Size Optimization of Plane Trusses subjected to Stochastic Loads

Student: Andrés Y. Moreno-Villa

Proffesor: Nicolás Guarín-Zapata, Ph.D.

Structural Optimization
Final Report

School of Applied Sciences and Engineering

Masters in Applied Physics
Noviembre del 2025
 1. Introduction

In real-world engineering applications, structural systems are commonly subjected to loads that are
inherently stochastic in nature rather than deterministic. This uncertainty poses significant challenges
in structural design, particularly when optimization techniques are employed to improve performance,
reduce weight, or minimize cost. For instance, consider a simple square truss subjected to a stochastic
point load at its bottom-right corner: half of the time the force has magnitude F and acts in the positive
x-direction, while the other half of the time it has the same magnitude but acts in the negative y-
direction. If topological optimization is applied to each of these loading conditions individually—or by
assuming an average load—the resulting optimal topologies differ markedly, and none provides robust
performance when the loading deviates from the case it was designed for. This discrepancy illustrates
that topological optimization is not a linear process, and that superimposing or averaging loading
conditions fails to capture the true design requirements under uncertainty. Moreover, simply designing
for a worst-case scenario does not necessarily yield optimal or reliable results. Consequently, there
is a need for improved design methodologies capable of explicitly accounting for stochastic loading
conditions to ensure structural robustness in real-world applications.

(a) Three main load cases that could be considered, given the stochastic nature of the load being applied to the
square truss.

(b) Applying topological optimization to the square truss, by considering each of the 3 cases separetely yields
completely different results.

Figura 1. Topological optimization of square truss subject to a single stochastic load.

2 / Propuesta de Tesis

2. Methodology

2.1. Implementation of FEM solver in Python

In accordance with what was proposed in the schedule for this project, a finite element code was
developed in Python, in order to solve 2D trusses. This code was then validated by running two static
analyses on two different trusses, seen in figures 2 and 3, and comparing the results with those obtained
by using the program developed by Valdivia (2025).

Figura 3. Validation test 2

Figura 2. Validation test 1

In both validation cases, the transversal area (A) of all bars was set to 5 × 10−4 m2 , the elasticity
modulus was set to 2 × 1011 Pa and the force applied (F ) had a magnitude of 1000 N. In the case of
Figure 2 the vertical and horizontal lines were the same length, equal to 1 m. On the other hand, for
case Figure 3, horizontal bars were equal to 2 m in length, while vertical bars were set to be a quarter
of that, being 0.5 meters in length. The results of the simulations performed using the in-house FEM
solver can be seen from Figures 4 and 5.

Truss Stress Distribution Truss Stress Distribution

1.0
0.4
0.8
0.2
0.6
0.0
Y

0.4
0.2
0.2
0.4
0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
X X

2 1 0 1 2 4 3 2 1 0 1 2 3 4
Stress (Pa) | Red = Compression, Blue = Tension 1e6 Stress (Pa) | Red = Compression, Blue = Tension 1e6

Figura 4. Validation test 1: Stress distribu- Figura 5. Validation test 2: Stress distribu-
tion of truss members using in-house solver. tion of truss members using in-house solver.

The difference between the stresses obtained for truss members using the in-house solver and the
solver developed by Valdivia (2025) was evaluated and no substantial difference between them was
found.

2.2. Case Study

As a case study for the present work, we propose the analysis of a truss cantilever beam subjected to
a point load at one of its ends. Figure 6 shows the geometric dimensions, material properties, boundary
conditions and loading conditions which are to be considered in the following analysis. In particular,
we require that the force F⃗ have magnitude r and direction θ, randomly chosen from some continuous
multivariate distribution with probability density function (pdf) given by D(r, θ; p⃗r , p⃗θ ), where p⃗r and
p⃗θ stand for all the parameters related to r, θ, respectively, which together fully characterize such pdf.

Figura 6. Caption

The stochasticity of this problem then arises from the specific choice of D. Whenever D is just a
Dirac delta, we recover the deterministic loading case. For the present case, we construct a multivariate
pdf by first defining the truncated normal distribution of some variable x with mean µ and standard
deviation σ as:

 ´ xmaxN (x;µ,σ) xmin ≤ x ≤ xmax
N (x;µ,σ) dx
N ∗ (x; µ, σ, xmin , xmax ) = xmin
0 elsewhere

Where N is just the pdf of the normal distribution. We then define the pdf of our continuous
multivariate distribution D as:

D(r, θ; p⃗r , p⃗θ ) = N ∗ (r; µr , σr , rmin , rmax )N ∗ (θ; µθ , σθ , θmin , θmax )

Where it follows that p⃗r = (µr , σr , rmin , rmax ) and p⃗θ = (µθ , σθ , θmin , θmax ). Figure 7 presents the
continuous (ϕ) and discretised (ϕ̂) versions of D that were used for the present case study.
4 / Propuesta de Tesis

Figura 7. Continuous and discretised versions of D used for the present case study. Here ’M ’ refers to
r.

The following values of p⃗r and p⃗θ were employed:

h i
p⃗r = 105 , 103 , 7 × 104 , 1.3 × 105 [N]
h i
p⃗θ = − 45, 60, −70, −20 [◦ ]

2.2. Survivability Metric

Given the stochastic nature of the loading described previously, it becomes necessary to define
a performance metric that quantifies the likelihood that a given structural topology remains viable
when subjected to random realizations of F⃗ ∼ D(r, θ). In this work, we assess performance through
a probabilistic survivability measure, defined as the probability that the structure does not exceed a
prescribed compliance threshold Cmax when subjected to a random sample drawn from D(r, θ; p⃗r , p⃗θ ).
Formally, this measure can be expressed as

ˆ θf ˆ rf h i
S= ⃗ dr dθ,
ϕ(r, θ)H Cmax − C(r, θ, A)
θ0 r0

where ϕ denotes the continuous probability density function, H(·) is the Heaviside step function,
⃗ represents the structural compliance for a truss with element areas A
and C(M, θ, A) ⃗ subject to a load
of magnitude r and direction θ. In practice, the compliance is computed as

C = f T u,

with f and u denoting the external force vector and the corresponding displacement response,
respectively.
 To render the computation tractable, the continuous integral is approximated using a discrete set
of load realisations sampled from D. The discrete analogue of the survivability metric is therefore given
by

XX h i
Ŝ = ⃗ ∆r ∆θ,
ϕ̂(r, θ)H Cmax − C(r, θ, A)
r θ

where ϕ̂ denotes the discretised probability distribution and (∆r, ∆θ) are the corresponding sam-
pling widths. A structural configuration is regarded as highly survivable when Ŝ approaches unity,
indicating that the design can withstand almost all expected loading scenarios.

2.2. Optimization Problem Formulation

The objective of the present case study is to determine the set of truss cross-sectional areas A ⃗
that yields a structure which is both lightweight and capable of withstanding uncertain loading condi-
tions. To this end, two alternative formulations are considered: a deterministic formulation, where only
one nominal load is applied, and a stochastic formulation, where survivability under random loads is
explicitly accounted for.
In the deterministic case, the optimization problem seeks to minimize a weighted sum of total struc-
tural volume and a barrier penalizing violations of the compliance constraint. This can be expressed
as

ne
X h i
mı́n L = ⃗ ,
li Ai + µ ln Cmax − C(A)
⃗i
A i=0

where li is the length of element i and µ is a penalty parameter that enforces compliance feasibility.
In contrast, the stochastic formulation incorporates the discrete survivability measure introduced
previously, penalizing designs that fail to meet a prescribed minimum survivability level Smin . The
resulting optimization problem becomes

ne
X h i
mı́n L = ⃗ − Smin ,
li Ai + µ ln Ŝ(A)
⃗i
A i=0

⃗ is the survivability associated with the current design. The inclusion of the logarithmic
where Ŝ(A)
barrier term discourages solutions whose survivability drops below Smin , thereby implicitly enforcing
robustness to uncertainty.
For both formulations, element areas are bounded to satisfy standard manufacturing or structural
constraints,

Amin ≤ Ai ≤ Amax , ∀i,

ensuring that the resulting topologies remain physically realizable.

6 / Propuesta de Tesis

3. Results

3.1. Deterministic Case

We consider the size optimization of a beam truss subjected to a deterministic load equal to the
average of ϕ̂. The applied load is therefore F⃗ (105 N, −45◦ ). All truss elements were initialized with
transverse cross-sectional areas of A0 = 1.5 × 10−5 , m2 and an elastic modulus of E = 2 × 1011 , Pa.
The pre-optimization stress and area distributions are shown in Figure 8a. The total volume of the
structure before optimization was 0.0619 m3 and its compliance was 483.21. For the optimization, we
require that the compliance does not exceed 1450, while the total volume of the truss is minimized.
Additionally, we set Amax = A0 and Amin = A0 /1500.

(a) Beam truss before optimization. (b) Beam truss after optimization.

Figura 8. Size optimization of a beam truss subjected to a deterministic load F⃗ (105 , N, −45◦ ).

For the optimization procedure, we utilized scipy’s minimize function. Figure 8b shows the stress
and area distributions of the beam truss post-optimization. Compared to the pre-optimization distribu-
tions, the stress field is much more uniform, while the area distribution is considerably more dispersed.
This indicates that the optimization procedure redistributes bar thicknesses to improve their structural
efficiency. After the optimization, the total volume of the structure was 0.00717, m3 , representing an
88.4 % reduction. Meanwhile, the compliance increased to 1430, but did not exceed the prescribed limit.
Therefore, we conclude that, for the deterministic case, our implementation performs as expected.

3.2. Stochastic Case

For the stochastic case, all bar elements of the beam truss were again initialized with transverse
areas A0 and elastic modulus E. The initial volume was therefore the same as in the deterministic
case, equal to 0.0619, m3 . We again set Amax = A0 and Amin = A0 /1500. The maximum allowable
compliance was set to 1450, but since we are now dealing with a stochastic load, we must also prescribe
a minimum survivability threshold. In this case, it was set to S = 0.8 (i.e., an 80 % probability of
survival). Because of the stochastic nature of the loads, the magnitude of the applied forces, as well as
the resulting compliance values, vary for each sampled load case. Figure 9 shows the stress response
of the beam truss when subjected to different loads sampled from ϕ̂. It also presents the probability
associated with each load case and the corresponding structural compliance.
 Figura 9. Stress distribution and compliance of the beam truss before optimization, when subjected to
different loads sampled from ϕ̂.

Initially, the beam truss exhibits a survivability of 99.9 %. After optimization, this value drops to
86.6 %, representing a 13.45 % reduction but remaining within the acceptable limits defined earlier. The
optimized volume is 0.0334, m3 , corresponding to a 46.1 % reduction. Interestingly, the area distribution
of the truss remains nearly uniform, indicating that the optimization algorithm primarily reduced the
cross-sectional areas of all bars by approximately the same proportion. The underlying reason for
this behavior is not fully understood, although it is hypothesized that the multiple loading directions
of F⃗ promote a more uniform allocation of material. Further investigation is required to assess this
phenomenon in greater detail.

4. Conclusion

This work presented a methodology for optimizing truss structures under stochastic loading by
combining probabilistic modeling, finite element analysis, and constrained optimization. The results
demonstrated that deterministic optimization can achieve substantial volume reduction but fails to
account for variability in loading, whereas the stochastic formulation incorporates survivability cons-
traints that improve robustness at the expense of higher mass. These findings highlight the importance
of explicitly incorporating uncertainty in structural optimization, particularly when designing light-
weight systems subjected to unpredictable loads. Future work should explore alternative formulations,
sampling strategies, and structural topologies to further enhance performance under uncertainty.

Referencias

Valdivia, A. (2025). 2d-truss analysis. [Link]

Accessed: 2025-11-25.