You are on page 1of 43

M.Sc.

Degree in Mechanical Engineering

METHODS FOR ADVANCED MECHANICAL DESIGN


M. Guagliano

MULTIOBJECTIVE STRUCTURAL OPTIMIZATION


Structural optimization
•  Topological optimization, does not cover all the
aspects of the optimal structural design.

•  Topological optimization can be seen as a part of the


wider discipline of structural design optimization.

•  This latter can be defined as the decision process


finalized to identify the “best solution”.

•  When more than one aspects have to be optimized, it


is necessary to develop methods able to consider all
POLITECNICO DI MILANO

the different aspects to be optimized and able to


guide the engineer toward a rational decision instead
of relying on personal experience and feeling.
M. Guagliano – Multiobjective optimization
Structural optimization
Structural design

•  The process of structural design is based on a series


of equations that describe the structure behaviour,
once some assumptions are defined.
•  During the sizing and strength assessment steps
these equations are treated as inequalities (i.e.:
σ<σamm) that are due to the design constraints, and
that must be satisfied with a reasonable safety
margin
•  Generally, the number of the associated variables
and the equations does not allow us to consider them
POLITECNICO DI MILANO

in analytical terms.

M. Guagliano – Multiobjective optimization


Structural optimization

Structural design

Thus, the first step is pre-sizing, which allows us to estimate


the value of the variables supposed to be the most important
ones.

The solution respects every constraint of the design, but is not


unique.

Comment: the control of the engineer on the solution is quite


poor. (i.e: in this way we cannot understand how to modify the
POLITECNICO DI MILANO

solution to improve the results)

M. Guagliano – Multiobjective optimization


Structural optimization

Structural design

Summary

The interconnections among the different constraints (variable


subjected to more than one constraint) do not permit, apart
very simple cases, to modify the value of one variable without
losing the certainty that all the design constraints are still
respected.

POLITECNICO DI MILANO

????
M. Guagliano – Multiobjective optimization
Structural optimization
Task: how to find an optimal solution in a multi variable,
multi-constraint and (sometime) multi-objective problem?

•  Design by trial and error approach: time consuming


and without the certainty to find a solution able to
guarantee the best performances with respect of the
function to optimize.

•  Need to define a rigorous approach able to find out the


maximum/minimum values once a function to be
optimized is identified. (OPTIMAL DESIGN).
POLITECNICO DI MILANO

M. Guagliano – Multiobjective optimization


Structural optimization
Once it is assigned a physical system, it is assumed we are always able
to define a mathematical model able to describe the behaviour of the
system. The optimization process requires a clear definition of:

SYSTEM, makes a function (input/output)

MATHEMATICAL MODEL: description of the real system and of its


functions (modelling of the problem)

FIGURES OF MERIT: (objective functions to be defined)


__________
What is the criterion to follow?
POLITECNICO DI MILANO

“SIMPLEST MEANINGFUL MODEL” !

M. Guagliano – Multiobjective optimization


Structural optimization

MODELLING – CRITERIA

1.  Do not build a complicated model when a simple one will


suffice.
2.  Beware of moulding the problem to fit the technique.
3.  The deduction phase of modelling must be conducted
rigorously.
4.  Models should be validated prior to implementation.
5.  A model should neither be pressed to do, nor criticized for
failing to do, that for which it was never intended.
POLITECNICO DI MILANO

M. Guagliano – Multiobjective optimization


Structural optimization

MODELLING – CRITERIA

6.  Some of the primary benefits of modeling are


associated with the process of developing the
model.
7.  A model cannot be better than the information that
goes into it (Remember: Garbage In, Garbage Out
(GIGO). Models do not create information, but
condense or convert information).
POLITECNICO DI MILANO

M. Guagliano – Multiobjective optimization


Structural optimization - example

POLITECNICO DI MILANO

M. Guagliano – Multiobjective optimization


Structural optimization
FIGURE OF MERIT: objective functions (if they can be
quantified)

The objective functions are identified by relations like the


ones that follows:

The behaviour of a physical system is described by a discrete


number of functions that can be assessed by a set of
principles/standard strictly related to the formulation of the
POLITECNICO DI MILANO

design problem.
That set is called “set of design criteria” and must be minimal
M. Guagliano – Multiobjective optimization
Structural optimization

If the performances of the system can be expressed in


mathematical formulas, these formulas are called
objective functions.

The objective functions are expressed as functions of a


set of parameters, called design variables.

The objective functions are the quantities that the


engineering designer wishes to optimize

POLITECNICO DI MILANO

M. Guagliano – Multiobjective optimization


Optimization of a cantilever

EXAMPLE: design of a cantilever with rectangular section.

Objective: to find the design variables, b and h, of the section


able to gurantee:

-  Minimum mass
-  Minimum deflection

By respecting the following constraints:

•  The maximum stress must be lower than the allowable one;


POLITECNICO DI MILANO

•  Buckling must be avoided.

M. Guagliano – Multiobjective optimization


Structural optimization

DESIGN VARIABLES
•  The design variables are the variables that are related
to the mathmatical model of the system during the
optimization design. These variables can be grouped
in a vector and can be identified as:

•  Sometime the engineer can choose to fix some of


those variables to simplify the optimization process.
Thse fixed variables can be re-classified as
parameters. The choice of the variables is not unique
POLITECNICO DI MILANO

but it is quite important to select independent


variables
M. Guagliano – Multiobjective optimization
Structural optimization

DESIGN VARIABLES- DEFINITION OF CREATIVITY

-  If the number of variables and the ranges of the values


remain the same during the design process, this is routine
design;

-  If the number of variables remains the same but the range


can varies, then this is innovative design,

-  If the number of variables is changing, this is creative


design.
POLITECNICO DI MILANO

M. Guagliano – Multiobjective optimization


Optimization of a cantilever

EXAMPLE: design of a cantilever with rectangular section.

Objective: to find the design variables, b and h, of the section


able to gurantee:

-  Minimum mass
-  Minimum deflection

By respecting the following constraints:

•  The maximum stress must be lower than the allowable one;;


POLITECNICO DI MILANO

•  Buckling must be avoided.

M. Guagliano – Multiobjective optimization


Optimization of a cantilever

DESIGN CONSTRAINTS
The constraints are conditions to guarantee the functions the
system is designed for.

Where x is the vector of the design variables. The constraints


are generally limitations to the range the variable values can
assume (side constraints, called lower bounds, upper bounds).

The constraints of the system model are constraints on the


POLITECNICO DI MILANO

output of the system (values of the objective functions.


Sometime such limits are given by the physical laws governing
the system
M. Guagliano – Multiobjective optimization
Optimization of a cantilever

EXAMPLE: design of a cantilever with rectangular section.

Objective: to find the design variables, b and h, of the section


able to gurantee:

-  Minimize the mass


-  Minimize the deflection

By respecting the following constraints:

•  The maximum stress must be lower than the allowable one;;


POLITECNICO DI MILANO

•  Buckling must be avoided.

M. Guagliano – Multiobjective optimization


Optimization of a cantilever

Example: design of a cantilever with rectangular section.

POLITECNICO DI MILANO

M. Guagliano – Multiobjective optimization


Optimization of a cantilever

Example: design of a cantilever with rectangular section.


To solve the problem it is possible to calculate m(b, h) and y(b, h) as a
function of all the possible combinations of b and h.

(a) Range of b and h values considered in the optimization step, (b) beam
POLITECNICO DI MILANO

mass m(b, h) and beam deflection y(b, h). The set of the values in (b)
considers the constraints on the maximum stress and on buckling,

M. Guagliano – Multiobjective optimization


Optimization of a cantilever

Example: optimized design of a rectangular cantilever


To generate the functions m(b, h) and y(b, h) all the possible combinationns
of the design variables b e h have been considered (after having verified
that m(b, h) and y(b, h) satisfy the design requirements).

The question is:


How can we find the values of b and h that minimise both the
objectives (minimum weight and minimum deflection ant the free end)
at the same time?

The solution is not trivial and requires some preliminary thought.


POLITECNICO DI MILANO

M. Guagliano – Multiobjective optimization


Optimization of a cantilever

Example: optimized design of a rectangular cantilever

Consider the set of values mA(bA, hA) and yA(bA, hA)


This couple of values bA, hA defines a design solution that should not be
considered by the designer. In other words it shuold not be considered for
POLITECNICO DI MILANO
optimization.

M. Guagliano – Multiobjective optimization


Optimization of a cantilever

Example: optimized design of a rectangular cantilever

If a not optimal solution A is considered, other solutions exist that fit better
than A at least one objective function.
The solutions that do not lie on the bold line (with ymin and mmin as extreme
values) should be not considered by the design engineer.
POLITECNICO DI MILANO

On the contrary, the optimal solutions lies on that line, called, detta “Pareto-
optimal set”. The solution the engineer should look at and choose should be
one of the Pareto optimal set the optimal solution is not unique.
M. Guagliano – Multiobjective optimization
Optimization of a cantilever
Example: optimized design of a rectangular cantilever - Summary
The solution of the problem where just the mass has to be optimized can be
expressed as:

In a similar way, if just the deflection is of interest and must be optimized:

This means that if the function to optimized is scalar, the optimal solution is
unique:

POLITECNICO DI MILANO

But, if the function to optimize is a vector, the solution is not unique,

M. Guagliano – Multiobjective optimization


Optimization of a cantilever
During the optimization of a complex system, the engineer, before taking any
decision, must get a deep knowledge of the system of interest and of its
performance. In particular must understand the limits of the system.
.

If the reference case is considered, assume that point D cannot be reached if


the system is not modified. Also mmin e ymin are important limits of the objective
POLITECNICO DI MILANO

functions (performances of the system). These limits depend on the system


investigated and could be changed by varying the admissible range of the
design variables.
M. Guagliano – Multiobjective optimization
Optimization of a cantilever

An important reference point is the so called “ideal point” or “utopic point”, U.


This is the point whose coordinates are (mmin, ymin): a cantilever with such
characteristics cannot exist but, if it were possible, would be the best solution.
U can be used to choose the optimal solution among the ones of the Pareto
set.
POLITECNICO DI MILANO

By intuition and instinct the engineer is guided toward solutions localized near
the point U.

M. Guagliano – Multiobjective optimization


I metodi evoluzionari

POLITECNICO DI MILANO

M. Guagliano – Multiobjective optimization


Optimal design – general method

OPTIMAL DESIGN – SINGLE OBJECTIVE


Objective of the scalar optimal design is to identify the values of the design
X with some assigned constraints g(x), in such a way that the one single
function is minimized (maximized). In formulas:

The subset is the space of the design variables.


(if the objective function must be maximized and not minimized, the dual
function can be introduced, f’(x)=-f(x)
POLITECNICO DI MILANO

M. Guagliano – Multiobjective optimization


Optimal design – general method

OPTIMAL DESIGN – SINGLE OBJECTIVE - DEFINITIONS

POLITECNICO DI MILANO

M. Guagliano – Multiobjective optimization


Optimal design – general method

OPTIMAL DESIGN – SINGLE OBJECTIVE DEFINITIONS

M. Guagliano – Multiobjective optimization


Optimal design – general method

OPTIMAL DESIGN – SINGLE OBJECTIVE


Existence of the solution
The problem must be well posed, the domain of the feasible solutions
(defined by the functional constraints) is not empty. The problem is well
constrained.

Example:
Min f(a)=a P/a<Sy/η a<Amax
(a=cross section – P=applied load - Amax =maximum section - η=safety factor)

If Pη/Sy>Amax no feasible solutions exist.

M. Guagliano – Multiobjective optimization


Optimal design – general method

OPTIMAL DESIGN – SINGLE OBJECTIVE


Active constraints
An active constraint is a constraint that, if removed, alterates the position of
the minimum value.

(this is important for the engineer).

M. Guagliano – Multiobjective optimization


Optimal design – general method

MULTIOBJECTIVE OPTIMAL DESIGN –


Pareto optimal solution
When a multiobjective problem is assigned with ndv design variables and nof objective
functions, the optimal Pareto solution x*i (not dominated, efficient and not inferior)
satisfies the following conditions:

The optimal Pareto solutions are infinite. The full set of Pareto optimal solutions
defines the set of the optimal solution according to the Pareto theory.
Assigned a solution x*i, if an objective function is changed to improve it, at least
another one is made worst.
For the non-optimal solutions the value of at least one objective function fl can be
reduced without increasing the vaues of the other functions.
The impossibility to optimize at the same time all of the objective functions is a
characteristic of the multi-objective problems (MOP).
M. Guagliano – Multiobjective optimization
Optimal design – general method

Summary

The “Pareto Optimum” or “Pareto efficiency” is a concept that is realized when


the resource allocation makes not possible to improve the system in the Pareto
way; this means that it is not possible to improve the value of an objective
function without worsen another one

The Pareto front (set) is the set of the optimal solutions, that is the front of the
“not dominated” solutions, where it is not possible to decrease the value of an
objective function without increasing at least one of the other objective functions.

According to the Pareto theory a solution dominates another solution if it is better


of this latter with respect of one of the objectives, without worsen any other
objective.

M. Guagliano – Multiobjective optimization


Optimal design – general method

MULTIOBJECTIVE OPTIMAL DESIGN


Example: optimal solution, two-objective
optimization

The dashed area represents the space of


the design variables in the space of the
objective functions. The set of the optimal
Pareto solutions is defined by the curve
b-c.

Solution a is not optimal: every solutionin the


indicated area can reduce both the objective
functions f1 e f2.

M. Guagliano – Multiobjective optimization


Optimal design – general method

OPTIMAL DESIGN– Implementation

The Pareto’s theory considers simultaneously contrasting objectives and does


not consider scalar functions; it requires particular algorithms to be
implemented

The problem of the research of the optimum design can be faced, from a
computational point of view, by using different optimization algorithms both
evolutionary, based on genetic routines and neural networks, and more
traditional deterministic algorithms.

In general, different algorithms are used to solve different multiobjective


linear/non linear/continuous/discrete/mixed, etc. problems.
Different strategies can be chosen to afford different problems.
M. Guagliano – Multiobjective optimization
Optimal design – general method

The concept introduced with the cantilever case can be generalized to


complex systems with many design variables and different objective
functions. The solving procedure changes with the complexity of the
problem.
Assume to face the problem of a system defined by 10 design variables
Ndv and 30 objective functions nof. Each design variable can assume nv=
10 different values within the definiton range. Assigned a combination of
the design variables, consider ts=1s the time needed for a simulation albe
to give the valu of a single deign function. With the general method, the
number of the design variables is:
ncdv=nvndv
And the total time needed to calculate all the objective functions as a
function of all the possible combunations of the design variables is:
POLITECNICO DI MILANO

Tt=tsnofnvndv, This means: 3E11s=9645 years!!! NOT AFFORDABLE!!

M. Guagliano – Multiobjective optimization


Optimal design – general method

OPTIMAL DESIGN – PARETO OPTIMAL SOLUTION


n-OBJECTIVE OPTIMIZATION
It is necessary to generate, first of all, a meaningful number of data to be
analyzed.
.

M. Guagliano – Multiobjective optimization


Optimal design – general method

OPTIMAL DESIGN – PARETO OPTIMAL SOLUTION


Pareto optimal solution – Method of the weighted sum

A popular method to find the set of the optimal Pareto optimized solutions is
the method of the weighted sum. By minimizing the weighted sum of the
objective functions with different weight factors, different points on the
Pareto sets can be generated
(the objective functions are normalized, the weights are between 0 and 1)
By changing the weights, different points of the optimal Pareto sets are
systematically generated..
The weights are not directly related to the importance of the objective
functions but allow us to localize some points of the optimal Pareto set.

M. Guagliano – Multiobjective optimization


Optimal design – general method

OPTIMAL DESIGN – PARETO OPTIMAL SOLUTION


Pareto optimal solution – Method of the weighted sum
The optimal Pareto set is obtained by changing the coefficients λ.

Es.: two objectives optimization

The Pareto set is found by varying the


Coefficients λi.

M. Guagliano – Multiobjective optimization


Optimal design – general method

OPTIMAL DESIGN – PARETO OPTIMAL SOLUTION


Pareto optimal solution – Method of the constraints
A more efficient method to find the Pareto optimal set of solutions is the
Constraints Method. Only one function is considered as objective function,
the other ones are considered as if they were constraints.

The problem is reduced to a standard non linear problem and the Pareto set
is obtained by varying the constraint levels εi, i= 2, 3…nof.
This makes of sure the existence of the solution for the constrained
problem. If the problem is solved for every possible value of εi and the
resulting solutions are unique, these solutions form the Pareto optimal set.
Otherwise the points of the set must be selected by direct comparison.
M. Guagliano – Multiobjective optimization
Optimal design – general method

OPTIMAL DESIGN – PARETO OPTIMAL


SOLUTION
Pareto optimal solution – Method of the
constraints
If just two objectives are considered, thus
the minimum of f1(x) is the point indicated in
the figure.
The region of the possible solutions is the
grey one. The minimum f1(x) is a not
dominated solution of the problem MOP. By
varying the constraint level ε a different
minimum is calculated for f1 and a different
not dominated solution.

M. Guagliano – Multiobjective optimization


Optimal design –

POLITECNICO DI MILANO

M. Guagliano – Multiobjective optimization

You might also like