Applying multi-objective optimization to a ventilation duct for an automotive vehicle

Motivations 1
1. Introduction
1.1 Motivations
Optimization is the process of searching for one or more solutions to a problem until no better
one can be found. The results are considered optimal in terms of superior characteristics with
respect to one or more objectives compared to all other solutions. The need to find optimal
solutions comes mostly from extreme purposes of finding, for example, the minimum cost of
fabrication or the maximum possible reliability or others.
Multi-objective optimization is strictly the task of finding the optimal solution in presence of
more than one objective. As the objectives are more than one, the optimum cannot be simply
found for only one objective when the rest of the objectives are also important. Different
solutions can represent trade-offs in conflicting scenarios among different objectives. A
solution with extreme characteristics with respect to one objective might not be (and usually is
not) an extreme for the rest of the objectives. It is commonly a compromise in other objectives.
For this reason we cannot simply choose a solution which is optimal with respect to only one
objective. Interestingly, most real world problems naturally involve multiple objectives.
This aim of this work is to apply optimization techniques to a real world problem: a
ventilation duct of an automotive vehicle. Automotive projects commonly pose several
problems to engineers. Among them, ducts of the ventilation system are required to provide
sufficient airflow to the outlets limiting the noise detectable in the cockpit. Airflow is required
to heat (or refresh) sufficiently a portion of the cockpit, while loudness should be avoided
mainly for passenger comfort. These qualitative requirements are translated into quantitative
objectives, namely minimal pressure losses (better airflow) and minimal noise. Project
prescriptions often require values of pressure loss and noise levels to be in a predetermined
target range. 
In this context, we seek to find one or more optimal solutions to the multi-objective problem
of minimizing the pressure loss and reducing the noise levels in the duct we were given to study.
Traditionally, optimization from the engineering perspective is driven by human knowledge
and applied when detail design is already performed or inherited from previous projects.
Another approach can be followed by introducing automated optimization in the early stages of
the engineering process to envision feasibility and obtain design indications. This approach will
inevitably be different.
Traditional approaches to optimization involve human factors as experience and insight.
Engineers foresee how a design can be improved and experiment different solutions. This
approach involves a “trial and error” iterative process where new designs are repetitively
verified until a satisfactory solution is found. Moreover, the approach does not ultimately
guarantee for a successful outcome as problems could be extremely difficult to solve. Generally,
an acceptable trade-off design is considered to be sufficient. This methodology is affected by a
couple of drawbacks: 
• The engineer experiments only a limited number of designs whereas combinations and 
possibilities could be much more;
• The search process is limited by human decision of which designs should be verified.
2I N T R O D U C T I O N
Differently from this approach, we claim that an automatic optimization process based on CAE
tools can search for solutions more extensively and pursuit optimal designs more efficiently.
Another problem with traditional optimization is the stage of the project at which
optimization is applied. Late optimization poses some drawbacks:
• Geometries are already developed in detail. Any change implicates acceptance from 
several departments and a significant number of persons involved;
• Changes imply higher costs in terms of rework and correction needed.
On the other hand, optimization performed in the early stages of the development process is not
affected, or rather marginally, from the outlined shortcomings. 
We are essentially advising for a new approach that combines the advantages of innovative
techniques and the benefits of early phase implementation. Not surprisingly, novel approaches
are expected to have bigger benefits if applied to the early phases of a project. However, as
geometries aren’t yet defined in the early stages, we can only aim at obtaining design indications
to be carried out and detailed in a later phase.
1.2 Scope of study
The scope of study is the front ventilation system, part of the cockpit dashboard assembly, of an
automotive vehicle which is supposed to provide sufficient air flow through the outlets. This
study is based on a real world design problem taken from an automotive engineering project. 
The project is a mainstream car for the public, a redesigned version of a famous model which
saw its infancies in the late 30ies. As it happens often for new designs of existing vehicles, the
car is totally re-engineered keeping resemblance with the original vehicle to preserve customer
attraction. The motivations, among many others, for a re-engineering process originate from
cost reduction initiatives which involve standardization and reuse of large number of elements
along different models of a car manufacturer (commonality). Not all elements are carried over
from project to project as proportions and dimensions of several areas of the vehicle change
quite considerably. This is the case of the ventilation system which is engineered uniquely for
each vehicle.
The cockpit dashboard, garbled for confidentiality reasons in figure 1 on page 3, covers a wide
area of the car and encompasses elements which are invisible to the passenger such as airbags,
crossbar, instruments, etc. The front ventilation system is positioned inside the dashboard
having four outlets as the only visible clues of its existence. Air flow coming from the lower
heater-conditioner group is channeled up into four ducts and directed to each of the four outlets.
The same dashboard seen from above and with the upper protection elements removed,
uncovers the ventilation system (see figure 2 on page 3) which appears heavily constrained into
the available space. The system is positioned in a densely populated area of elements, where
other components rigidly define the path of the ducts and provide small margins of freedom.
This is a major factor involved in the optimization process as the feasibility of solutions is
strictly related to the geometrical clearance of the boundaries.
Scope of study 3
figure 1 - Dashboard front
figure 2 - Dashboard upper
4I N T R O D U C T I O N
When observed isolated from its designated position in the dashboard, the front ventilation
appears as in figure 3 on page 4.
The system is composed of several parts including a mixer interface to the air supply, a duct
adapter, four air ducts and their respective outlets. A study for optimal air flow should include
by nature the whole system as depicted in figure 3 on page 4. However, to avoid the complexity
of the parameters involved or the necessary trade-offs to simplify the problem and reduce the
degrees of freedom, decision has been made to focus on the sole right ventilation duct (see
figure 4 on page 4).
figure 3 - Vent system
figure 4 - Right vent system
Objectives 5
Hopefully, a successful optimization of a portion of the entire system can allow for extension to
further areas encompassing the left duct and the front pipes or eventually the whole system. That
is, considering that the process has proved its validity and perceived as useful.
Adopting a real world design problem offers the advantage of having project geometries both
as a starting point and for constraints definition. Furthermore, comparison of the obtained
results with the original design is also possible. 
Nonetheless, instead of optimizing the existing ventilation duct, we started with a simplified
geometrical model. This approach is justified by the fact that we are considering the early
phases of a project where definitive geometries aren’t yet defined. Furthermore, our mission is
to provide feasibility and design indications, not yet a fully defined ventilation duct.
Consequently, real project geometries have been limited to boundaries control (contact or
interferences with the surrounding elements) and geometrical interfaces (inlet and outlet
shapes).
1.3 Objectives
The objective of the study is to apply optimization techniques to the right duct of the front
ventilation system which is supposed to provide sufficient air flow through the outlet, by
minimizing the pressure loss and keeping the noise levels to an acceptable minimum.
In a more general formulation the optimization task can be defined as a problem where
design variables are altered with the aim of reaching objectives while satisfying necessary
constraints. Schematically:
6I N T R O D U C T I O N
Design variables involve the geometrical parameters which ultimately define the duct shape,
while the objectives include the principles we seek to optimize. Constraints define the
restrictions of the problem, in our case the geometrical bounds and extremes. 
The objectives, as defined here, have been translated later in the work in favor of geometrical
principles. The new objectives and the motivations for such decision are detailed extensively in
the study. Correlation with the original objectives will be provided in order to prove that they
still preserve overall validity.
In broader terms, the tasks of the study were:
• Develop a simplified geometrical model of the right ventilation duct;
• Determine a valid approach to duct optimization using our model;
• Define an optimization process to achieve the objectives;
• Verify the effectiveness of the process.
We needed to define both an approach to optimization and the optimization process itself. The
approach defines which techniques and whatever combination of them is needed. The process
is the effective automatic mechanism to handle numeric values and interaction between
software modules to achieve optimization.
1.4 Operative approach
Operatively, we have followed an incremental approach to face the tasks of our study
progressively. Improvements are introduced gradually and issues are solved while moving
ahead.
Given the objectives of this work, the following steps were taken:
• Collection of necessary geometrical information;
• Development of the simplified model;
• Formulation of the problem we were seeking to optimize;
• Identification of the optimization steps;
• Definition of the optimization process;
• Execution of the process and verification of the results.
Each phase involved analysis of the data collected and judgment the results obtained before any
progress was made into further steps.
Several commercial products have been used to implement the optimization process. Among
these, a CAD platform from Dassault Systems (CATIA), a process automation and optimization
software from Simulia (Isight) and a computational fluid dynamics simulation package from
Cd-Adapco (StarCCM+).
Operative approach 7
The relation between the software modules is outlined by the following scheme:
Direct interaction occurs only between the optimization tool (Isight) and the geometrical
module (CATIA). The former acts as the operational unit exchanging information with the
latter. The geometry itself is an input to both the geometrical and the simulation module. 
At close observation the absence of the CFD module from any interaction with other tools
could sound alarming as both objectives defined above (minimize pressure loss and noise level)
belong to the fluid dynamic domain. As a matter of fact, CFD was excluded from the
optimization process and the objectives were changed from fluid dynamic to geometrical as we
will see later. As such, the fluid dynamic simulation package (StarCCM+) was used as a
validation tool only.
The reason for this choice is directly related to the computationally expensive resources
needed for three dimensional CFD simulations. Performing optimization, which necessarily
requires iterations, with the CFD module can hinder any progress for all but the most simpler
problems. For medium sized, averagely complex geometries, a single CFD simulation can take
up to hours in order to complete. As we aim to implement an automatic process with multiple
evaluations for each generated solution, even with few iterations the time spent waiting for
results can rapidly rise to an unacceptable amount.