Conceptual Design Part 2: Towards an Initial Design
The parallels between designing an aircraft and drawing an owl are more than you might expect...
Hello again, it's time for more aircraft design! This time, we're diving back into conceptual design once again, looking at the first major stage of the process: initial sizing and creating the first aircraft design. We'll look at what initial sizing is, why it's used, and some of the limitations of the process as it stands.
This post has taken a while to put together, from getting my head around the process to figuring out how best to explain it (moving house and travelling have also played their part in the delay!). I had a ~5,000 word draft written explaining the ins and outs of how initial sizing worked, before realising that I was starting from the wrong point of view. This is still a long one, but hopefully this version is clearer!
I should note at the outset that whilst I've picked up what I know of conceptual design from a number of sources, I'm basing much of my knowledge and the examples here on Daniel P Raymer's Aircraft Design: A Conceptual Approach. There's no one true way to approach this stuff, but Raymer's book is often regarded as the 'bible' of conceptual design, and seems a good starting point from which to set out.
Recap time
Right, let’s start with a quick recap. I’ve covered the main stages of aircraft design before, but broadly, the design process can be thought of as an iterative loop, refining details and getting progressively more detailed over time. Conceptual design sits at the top of that process, and is all about taking the vague idea of an aircraft, working up a bunch of different alternatives, and finally narrowing down to one main configuration to take forward.
Essentially, we figure out what we want our aircraft to do (based on initial requirements – more on that in a bit), and then work up some initial designs that could fit the bill. At the conceptual stage, the process is fast and relatively rough – trading accuracy for speed to allow iterating through a wide variety of concepts (hence the name). We want to understand roughly how each performs, and how they trade-off against one another and versus different requirements – payload, range, speed, etc. Ultimately, one configuration is chosen that best fits the bill, the broad specifications are locked in, and more detailed trade studies and analyses are performed to refine the finer details – beginning the process of preliminary and then detail design.
Initial Requirements
Initial requirements are a bit of a chicken and egg situation. To design an aircraft, we need to know what it should be able to do – aircraft are highly specialised for their design mission – there’re many good reasons an A380 looks different to an F-35 to a Cessna 172! However, the process of designing an aircraft is iterative in terms of requirements as well as design – a company will take potential designs to a customer, who will likely change their requirements and desires based on the opportunities offered by what they see.
So to start the process off, we need some rough idea of what our aircraft should do. We could pick numbers out of the air, but we could then end up with something unachievable. So instead we generally look to existing aircraft in the market we’re looking to enter. If we’re not sure which market exactly, then we may need to generate several sets of requirements, and repeat the process for each.
Our ultimate mission here is to decarbonise the commercial airliner market, but jumping straight to that is an ambitious and likely ill-fated idea, so instead our first market will be one which shares some of the key characteristics of our target, but in smaller airframes which are more approachable as an initial goal. To give a couple of examples:
Light Utility Aircraft
Small, unpressurised passenger or cargo propeller aircraft, such as the Cessna Caravan, Britten-Norman Islander and similar. This is about as simple as commercial aircraft get, but would still involve passenger considerations in a small package, whilst dropping the challenges of pressurisation, speed, range and altitude.
Turboprop Executive Aircraft
This category covers faster pressurised prop aircraft for passengers and cargo, generally using turboprop propulsion. Examples here might be the Beechcraft Super King Air, the Pilatus PC-12 or the Piaggio Avanti. These bridge the gap between the light aircraft above and small jets, travelling at up to Mach 0.5-0.6 or so. They are more complex than their light utility brethren, incorporating pressurisation, longer range and higher speeds, whilst not pushing performance as far as jets.
Light Business Jets
Business jets exist on a huge spectrum, from the tiny VLJs all the way up to world-hopping aircraft that are airliners in all bar name. The smaller end of this range offers perhaps the closest to a commercial airliner at this scale, with aircraft flying at Mach 0.7-0.8 carrying up to 10 passengers. The market is crowded, but there’s plenty of alternatives here, with aircraft broadly following a trend of increasing speed and range as size and cost increase.
Towards an Initial Design
Right, so we’ve sussed out a few initial markets and have a good idea of the major requirements of our aircraft – payload, range, cruise altitude and speed. How do we turn that into something resembling an aircraft that we can start assessing and iterating upon?
The best analogy I can come up with for this process is the well-known ‘How to draw an Owl’ meme. Broadly, the process consists of calculating a couple of key metrics and numbers, and then… designing everything else (based on a host of heuristics and experience). I’ll try and break down the process a bit, then look at how this affects trying to design zero-emissions aircraft, especially those incorporating more novel technologies.
Pinning the tail on the donkey
The most important number for an aircraft at the design stage is the gross take-off weight, unsurprising perhaps given how weight-dominated aircraft are. Once we have an idea of the take-off weight, we then try and roughly figure out two other key parameters – wing loading (the ratio of wing area to weight) and thrust-to-weight ratio. These two parameters determine how the aircraft will perform in various different situations, from setting the stall speed to take-off and landing field lengths and much more. They are trade-offs, and there’s a balance to be had between the two depending on the particular priorities for our aircraft.
These three parameters make sure that the aircraft performs approximately to requirements, but say nothing about the shape, its planform, materials, propulsion, or really any concrete details. So how do we make decisions about all of that? That’s the ‘draw the rest of the f****** owl’ stage. There’s a vast number of decisions to be made, and whilst conceptual design is meant to be high level, choices made here can influence hugely how the aircraft will fly and handle, and can make the difference between later stages of design and manufacture being simple, or involving countless revisions and changes.
Taking a closer look
I’m approaching this as a relative newcomer to the design process and to aerospace engineering as a whole. As such I’m trying to figure things out – understanding how decisions are made, what trade-offs exist, and how to adapt this process to novel aircraft. With this perspective, I’ve a few take-aways from exploring thus far.
The conceptual design process is very much about speed – getting results quickly, so that designs can be analysed and iterated. Aircraft are really complex machines that have to perform many tasks safely in a highly dynamic environment where things can go sideways quickly. As such, figuring out the implications of a decision is tricky – full analysis of many design decisions is hugely complex, and involves a mix of computational techniques and practical testing. Even then, sometimes outcomes only emerge once the plane has been built and is flying! The history of aviation is littered with examples of people trying new things, then discovering some unpredicted side-effect that had catastrophic consequences.
This puts us in a bind (as I’ve mentioned before) – we want to explore different designs rapidly, but to properly understand each design is nigh on impossible without fully building it and flying it. The traditional design process has solved this by leaning on historical convention heavily. If it’s flown, then we have data about how it performed, and can use that as a statistical basis to figure out how a new design might perform.
To illustrate this, let’s take a look at how the initial gross take-off weight (GTOW) calculation is performed. As mentioned at the start, this is based on Raymer, but from what I’ve seen, this approach to design is broadly typical.
The process is iterative – using an initial guess, we estimate the rest of the factors, calculating a new weight value. If our output differs from the guess, we pick a new input and iterate again until guess and calculation converge. This looks something like this:
Make an initial estimate of the GTOW from existing similar aircraft.
We use that GTOW and trend lines of similar aircraft to estimate the empty weight fraction. We can apply modifiers for factors such as the aircraft being composite (making it lighter) or having variable sweep wings (making it heavier).
We then figure out the approximate mission for the aircraft – what flight phases will this involve? For commercial aircraft, this is fairly simple – take-off, climb, cruise, descend, land (with a reserve for safety).
This is broken down by stage, and used to calculate fuel burn. Some stages we’ll simply use guesses based on trends, but cruise and loiter dominate, so we’ll estimate those based on our desired range and loiter time, using the Breguet range equation. However, to do so, we need a couple of things – our specific fuel consumption (SFC) and maximum L/D.
SFC is the rate of fuel consumed per unit thrust per unit time, and broadly correlates to the type of engine we're using. Jets can just use estimates based on previous aircraft, whilst prop aircraft need to also account for propeller efficiency and speed.
Max L/D is where our design comes in. Here we need to estimate the wingspan and the wetted area (the area exposed to airflow), which broadly correlate to the performance we will achieve. Again, much like aircraft empty weight fraction, we use trends of similar classes of aircraft to adjust the values here.
Now we've got all the fuel fractions, we sum them up, and add an extra fudge factor to account for unused fuel – reserves, fuel remaining in the tank that can't be consumed, etc.
Calculate the GTOW, based on the empty weight fraction, fuel fraction, and the requirements for crew and payload.
As you can see, we’re leaning heavily on historical data throughout this process, which makes sense if we’re doing a quick pass – historical data ensures that the outcome should be sound without requiring computational heavy lifting or wind tunnel tests.
However, if you’ve worked with data before, you can possibly spot a bit of an issue – it’s going to be really hard for this type of system to make guesses about anything that differs significantly from what has come before. Unfortunately, zero-emissions propulsion, novel designs and technologies all fit that description rather well.
The rest of the owl
This problem compounds when we start exploring the rest of the design. As I’ve already touched upon, the trade-offs involved in designing aircraft are many and complex – you can’t just slap a wing here and a tail there and call it good. You have to consider stability, weight, aerodynamics, handling, and so much more. Something seemingly minor like the design of a wingtip could affect a host of different areas – lift, structural weight, flutter, roll rate, landing gear length, stall recovery...
As a result of all of these interleaved trade-offs, most design processes rely on skilled designers who have a wealth of theoretical and practical knowledge and experience about what does and doesn’t work in any given situation. This makes sense, but has a few issues. Firstly, it isn’t very automatable or scalable – if we want to test a wider array of designs, we need to speed up the assessment process, which means we want to reduce the role of humans in the loop wherever possible. Secondly, that human judgement and intuition imparts bias. Bias is almost inevitable, but in this case the designer will likely have their own preferences and experiences that will skew their judgements towards certain areas.
A different way
So how do we get past this seeming impasse?
Ultimately, it’s not an easy problem to solve. As I alluded to above, the simplifications in the current process are there for good reason – the underlying systems are incredibly hard to model and simulate, especially rapidly and accurately. However, an approach that might work is trying to unpick each of the decisions being made, and what factors are involved, and any broad heuristics that exist already, and rework those so they’re more tied to the underlying motivations rather than to the elements of a traditional aircraft.
To take an example, let’s look at tail design. Aircraft tails, or empennages, can come in a whole host of different designs, but the majority will use a horizontal and a vertical stabiliser, each with flight surfaces. The conceptual design of these scales them based upon the wing area combined with the lever arm (how far they are from the aerodynamic centre, determining the torque they can exert on the aircraft). This is then adjusted using coefficients based on the type of aircraft we’re designing (sailplane, flying boat, dogfighter, airliner, etc).
But what is the tail for? It performs a host of functions, the most obvious being controlling yaw and pitch (at least in some phases of flight), but also trimming pitch from the wing and the weight distribution of the aircraft, countering forces from high-lift systems like flaps, countering p-factor from props, controlling the aircraft in engine-out scenarios… the list is long.
It’s easy to see why, at least in the early stages, it’s appealing to simplify a lot of this away. However, if we can break down all the main needs for a tail, then we can hopefully devise more general heuristics and rules for each specific function that are still fairly lightweight, but that don’t require a particular tail design (or even a tail at all, potentially!). This could allow for a set of separate changes and new ideas to be stacked, as long as the results from the broader heuristics add up.
This also allows us to figure out where we’re lacking information. After all, we’ve seen how useful the wealth of data from existing aircraft is to intuit how a new design might perform, but we likely lack that data for new technologies. However, if we break down the short-cut rules used now into their component parts, we can start to understand the needs of different technologies, and what practical testing and further study might be needed for them to be a viable option.
The Bigger Picture
This is a truly massive task, but by moving fast and taking advantage of the best of modern computational techniques, there may be a way forward. One example that springs to mind is the approach taken by the latest generation of biotech companies such as Ginkgo or Benevolent AI – these companies are tackling a similarly complex problem, albeit at the micro- rather than the macro-scale. They combine modern ML techniques and search algorithms with rapid real-world testing at scale, combining existing expertise with computation to massively speed up and widen their search.
Translating such an approach to the aerospace world is undoubtedly hard, but offers a way to break past the massive constraints imposed by the industry – the huge capital costs, long timelines, complexity and high regulatory burden – whilst injecting the novelty that’s sorely needed to create truly zero-emissions aircraft. It’s too early to say whether such a radical approach can have the impact required, but that’s what we’re here for – to figure that out!
Thanks for reading, and well done if you made it all the way to the end – this was a long one indeed! I’m hoping next time I can dive into a few more of the details of how this might work, but that may be a while – I’m still very much figuring this out as I go!
As always, I’d love to hear your thoughts – please do reach out if you’ve got questions or would like to discuss this more!
Oli