You are on page 1of 5

Design Engineering – DE3 Optimisation

Tutorial Sheet 7 (Lectures 13 & 14)

Lecturers: Dr Nan Li, Dr Michel-Alex. Cardin

Teaching Assistants: Cesare Caputo, Haosu Zhou, Wenbin Zhou

Provided: Week 10

Notes:
- Students should solve Part A individually and work on Part B individually or together
with your project partners.

Topics: Multi-objective Optimisation; Parametric Study & Sensitivity Analyses; System


Optimisation; Optimisation with Uncertainty.

PART A

1. Consider the problem

min 𝑓(𝐱)
𝐱

s.t. ℎ1 (𝐱) = 𝑝𝟏
ℎ2 (𝐱) = 𝑝𝟐

where 𝑝𝟏 and 𝑝𝟐 are the model parameters that have been set for the optimisation
problem. If Lagrange Multipliers 𝜆𝟏 < 0 and 𝜆𝟐 > 0 at the solution, how could we
potentially improve the solution?

𝜕𝑓𝑇
Since −𝝀 = , this implies that we could decrease 𝑝1 and increase 𝑝2 to improve the solution.
𝜕𝐡

DE3 Optimisation. 2020-2021. V1.0.


2. Consider the beam problem that we have encountered before (equations below).
Variables or parameters may include force 𝐹, width 𝑤, height ℎ, length 𝑙, material
density 𝜌, Young’s modulus 𝐸, and cost per unit mass of material 𝑐. Assume that the
force is prescribed as 100 N, and the maximum internal stress is not to exceed 100 MPa.
Neither the width nor the height may exceed one tenth of the length.

𝐹𝐿3
deflection 𝛿=
3𝐸𝐼
6𝐹𝐿
maximum internal stress 𝜎𝑚𝑎𝑥 =
𝑤ℎ2
volume 𝑉 = 𝐿𝑤ℎ
𝑤ℎ3
area moment of inertia 𝐼=
12

a. Write a full formulation for a multi-objective optimisation problem that minimises


material cost and beam deflection. Treat material properties as independent and
continuous variables.

min {𝐶𝑡𝑜𝑡 ,𝛿}

w.r.t 𝑤, ℎ, 𝑙, 𝜌, 𝐸, 𝑐

where 𝐹 = 100 N
𝑤ℎ3
𝐼=
12
𝑉 = 𝑤ℎ𝑙
𝐶𝑡𝑜𝑡 = 𝑉𝑐𝜌
6𝐹𝐿
𝜎𝑚𝑎𝑥 =
𝑤ℎ2
𝐹𝐿3
𝛿=
3𝐸𝐼

s.t. 𝜎𝑚𝑎𝑥 − 100 𝑀𝑃𝑎 ≤ 0


𝑙
𝑤− ≤0
10
𝑙
ℎ− ≤0
10

b. Explain how you would solve the bi-objective problem from part a. to find a Pareto
set using a single-objective algorithm.

Include a formulation that can be solved parametrically (if you did not provide it in
part a.)

Note there is more than one way to do this, but only one is needed.

Using the weighted sum approach, we can create a combined objective function and solve
parametrically with different values for the weights (𝑤1 , 𝑤2 ).

min 𝑤1 𝐶𝑡𝑜𝑡 + 𝑤2 𝛿
𝑤,ℎ,𝑙,𝜌,𝐸,𝑐

DE3 Optimisation. 2020-2021. V1.0.


where 𝐹 = 100 N
𝑤ℎ3
𝐼=
12
𝑉 = 𝑤ℎ𝑙
𝐶𝑡𝑜𝑡 = 𝑉𝑐𝜌
6𝐹𝐿
𝜎𝑚𝑎𝑥 = 2
𝑤ℎ
𝐹𝐿3
𝛿=
3𝐸𝐼

s.t. 𝜎𝑚𝑎𝑥 − 100 𝑀𝑃𝑎 ≤ 0


𝑙
𝑤− ≤0
10
𝑙
ℎ− ≤0
10
0 ≤ 𝑤1 ≤ 1

c. Optional: Identify 2 potential sources of uncertainty and explain which variable(s),


parameter(s), constraint(s), and/or objective(s) they correspond with.

Then, formulate this problem as a reliability-based design optimisation problem


(with respect to those sources of uncertainty) that seeks to minimise cost with a
maximum deflection of 1mm.

Make and explain any assumptions that you deem necessary for this formulation.

There may be uncertainty in:

1. Physical dimensions due to manufacturing tolerances, which affects variables 𝑤, ℎ and 𝑙, and
outputs 𝜎𝑚𝑎𝑥 and 𝛿.

2. The force applied 𝐹, which affects 𝜎𝑚𝑎𝑥 and 𝛿.

min 𝐶𝑡𝑜𝑡 = 𝑉𝑐𝜌


𝑤,ℎ,𝑙,𝜌,𝐸,𝑐

where 𝐹 = normal dist. with mean 𝜇𝐹 = 100 N and std. dev. 𝜎𝐹 = 10 N


𝑤ℎ3
𝐼=
12
𝑉 = 𝑤ℎ𝑙
6𝐹𝐿
𝜎𝑚𝑎𝑥 = 2
𝑤ℎ
𝐹𝐿3
𝛿=
3𝐸𝐼

s.t. 𝑃(𝜎𝑚𝑎𝑥 > 100 𝑀𝑃𝑎) − 0.05 ≤ 0


𝑃(𝛿 > 1𝑚𝑚) − 0.05 ≤ 0
𝑙
𝑤− ≤0
10
𝑙
ℎ− ≤0
10
Assumptions:

1. We can accept a 5% probability of failure for stress and deflection requirements,


independently.

2. Dimensions 𝑤, ℎ and 𝑙 may vary from the nominal design by 0.1 mm. Therefore, when
calculating 𝜎𝑚𝑎𝑥 and 𝛿, we use a normal distribution with mean at the nominal design and
0.1
standard deviation at mm.
3

DE3 Optimisation. 2020-2021. V1.0.


3. Optional: a) Download the Tutorial 7 MATLAB code from BB. Explore how the different
strategies (i.e. constraint-based and weighted sum) to achieve multi-objective
optimisation for the design of a cantilever beam can be implemented. b) Try using
alternative algorithms.
NSGA2: https://uk.mathworks.com/matlabcentral/fileexchange/31166-ngpm-a-nsga-ii-
program-in-matlab-v1-4

5 methods (2 for Q3a and 3 for Q3b) are used to solve the problem. Code posted on BB.

DE3 Optimisation. 2020-2021. V1.0.


PART B

Having solved your optimisation through Tutorials 3-6, you can now conduct post-optimality
analysis in this tutorial. For example, analyse which is the most important active constraint?
(This can possibility be done through sensitivity analysis). Try changing the most important
active constraint by some amount, and re-optimise, and compare the new optimum with
the previous optimum. What do you observe?

Select two objective functions for your problem. Optimise your problem for both objectives
simultaneously using any multi-objective method of your choice. Try to answer the
following questions: (i) Are your objectives mutually supportive or opposing? How do you
know? (ii) What is an estimate of the Pareto frontier? Show this graphically. (iii) How
sensitive is your answer to the weights (if you use a method that requires weights as in the
weighted sum method) you have chosen? (Optional)

Analyse the potential sources of uncertainty of your problem and explain which variable(s),
parameter(s), constraint(s), and/or objective(s) they correspond with. Then, formulate your
problem as a robust design optimisation problem, or a reliability-based design optimisation
problem, with respect to those sources of uncertainty. Make and explain any assumptions
that you deem necessary for this formulation. (Optional)

If you work on both system and sub-systems, think about the coordination between them.
Think about how to conduct your system-level design and optimisation collaboratively with
the sub-system-level designs. What strategy would you choose and why? What design
variables would you choose from your subsystems as the design variables for your system?
Then, explain the rationale behind your choices; optimise the problem; conduct necessary
post optimality analysis as appropriate. (Optional)

No standard solutions available. If you have any issues, feel free to post your questions to
the Piazza forum, or discuss with one of the instructors during tutorial hours.

DE3 Optimisation. 2020-2021. V1.0.

You might also like