
q max, q

⎝ σ th ⎠
This approach was previously successfully demonstrated via 2D numerical examples for only one subregion [6]. We have now
extended it to quasi-3D cases involving multiple subregions. At any angular position i around the boundary, the maximum nodal
stress σ imax, q, j = max σ lq=1Km , j =1Kv ,i through the thickness, j = 1…v, is used as the basis for the optimization calculation. Hence, the
nodal movements for each node through the plate thickness, at a given angular position i, are given by:
⎛ σ max, q, j − σ thq , j ⎞ where σ thq , j = max( σ imax,q , j ) , q = 1, 2, …, n and j = 1, 2, …, v (4)
d iq , j = ⎜⎜ i q, j ⎟⎟ sc
⎝ σ th ⎠

Remote
max(σ[l = 1...m]) peak
uniaxial

Tangential stress
stress distribution
load
m 1
2
l
Initial arbitrary
hole shape for l
α σ[1]
2 optimization
σ[2]
1 m σ[l]
σ[m]
Position along hole boundary
(a) (b)
Figure 4. Description of robust stress optimization, for one subregion: (a) plate with remote uniaxial load of varying orientation α;
(b) typical stress distributions about part of an initial arbitrary hole shape for different load condition cases.

6. Fatigue analysis method

Airframe components are usually certified and managed on either a ‘safe-life’ or ‘safety-by-inspection’ basis [15]. Key aspects as
they relate to the estimation of fatigue lives in the present work are described briefly below.

6.1 Safe-life approach

Under a safe-life approach, the total life (i.e. crack nucleation plus crack growth life to failure) of a component is estimated and then
scaled by an appropriate safety factor in order to provide a safe estimate of the service life. After the safe life is reached, the
component is automatically withdrawn from service. The result of calculating the total life of the crack nucleation process under the
safe-life approach is the stress-life curve, which is shown schematically in Figure 5a. This curve can be derived from basic coupon
test data of constant amplitude cycles to failure, or more directly by coupon testing using the particular variable-amplitude flight
spectrum for the location. The life estimation approach can be based either on local stress (through the use of Kt, the elastic geometric
stress concentration factor), or local strain (using Neuber’s notch factor Kn), as the input spectrum. A simple empirical expression for
rewriting the stress-life relationship under a flight spectrum can be given as:
⎡ K −1 ⎤ (5)
ln(σK n ) = Α ln( t t ) + Β where K n = 1 + ⎢ t ⎥
⎣⎢ 1 + a
⎥
m ⎦

Here A and B are derived constants that are material and flight-spectrum dependent, σ is a reference remote stress level for the
variable-amplitude spectrum of interest, tt is the total life, m is the radius of the notch root, and a is a characteristic dimension of the
material. Hence, shape optimization is used to increase the total life by minimizing the geometric concentration factor Kt and hence
Kn. This is shown schematically in Figure 5a, where the decrease in Kn from Kn1 to Kn2 results in an increase in total fatigue life from
tt1 to tt2. In the present paper, the stress life plot is used in a manner consistent with the approach adopted by the authors’ colleagues
in the AP-3C Aircraft Fatigue Assessment Group [16]. For the numerical examples 1 and 2 (see Sections 7 and 8, respectively), we
use Figure 6b, and also simplify by taking Kn = Kt, since we are only concerned with lives relative to the initial circular hole, and
since the ratio of Kn/Kt will change little with the changes to Kt due to the optimization process. It is noted that, instead of minimizing
Kt, it is also possible to make further life improvements by minimizing the accumulation of non-linear damage at each point on the
boundary [17]. However, this is not the focus of the present work.

6.2 Safety-by-inspection approach

In contrast to the safe-life-approach, the safety-by-inspection approach treats fatigue as growing cracks in progress, and it relies on
inspections to demonstrate that the initial size of assumed existing cracks (often chosen to match the inspection threshold) are small
enough to achieve a further period of safe operation. After that period has expired, the location is inspected again to assess if it can
remain in service for another period, and so on. Many aircraft operators use this approach for ageing aircraft, sometimes because the
airframes are being kept in service past their respective safe-life design, but also because of limitations with the safe-life approach in
that it does not explicitly consider in-service mechanical damage or corrosion – which can be highly relevant for ageing aircraft. The
basis for calculating the residual fatigue life under a safety-by-inspection approach is the crack growth curve as shown in Figure 5b.
These curves can be estimated from crack growth programs or derived from coupon tests using the particular variable-amplitude
flight spectrum for the location. In the present work, the crack growth program FASTRAN [18] was used, hence the crack growth
rate is expressed in terms of the effective stress intensity factor for a cracked plate, so that:
da / dN = C1 ( ∆Keff ) n (6)
where: (i) C1 and n are experimentally determined constants; and (ii) Keff is the effective stress intensity factor, which depends on the
maximum stress intensity factor in the current load cycle Kmax, and its stress ratio, to represent the existence of a physical threshold
stiffness intensity factor range below which the crack does not grow. The maximum stress intensity factor Kmax may be calculated as:

K max = σ max β π a (7)

where β depends on the notch geometry and surrounding geometry, and is a function of crack length, and σmax is the maximum
remote stress in the current load cycle. For complex geometries β is determined by linear superposition of simpler cases [19]. For
the optimal case it is then scaled by the ratio of the stresses along the prospective crack path, for the optimal and circular cases, as the
value of a is varied. The assumed crack configuration is that of a semicircular edge crack emanating from the hole edge, at the
location of peak stress, in a 3D plate. Here the crack depth from the hole edge is a, and the surface length is c = 2a. The crack growth
curve can be calculated by numerically integrating Eq.(6). The residual life, tr, is taken as the time in flight hours to grow the fatigue
crack from the NDI maximum undetectable size andi to a critical size ac, as shown in Figure 5b. Here ac is the maximum crack size
such that the structure has sufficient residual strength that it can still withstand the required loading. In aircraft practice, a safety
factor (often equal to 2) is used to determine inspection from the total inspection period. Shape optimization is used here to increase
the residual life by minimizing the free-surface peak stress, and this primarily minimizes the magnitude of Kmax when the crack size
is small. Hence, the difference in Kmax between a non-optimal and an optimal notch is most significant when the crack is small. This
is highlighted in Figure 5b, which shows a relatively higher crack growth rate for smaller crack lengths for the non-optimal case
compared to the optimal case. A typical value is andi = 1.27 mm. In practice, the magnitude of andi varies, depending on the particular
NDI technique, the accessibility to the location, the experience of the operator, the material, and importantly the shape and size of the
notch feature. For all inspection intervals calculations in the present paper, the spectrum loads (see Figure 6a) are scaled so that the
peak stress on the edge of the circular holes matches that for the practical example in Section 9. This same loading is applied to the
plate with the optimal shapes.

tt2 tr2
tt1 tr1
ac
Crack size, a

σKn1 Non-optimal
σKn

Non-optimal

Optimal Optimal
σKn2 andi

Flight time, t (hours) Flight time, t (hours)

(a) (b)
Figure 5: Schematic showing the relationship between reducing the notch stress level and the life extension: (a) stress-life curve used
by a ‘safe-life’ approach, and (b) crack growth curve, starting from a = andi, used in the ‘safety-by-inspection’ approach.

10 7 7.5
Exceedences per 15000 hours

7.0
10 6
6.5
10 5 6.0
10 4 5.5
5.0
Kn

10 3 4.5
10 2 4.0
3.5
10 1 3.0
1 2.5 4
-10 -5 0 5 10 15 20 25 10 10 5 10 6
Stress (ksi) Simulated flight hours
(a) (b)
Figure 6. Fatigue analysis for AP-3C load spectrum: (a) spectrum stress exceedences, and (b) Kn versus simulated flight hours.

7. Numerical example 1 – stress robust hole in 3D loaded plate

Consider a large circular plate with radius Rb = 5h containing a geometrically constrained hole of aspect ratio h/w = 2 (see Figures 7a
and 7b, respectively). The remote uniaxial stress, σnom, is applied via distributed forces [6], as shown in Figure 7a for the nominal
case, where the inclination angle is α = 0°. It is considered that in practice this inclination angle may vary over a continuous range,
and this effect is studied here. The starting hole shape for optimization was an ellipse, and the half plate thickness was 0.25w, as
shown in Figure 7b. A minimum radius of curvature constraint of ρmin/w = 0.3 was applied. The 3D model was initially optimized
with a standard (i.e. non-robust) analysis for α = 0°, and then with a robust analysis where the load was applied in 1° increments from
-5° to +5°. The profiles and stresses resulting from this analysis are shown in Figure 8. The non-robust model experiences an
increase in the peak Kt from 1.83 to 2.27 upon application of the α = 5° loading, while the robust case had a consistent peak Kt of
1.94 for all α in the design range -5º < α < 5º. The benefit of the robust optimization analysis is clearly evident.

y Remote boundary
Rb nodal forces,
equivalent to
a uniform j=v
uniaxial
θ stress :
σnom j=2
j=1
α
x
w
i−1
i i+1
h

(a) (b)
Figure 7. Robust hole analysis. (a) overall model with the remote stress applied at the nominal inclination angle of α = 0º; and
(b) local view around initial elliptical hole for half-plate-thickness 3D model.

0.3 2.5
2.0

0.2 1.5
1.0
y/h

Kt

Robust shape with α = 0°

θ 0.5
Robust shape with α = 5°
0.1
0.0 Non-robust shape with α = 0°
Robust shape (optimised for −5°≤ α ≤ +5°) Non-robust shape with α = 5°
Non-robust shape (optimised for α = 0° only) -0.5
0.0 -1.0
0.0 0.1 0.2 0.3 0.4 0 .5 0° 10° 20° 30° 40° 50° 60° 70° 80° 90°
x/h Angular position around hole, θ
(a) (b)
Figure 8. Robust stress minimization results for 3D case showing: (a) optimal hole shapes, and (b) stress distributions about the hole
boundaries for both robust and non-robust cases when the remote load angle is at the nominal orientation of α = 0º and α = 5º.

The estimates for relative lives as compared to the initial circular hole are summarized in Table 1, where the Kt values are given in
parentheses. It can be seen that, for both the standard and robust optimal shapes, there is a significant increase in life compared to the
initial circular or even the elliptical hole. The robust optimal offers a four-fold increase in life as compared to the circular hole. As
required, there is an advantage in using the robust optimal as compared to the non-robust optimal, for the perturbed load case, α = 5º.

Table 1. Relative safe lives, tt2/tt1, and (peak Kt) for nominal and robust 3D hole optimization, ρmin/w = 0.3.
Circular hole Elliptical hole Non-robust hole Robust hole
Load case relative life relative life relative life relative life
h/w = 2.0 h/w = 2.0 h/w = 2.0
α = 0º 1.0 (3.10) 3.79 (2.04) 5.36 (1.83) 4.45 (1.94)
α = 5º 1.0 (3.10) 3.79 (2.04) 2.70 (2.27) 4.45 (1.94)

8. Numerical example 2 – hole close to edge in uniaxially loaded 2D plate

Figure 9a shows a large rectangular plate subjected to uniaxial loading, S, containing a hole near one edge. A key aspect is that the
hole cannot grow beyond the bounding constraint line x = -w/2. The plate geometry is W/H = 0.6, the initial hole width is w/W = 1/6,
and the hole center is located at e = w. Stress results only for one h/w and one ρmin have previously been given in [13]. In the present
investigation, peak Kt and fatigue results for optimized holes are given for a range of h/w and ρmin, and are compared to an initial
circular hole. As a useful limiting case, standard optimization is done to investigate this effect. Figure 9b shows a typical comparison
of the shape of a nominal 2:1 elliptical hole (A) with that of the optimal hole (B), and Figure 9c shows the corresponding variation of
Kt around the hole boundary, as a function of arc length, µ. The initial hole has four major stress peaks, which from left to right are Kt
= 3.09, -1.51, 2.33 and -1.51. The optimal shape reduces these peaks to flat-topped regions that are uniform to within a tolerance of
0.1%, representing reductions in peak Kt for the corresponding ellipse of 24%, 29%, 11% and 29%, respectively.

S 1.6
3.5
Constraint line Initial 2:1 elliptical hole
1.2 3.0
e
2.5
y 0.8
H/2 2.0
A 1.5
0.4
1.0
y/w
h

Kt
0.0 0.5
x Optimal
0.0
u hole
w -0.4 -0.5
H/2 B -1.0
-0.8
W -1.5
-1.2 -2.0
-0.8 -0.4 0.0 0.4 0.8 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
S x/w Arc length µ around hole boundary
(a) (b) (c)
Figure 9. Optimization of constrained hole near edge in uniaxially loaded plate: (a) geometry, (b) typical comparison of initial
elliptical hole (A) with optimal hole (B) for 2:1 aspect ratio case, and (c) variation of Kt around boundary for initial and optimal hole.

The peak Kt results are summarized in Figure 10a. As h/w is increased, large stress reductions are demonstrated compared to the
initial circular hole, and also to the corresponding ellipses. It is also evident that, as h/w is increased, the effect of different ρmin
constraints diminishes; however, as expected the peak Kt is always higher for higher value of ρmin. Figure 10b gives stress decay plots
along the x-axis, for a typical case of ρmin/w = 0.3. The trend is such that, while stresses for optimal holes are reduced at the hole
edge, as the distance away from hole edge is increased the stresses for optimal shapes are greater than for the initial circular hole.

4.0 4.0
ρmin/w = 0.1
3.5 Circular hole
ρmin/w = 0.3
3.5 1:1 optimal hole
ρmin/w = 0.5 3.0 2:1 optimal hole
Ellipse 2.5
Kt

3.0
Kt

2.0
1.5
2.5 1.0
0.5
2.0 0.0
1.0 1.5 2.0 2.5 3.0 3.5 -1.0 -0.9 -0.8 -0.7 -0.6 -0.5
Hole aspect ratio h/w x/w
(a) (b)
Figure 10. Stresses for optimized constrained hole near edge in uniaxially loaded 2D plate, compared to ellipses: (a) effect of h/w and
ρmin, (b) sample stress decay plots for ρmin/w = 0.3, along x-axis on left-hand side of the hole.

The fatigue calculations are given for prospective cracking on the left-hand side of the hole (i.e. closest to the plate edge). The
estimates for relative lives as compared to the initial circular hole are summarized in Table 2, using Kn = Kt for all cases, where the Kt
values are given in parentheses. Even for the large constraint case, ρmin/w = 0.3 (which is desirable for ease of in situ manufacture),
there are very large increases in life, and the benefit increases with increasing aspect ratio. Here a 2:1 aspect ratio is sufficient to
provide a five-fold increase in life, while for ρmin/w = 0.1 a six-fold increase in life is attained.

Table 2. Comparison of relative safe lives, tt2/tt1, and (Kt) for optimized hole near edge of plate.
Minimum radius of Circular hole Optimal, h/w ≈ 1.0 Optimal, h/w ≈ 1.5 Optimal, h/w ≈ 2.0 Optimal, h/w ≈ 2.5
curvature constraint relative life relative life relative life relative life relative life
ρmin/w = 0.1 1.00 (3.99) 3.12 (2.79) 4.73 (2.45) 6.11 (2.26) 7.27 (2.14)
ρmin/w = 0.3 1.00 (3.99) 2.14 (3.14) 3.82 (2.62) 5.25 (2.37) 6.56 (2.21)

The predicted inspection interval results are summarized in Table 3 for a typical large-hole case with w = 100 mm, where they are
normalized with respect to the circular hole case with andi = 1 mm, and for a plate thickness of t=w/2. It can be seen that the relative
inspection interval increases with increasing h/w, for a given andi. As expected for a given h/w, the inspection interval decreases with
increasing andi. Useful trend plots are given in Figure 11a and 11b, which summarize the required h/w to achieve a doubling of
inspection interval as a function of andi. Here results for both a small hole (w = 25 mm) and a large hole (w = 100 mm) are given. As
expected, a greater h/w is needed for the small hole to achieve a doubling of inspection interval at the same value of andi.

Table 3. Relative inspection intervals, tr2/tr1 for optimized hole near edge, compared to circular hole with andi = 1 mm, ρmin/w = 0.3.
Inspection crack Circular hole Optimal, h/w=1.02 Optimal, h/w=1.52 Optimal, h/w=2.01 Optimal, h/w=2.51
threshold, andi (mm) relative interval relative interval relative interval relative interval relative interval
1 1.00 1.85 2.94 4.32 4.94
2 0.63 1.13 1.86 2.71 3.08
3 0.49 0.82 1.33 2.12 2.23
4 0.42 0.64 1.01 1.71 1.71
Approximate profile aspect ratio

Approximate profile aspect ratio

2.5 2.5
Design
Range Design
2.0 2.0
Range

1.5 1.5

1.0 1.0
1.0 1.5 2.0 2.5 3.0 3.5 4.0 1.0 1.5 2.0 2.5 3.0 3.5 4.0
a ndi (mm) a ndi (mm)
(a) (b)
Figure 11. Design range to double inspection interval for optimized hole close to edge in 2D plate: (a) small hole with w = 25 mm,
and (b) large hole with w =100 mm.

9. Numerical example 3 – fuel flow hole #15 in AP-3C lower wing skin stiffener
Fatigue life extension enhancements to maintain aircraft availability may be required for this and other circular holes in lower wing
skin stiffeners (integrally machined) of Lockheed Martin AP-3C Orion aircraft, in service with the RAAF. This location was
identified as a fatigue critical location during testing conducted by Lockheed Martin, Marietta, as failure of the test article resulted
from cracking at this hole [16]. In the present investigation, optimal rework hole shapes are determined as a potential life extension
option. Some preliminary peak stresses from shape optimization have been reported previously [20]. In the present work, we include
robust stress optimization, along with fatigue calculations, for various hole aspect ratios and crack detection thresholds.

9.1 Stress optimization analysis for AP-3C hole

Figure 12a and 12b indicates the local geometry and location of the hole in the stiffener. The dominant loading is uniaxial, applied in
the x direction at x = -81.7 mm, the left hand end was restrained at x = 81.7 mm. For optimization analysis the following constraints
applied: (i) only material removal; (ii) the lower hole bound was at y = 2.60 mm; (iii) the minimum radius of curvature was ρmin/w =
0.236, and (iv) hole convexity was not allowed. Optimal hole aspect ratios of h/w = 1.5 and 2.0 were considered, where h and w are
relative hole lengths in the x and y directions, respectively. Typically, the starting shapes for optimization were ellipses of the given
aspect ratio, with w = 12.7 mm. A robust shape optimization was also undertaken for the 2:1 case, where a bending moment at the
end was superimposed, such that the peak stress at (x, y) = (0, 0) varied by ±20%, in increments of 5%.

BL 52.46 y BL 48 BL 46.02

81.79 31.50 50.29

y
3.05

4.09
7.87 Dia

x z
31.58 0.51 32.26
46.99
2.21 2.08

(a) (b) (c)

Figure 12. Critical hole in stiffener #15 of AP-3C lower wing panel: (a) side view, (b) end view, and (c) local detail of FE model with
initial circular hole. All dimensions are in mm.

The various hole shapes and corresponding Kt results are given Figures 13a and 13b, respectively. The optimal holes provide
significant stress reductions. For example, the 2:1 optimal hole gives peak tensile Kt reductions at the top and bottom of the hole of
41.4% and 35.5%, respectively. Figure 14a gives sample stress decay plots below the hole along the y-axis. It is noted that there is
some difference in the stress decay rate between the standard and the robust optimal for the 2:1 case, due to their different shape.
20 5.0
Initial 1:1 hole Initial 1:1 hole
Optimal enlarged 1.5:1 hole 4.0 Optimal enlarged 1.5:1 hole
Optimal enlarged 2.0:1 hole Optimal enlarged 2.0:1 hole
15 3.0
y (mm)

2.0

Kt
10
1.0
θ
0.0
5
-1.0

0 -2.0
-15 -10 -5 0 5 10 15 0° 45° 90° 135° 180° 225° 270° 315° 360°
x (mm) Angle around profile, θ
(a) (b)
Figure 13. Comparison of initial circular hole and standard optimal holes in AP-3C stiffener, for various aspect ratios: (a) hole
shapes, and (b) Kt around hole boundary.

3.5
1.0:1 nominal
3.0
1.5:1 optimal
2.5 2.0:1 optimal
2.0:1 robust optimal
2.0
Kt

1.5
1.0
0.5
0.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0
y (mm)
Figure 14. Kt decay below the hole for a circular hole and optimal holes of various aspect ratios in the AP-3C stiffener.

9.2 Fatigue analysis for AP-3C holes

The fatigue calculations are given for prospective cracking below the hole. The estimates for relative lives as compared to the initial
circular hole are summarized in Table 4, where the Kt values are given in parentheses, and using Kn = 1.314Kt for all cases, based on
experimental data for the circular hole in a full scale fatigue test [16]. It can be seen that there are significant increases in life, where
the benefit increases with increasing aspect ratio. A 2:1 aspect ratio is sufficient to provide a five-fold increase in life. For the
inspection interval calculations, the β function for the circular hole case was determined separately [16], with good agreement
between the crack growth curve measured fractographically. The inspection interval results are summarized in Table 5. It can be seen
that for a given andi the inspection intervals are negligibly improved by using a stress optimal shape. This is due to the very small
relative distance between the bottom of the hole and the skin outer surface, as compared to andi. Hence, while optimal shapes for this
location would not increase inspection intervals, they will minimize the likelihood of a crack developing and requiring a repair, as a
result the safe-life increase. Also as expected, for a given h/w, the inspection interval decreases with increasing andi.

Table 4. Relative safe lives, tt2/tt1, and (Kt) for optimized holes in AP-3C stiffener.
Circular hole Optimal, h/w = 1.52 Optimal, h/w = 2.01 Robust optimal, h/w = 2.00
Relative life Relative life Relative life Relative life
1.0 (3.37) 4.38 (2.12) 5.71 (1.95) 5.62 (1.96)

Table 5. Relative inspection intervals, tr2/tr1, for optimized holes in AP-3C stiffener, compared to circular hole with andi = 1 mm.
Inspection crack Circular hole Optimal, h/w = 1.52 Optimal, h/w = 2.01 Robust optimal, h/w = 2.00
threshold, andi (mm) relative interval relative interval relative interval relative interval
1 1.00 0.93 0.96 1.00
2 0.57 0.55 0.54 0.54
3 0.38 0.41 0.40 0.38
4 0.29 0.32 0.32 0.29

10. Discussion and conclusion

In the ageing airframe context, some key points regarding the practical implementation of rework shape optimization arising from the
present paper are as follows:
(i) Even though we are constrained to only removing material, the optimal rework shapes typically provide large stress reductions
at the hole boundary, say 25–50% as compared to a nominal rework and initial circular holes. The minimum radius of curvature
 constraints are readily met, because as the hole aspect ratio is increased (say to 2 or above) its influence on stress is negligible.
(ii) For components managed under a safe-life regime, large life improvements are typical, since lives are driven essentially only by
the reduction in peak stress. Also, if the allowable life has been expended before reworking is undertaken, it may become
feasible to continue the life management of the location on a safe-life basis since the reworking will typically remove the locally
‘damaged’ material. However, in some cases the location may revert to management under safety-by-inspection, which is less
desirable from the viewpoint of minimizing maintenance costs.
(iii) For components managed by safety-by-inspection, for most examples, inspection intervals can be doubled, and from a life
management perspective this would typically be an excellent outcome. However, from the results given in this paper, there are
essentially three scenarios where there will be negligible benefit to inspection intervals: (1) if the relative size of andi compared
to the hole size is high; (2) the relative edge distance of the hole to the plate edge is small compared to the hole size; and
(3) available inspection techniques are such that andi for the optimal hole is significantly greater than for the optimal hole. In
these cases the problem arises because the stress optimals do not minimize the relative stress of the optimal hole compared to
the initial hole along the prospective crack path, only at the hole edge. To overcome this issue one option is to minimize, by
optimization, the stress intensity factor for a given andi. However, this results in shapes that do not maximize safe lives [21, 22].
Hence, there is a trade off between maximizing safe-life or inspection intervals, and the focus will depend on the relative cost of
inspections and on having to do repairs.
(iv) To make effective use of this technology for in-service applications, a quick turnaround time is needed. This requires efficient in
situ manufacturing methods. We are currently focusing on this for aluminum alloys, making use of templates in conjunction
with small automated hand-held tools. The optimal shapes also need to handle fleetwide variability of the surrounding structure;
hence the developments shown herein regarding in-built robustness can be very useful.
(v) In some cases, the obvious way to overcome some difficulties given above in (iii) is to locally rebuild some material at the stress
concentrator, and this option is currently being investigated. However, this needs be easy to be done in situ. Metallic rebuilding
using laser cladding may be an appropriate option, where ideally the shape of the rebuilt hole would be optimized.

12. References
1. Heller M, Kaye R, Rose L R F. A gradientless procedure for shape optimisation. Journal of Strain Analysis for Engineering
Design, 1999, 34(5): 323–336.
2. Waldman W, Heller M, Chen G. Optimal free-form fillet shapes for tension and bending. International Journal of Fatigue,
2001, 23: 509–523.
3. Burchill M, Heller M. Optimal free-form shapes for holes in flat plates under uniaxial and biaxial loading. Journal of Strain
Analysis for Engineering Design, 2004, 39(6): 595–614.
4. Burchill M, Heller M. Optimal notch shapes for loaded plates. Journal of Strain Analysis for Engineering Design, 2004, 39(1):
99–116.
5. McDonald M, Heller M. Robust shape optimization of notches for fatigue-life extension. Journal of Structural and
Multidisciplinary Optimization, 2004, 28(1): 55–68.
6. Heller M, McDonald M, Burchill M, Watters K C. F-111 airframe life extension through rework shape optimisation of critical
features in the wing pivot fitting. In: Proceedings of the 6th Joint FAA/DoD/NASA Aging Aircraft Conference, 16–19 September
2002, San Francisco, USA.
7. Baud R V. Fillet profiles for constant stress. Product Engineering, 1934, April: 133–134.
8. Neuber H. Zur optimirung der spannungskonzentration. Continuum Mechanics and Related Problems of Analysis, Nauka
Publishing House, Moscow, 1972, 375–380.
9. Vigdergauz S B, Cherkayev A V. A hole in plate, optimal for its biaxial extension-compression. Journal of Applied Mathematics
and Mechanics, 1986, 50(3): 401–404.
10. Schnack E. Computer-simulation of an experimental method for notch-shape-optimization. In: Proceedings of the International
Symposium ‘Simulation in Engineering Sciences,’ J Burger and Y Jarny (Eds), May 1983, Nantes, 311–316.
11. Mattheck C, Burkhardt S. A new method of structural shape optimisation based on biological growth. International Journal of
Fatigue, 1990, 12(3): 185–190.
12. Heywood R B. Photoelasticity for Designers (Chapter 11: Improvement of Designs), Pergamon Press, 1969.
13. Waldman W, Heller M. Shape optimisation of holes for multi-peak stress minimisation. In: Proceedings of the 4th Australasian
Congress on Applied Mechanics, 16–18 February 2005, Melbourne, Australia, 189–195.
14. Waldman W, Heller M. Shape optimisation for simultaneously minimizing multiple stress peaks around hole boundaries.
Journal of Structural and Multidisciplinary Optimization (submitted for publication, 2005).
15. Milne I, Ritchie R O, Karihaloo B (Eds). Comprehensive Structural Integrity, Vols 1–10. Elsevier Pergamon, 2003, ISBN
0080437494 (set). See Vol 4: Cyclic load and fatigue.
16. Hartley D, Jackson P. Personal communication. AP-3C Airframe Fatigue Assessment Group, DSTO, Australia, May 2005.
17. Grunwald J, Schnack E. A fatigue model for shape optimization. Structural Optimization, 1997, 14: 36–44.
18. Newman J C. FASTRAN-II – A fatigue crack growth structural analysis program. NASA Technical Memorandum 104159,
February, 1992.
19. Murakami E (Ed). Stress Intensity Factors Handbook, Vols 1 and 2. Pergamon Press, Oxford, 1987, ISBN 0080348092.
20. Evans R, Braemar R, Heller M. P-3C Aircraft life extension via shape optimization of holes in lower wing skin stiffeners. In:
Proceedings of the 4th Australasian Congress on Applied Mechanics, 16–18 February 2005, Melbourne, Australia, 801–806.
21. Peng D, Jones R, Xie M. The biological algorithm for solving shape optimisation problems with fracture strength and fatigue life
constraints. In: Proceedings of the 4th Australasian Congress on Applied Mechanics, 16–18 February 2005, Melbourne,
Australia, 711–716.
22. Jones R, Chaperon P, Heller M. Structural optimisation with fracture strength constraints. Engineering Fracture Mechanics,
2002, 69: 1403–1423.