Engineering design programs normally
tackle only one problem at a time - a significant drawback when most
real-world design problems involve juggling with many different factors
such as cost, weight, strength and noise. OPTIM is a successful attempt
to link several design programs and automate their operation, allowing
design engineers to concentrate on the more creative parts of the
OPTIM, a collaboration between two SMEs, two university departments
and a large aerospace manufacturer, developed the world's first commercial
concurrent optimisation system for engineering design. The partners
estimate that they have a two-year lead in this technology and expect
to be selling licences by early 1997.
Engineering design is a process of synthesis.
An engineer planning a new road bridge, for example, must take into
account a whole range of interrelated factors such as the bridge's
deflection under different loads, its resistance to earthquakes,
the building cost and the noise produced by traffic travelling across
it. Unfortunately, juggling all these factors at once can be time-consuming
and the engineer may not even be sure that the final design will
meet all the constraints imposed on it.
The future of engineering design lies in optimisation, a computer
technique that allows engineers to be confident that they have found
the best possible solution to a particular aspect of the design.
By optimising simultaneously for cost, weight, strength and all
the other factors that are important to the final design, the engineer
can make sure that nothing has been overlooked and that the finished
product is indeed the best for the job.
The partners in OPTIM are world leaders in the difficult techniques
of concurrent optimisation. They have developed a saleable product
that has already created great interest in the aerospace industry
and has a promising future in the design of cars, ships and offshore
Closing the loop
Computers can help with many design tasks, yet typical engineering
design software takes a narrow, analytical approach that is a long
way from the broad view that characterises the best human designers.
A structural analysis program, for instance, can calculate the deflection
of a bridge but takes no account of costs, road noise or other important
factors. Even in its own area of expertise, the program is limited:
it can calculate the deflection for a given bridge design, but it
cannot design a bridge to match a given deflection.
So if the deflection turns out to be too great, the structural engineer
has to adjust the design and run the analysis program again. After
a process of trial and error a satisfactory structural design emerges
- but by this time the bridge is heavier than originally planned,
which upsets the calculations of the costing specialist in the next
office. And so the process continues.
The lead partner in OPTIM is SIREHNA, a French company specialising
in technology transfer between research and industrial applications
in the area of fluid mechanics. SIREHNA recognised that there was
an opportunity to use the latest optimisation technology to link
all the different analysis programs used in a typical engineering
design. Closing the loop by reducing the need for human input to
the repetitive parts of the design process would leave engineers
free to do what they do best: specifying the overall design envelope
and coming up with truly innovative solutions.
The OPTIM project, which started in December 1992, brought together
SIREHNA, aircraft manufacturer British Aerospace, Belgian structural
analysis software supplier SAMTECH, and two universities. The aerospace
laboratory at the University of Liège in Belgium specialises
in optimisation of finite-element systems and was in charge of the
algorithms for local optimisation. Eindhoven University of Technology
in the Netherlands, which has good links with industry, looked after
If it isn't broken, don't fix it
Throughout the project the aim wherever possible was to allow
engineers to continue working with their existing analysis software.
The core optimisation program, a SAMTECH product known as BOSS/Quattro,
does not attempt to take over the work of the analytical packages;
instead it simply feeds them with the details of the design and
waits for them to calculate their results. Depending on the answers,
BOSS/Quattro then adjusts the design and sends the new details back
to the analysis packages, a process it repeats until it is satisfied
that it has the best design that meets all the engineers' constraints.
BOSS/Quattro communicates with the various analysis programs through
a series of purpose-written translation programs or drivers. Because
the drivers are small and easy to write, BOSS/Quattro can be adapted
to work with almost any existing or future analysis program. The
design programs themselves do not need to be altered in any way,
so engineers can continue to work with tried and tested software,
including programs they have developed themselves.
The algorithms used for the optimisation can be varied to suit the
job in hand, or updated as mathematicians develop new optimisation
techniques. A good example of this is simulated annealing, an optimisation
technique based on stochastic (random) processes. At present, simulated
annealing is not widely used with time-consuming analysis programmes
because it needs large amounts of computing power, but for the future
it promises to increase designers' confidence that they have indeed
found a true global optimum.
Think globally, think locally
The optimisation process starts with the objective function and
constraints, a mathematical statement of the various factors that
are important in the design (strength, weight, cost and so on),
and the importance and limits attached to each. Next, the program
uses a technique known as sensitivity analysis to find out which
of the parameters used to define the design is the most important.
Finally comes the optimisation itself, which falls into two types:
local and global.
Local optimisation tries to locate 'valleys' in a multi-dimensional
plot of the objective function by finding the steepest slopes and
following them downhill until it can go no further. Local optimisers
are quick and efficient, but cannot be relied on to find the deepest
valleys because they may become trapped at the bottom of other,
smaller, valleys. Global optimisation is like flying over the terrain
in a helicopter: it is easier to find the deepest valleys but the
process is time-consuming and needs powerful computers.
First aircraft, then the world
The aerospace industry has been optimising individual design factors,
such as the airflow over a wing, for around five years. Aircraft
designers are now starting to look seriously at optimising several
factors at once, so aerospace was a natural place to start the OPTIM
project. By the time the project finished in November 1995, the
partners had developed a new approach to the design of transonic
aerofoils, optimising the wing shape for drag but also taking into
account structural constraints in a way that was simply not possible
with previous techniques.
The OPTIM system is also more flexible and cheaper than the previous
in-house systems. Since the commercial launch of the system in June
1996 many aerospace companies have shown serious interest.
Naval architects have only recently begun to look at optimising
the shapes of ships' hulls to minimise drag, so it will be several
years before concurrent optimisation becomes widely used. Nevertheless,
the partners expect to be selling the system within six months.
Interest from companies designing offshore structures has also been
strong, and further into the future the software may find applications
in the automotive industry, for instance in coupling aerodynamic
design with acoustics to produce cars that are quieter as well as