xiv

| The Institution of Structural Engineers

Computational engineering

Preface
“A lot [of structural engineers] are acting as human calculators and that’s not engineering. If that is what
people are doing, they will soon be replaced by computers, and that’s a good thing. Because then they will be
free to do what humans do best: complex problem solving; dealing with new phenomena; and being human.”
Chris Wise1

The engineer and the computer

In my first year at university I dutifully attended the maths classes. Although I had been top of my school in maths at
16, I disliked A-level calculus and despaired to learn that I would need to do a lot more of it, both at university and in
my career. It was a relief then, when I went into the subsequent structures class and was told that we could ignore
calculus; instead there were plenty of standard formulas that we would use to calculate the results. This, I thought,
was more like it. Finally, I attended the computing class, where we were told that we did not even need to know the
standard formulae but could just put the structure into the computer and let it do all the calculations. This was the
answer I was looking for!

Of course, as I subsequently learned, these statements were not entirely correct. Rather, like so much in
engineering, they were ‘near enough’. I have subsequently found occasional uses for calculus, and standard formulas
can be incredibly useful. Where would we be without M = wL 2/8 for example: it is the structural engineer’s
equivalent to E = mc 2! Such a simple formula yet it allows us to rapidly analyse not only beams, but also trusses,
arches and catenaries, if you know how to apply it.

But it was to the computer that I kept returning, whether for creating computer-aided drawings or finite element
analysis models, many of which would be extremely difficult, if not impossible, to design with paper and pencil.

I also discovered that computers are not the answer to every problem and, like the oracles of old, their answers need
the most careful scrutiny and thought. They will, usually, give you an answer, but it is for you to ensure that you asked
the right question and received the right answer. The more experienced you are as an engineer the better you become
at judging the results of computer calculations. “The purpose of computing is insight, not numbers” said Richard
Hamming back in 1962 and he was, in many respects, correct, though numbers are still particularly important to the
engineer2. I have learned a lot from building computer models of structures to see how they behave, but usually the
models are of structures that I already understand, so I use them to extract the value of the forces and moments.

As a graduate I focused a lot on engineering computing, so it came as quite a shock when I started preparation for
the IStructE Chartered Membership exam: computers were not allowed. But as my preparations continued, I realised
that this was quite correct. To be a chartered structural engineer you must know the answers before putting pen to
paper, or mouse to computer. Also, you cannot analyse a structure until you’ve designed it, and the design needs to
be decided by you. Computers are an excellent tool for engineers to fine-tune their designs: check exactly what
section sizes are required, and perhaps tweak the geometry to maximise efficiency. They also help when it comes to
revising the design when the architect has, invariably, changed their mind again.

Computer programs are not universal panaceas: they will help a good engineer produce better designs and a bad
engineer produce worse ones. They are only tools to magnify our abilities. There is no substitute for engineering skill,
but the best engineers also make the most of computers.

Workshop manual or driving handbook?

Computers have enabled the modern engineer to create more efficient, more elegant, and more sustainable
structures faster than they ever could have done using the methods of previous generations. While we may know
less now about advanced mathematical and empirical methods than our forebears, who had no choice in the matter,
we instead need to know more about the electronic computer, the programs that run on them, the methods they
use, and how to use them all to best advantage. Among everything else that project work demands, today’s
graduate engineer must also be knowledgeable in computational engineering. So how is this knowledge acquired?
 The Institution of Structural Engineers
Computational engineering | xv

One of the primary tools of the modern computational engineer is finite element analysis (FEA), which allows us to
analyse almost any structure in multiple ways. Most universities and textbooks approach FEA by teaching about its
inner workings in detail but then ignore how to use it. This is like teaching someone how the internal combustion
engine works, then after passing a written exam on the advantages of fuel injection over a carburettor and calculating
torque from a diesel motor, awarding them a driving licence. We would not let anyone drive a car unless they had
proved their competence. Yet junior engineers are often just as inexperienced in ‘driving’ engineering software.

So, what does a driver need to know? Do not put diesel into a petrol engine and vice versa; keep the oil and water
topped up; and get the car serviced regularly. You also need to know which side of the road to drive on, how to
behave at junctions, and how to stop in an emergency. Driving an engineering computer program is just the same.
You need to know how to build good models and avoid bad ones, to let the computer deal with the repetitive and
mundane tasks, and to use the computer to explore efficiencies and possibilities in design that were just not possible
a generation ago. As practicing engineers, we need to have a working knowledge of FEA even if we are not experts,
just as we need to know something about materials, construction methods, fire, architecture, and other problems
that we must deal with on a typical design project.

There are many excellent textbooks available on the detailed workings of FEA, on optimisation methods, on the detail
of the Industry Foundation Classes that allow communication between engineering programs, and so on. These are
all invaluable if you are going to write your own programs or want to deepen your understanding of how those
programs work. If the chapters in this book ignite your interest, then do seek them out. Likewise, talk to the highly
skilled engineers and mathematicians who teach FEA modules and write programs — they’ve spent years studying
engineering computing in detail. Like all engineering experts they are a great resource for those occasions where we
need to dig deeper into a project problem. Meanwhile this guidance will introduce you to the many aspects of
computing for structural engineers, bringing them together to look at the whole design process, with a minimum of
maths but a maximum of explanation and context.

Computational engineering
The structural engineering profession is on the verge of a new generation of technology and innovation: the fourth
industrial revolution. The 1960s and 70s gave us the mainframe and the first finite element analysis; in the 1980s
CAD became mainstream and personal computers (or PCs) replaced the behemoths; mobile phones in the 1990s
lead to the mobile technology revolution and smart phones of the new century; and more recently the industry (finally)
woke up to the communication and workflow possibilities afforded by BIM. Now we are beginning to see artificial
intelligence (AI) and machine learning (ML) applied to everyday engineering tasks.

Artificial intelligence has been an ambition of computing since its very beginnings with Alan Turing’s 1950 paper
Computing Machinery and Intelligence 3. AI has had its setbacks, and while it has not achieved the original goal of
creating a general intelligence, certain aspects have been phenomenally successful. Computing power, especially the
parallelisation possible with distributed cloud computing, is now making optimisation a working proposition for
everyday engineering. 3D printing, robotics, and zero-carbon construction means that very soon material usage will
dominate over labour costs, necessitating the application of optimisation techniques that our mechanical engineering
colleagues have been using for some years. And the computational power of modern engineering computer
programs means that the traditional tasks of sizing steel members and rebar quantities are now regularly automated.

This all means that the time has come for a fresh look at computing for structural engineers and what our role is in
this age of machine learning and automation. What does the graduate of today actually need to know and how can
they compete? The engineers of today must embrace computers as digital engineering assistants if they are not to
replace us. We must use their strengths and know their limitations, let them free us up to do that what computers
can (probably) never do — understand the client’s needs — then use imagination and ingenuity to solve their
problems.

So, what do engineers need to know about these computer programs? As always it is best to start at your goals
and work backwards. Think of it like those puzzles we had as children, where there are several staring points and
some end points, joined by lines resembling a plate of spaghetti. I realised quite young that the way to solve those
puzzles was not to try the start points, but to instead start at the end and work back. Sometimes trial and error can
xvi
| The Institution of Structural Engineers
Computational engineering

be a useful thing though, as we will explore in the chapter on optimisation: answers can sometimes depend on
where you start from. There is rarely only one ‘correct’ answer.

Apart from computing, this book is about design, but that also raises the question: ‘What is design?’ If you ask any
civil engineering student, they might tell you that design is determining the size of the steel beam or the
reinforcement needed in a concrete beam using a design code. But these are both trivial problems. You put the
forces and moments into the formula, turn the handle, and the answer drops out. In the case of a steel beam you
may then need to choose a section that meets the requirements from a table, but little more than that. It is just a
mechanical process with a single answer. Or is it that simple?

It is easy to find the lightest steel section, but is it the cheapest or the one with the lowest environmental impact?
What happens when we take the connections into account, is it still the best, or are we better-off using a larger
section and make the connections simpler? Should we make the section the same as some others to increase
manufacturing efficiency? But clearly making all the beams on the project the same depth, whether they are spanning
1m or 10m, is not going to be efficient — so what is the answer?

If you are just focusing on optimising the section sizes you are missing the bigger picture: should the member even
be there, or should it be somewhere else? You may tune the chord sizes on a truss but varying the depth of the
truss will have a far greater impact on the end result; but did you test that or just use a standard span-to-depth ratio
and make it work? What if you change the number of diagonals in the truss, will that make it better or worse?
Does the bottom or top of the truss really need to be straight, or would a different shape be better? Should it be a
Warren, a Fink, a Prat, or even a Michell truss? We have not seen many Michell trusses so far, but I expect that they
will be far more popular in the future. What if the columns were further apart, or closer together? What if the beams
spanned the other way? What if we change the roof slope or even invert it? Should we use steel, concrete, timber,
or consider an alternative material?

The lightest design is not the cheapest design is not the design with the lowest environmental impact is not the
design with the shortest time on site is not the design with the simplest construction. Usually. Design codes do not
tell us how to design, despite their name, but just give us the basics: strengths, formulas, minimum requirements,
and so on. But how do we find the best design?

“Engineering problems are under-defined, there are many solutions, good, bad and indifferent. The art is to
arrive at a good solution. This is a creative activity, involving imagination, intuition and deliberate choice.”
Ove Arup

Despite our training in maths and physics, where we are taught that there is one answer, design is not so simple as
we may have been led to believe. As some of my colleagues would put it: it is “non-trivial”. Personally, I would say
that the challenge makes design more interesting. How do we produce good designs when there are so many
variables and no single answer? It takes a long time for us to check a single design. We cannot check them all,
but this is where computers can help us.

One advantage of computers is that they are incredibly good at doing the maths for us — the clue is in the name.
Another is that they do the maths very quickly. And a third is that they do not tire or complain about doing the same
thing over and over again. While we, if working with pencil and paper, might choose just one design, we can tell the
computer to test 100, then take the best and from those make 100 more. Or we might ask it to keep adjusting the
position of a connection until it has found the best location. Or we might ask it to examine all the designs that we
have done in the past and suggest what the best answer is likely to be this time.

As a young structural engineering graduate, I started in the industry in the early days of the computer analysis and
drafting revolution. Now we see the same happening to design, just at the time when design, especially the
environmental impact of our designs, is becoming so important. We have the responsibility and we have the tools at
hand to make this world a better place, if we know how to use the tools wisely. Let’s get started.