Starting 1984, I worked at India’s National Aerospace Laboratories (NAL) for over two decades. When I look back to the 1980s I can recall one enduring debate that never failed to stir passions: What matters more in aerospace engineering … experiments or computations?
As someone with degrees in statistics and applied mathematics I found myself agreeing more with the group advocating computations. But our group was losing! This wasn’t such a surprise either, given the meager computing resources – one megaflop of computing power was still a dream in the world we then lived in.
We’ve since come a long way. Computations have invaded aerospace engineering – and indeed every domain of human endeavor – in a big, big way. In fact Boeing 777 was designed entirely using computations.
But experiments won’t go away … and shouldn’t. The best engineering decisions require a judicious mix of computations and experiments. This mix will get even more involved and intricate as we go along, and we will one day find ourselves in a situation where the dividing line between the two will practically vanish … with big data acting as that powerful intermediate catalyst.
I guess that day is still some years ago. But aerospace will continue to be a fascinating workshop for analytics and big data applications. In this post I will discuss some promising interactions between aerospace and analytics.
I must begin by lamenting at what I see to be a sad disconnect. Engineers just don’t seem to know enough about what probability and statistics can do! In fact I recall a story (which I hope is not true): NASA once estimated the probability of a shuttle failure to be 1/100,000, when it was actually 1/100. Why? Because they thought that P(AUB) was equal to P(A). P(B)!
And then there’s this story that is indeed true: In 1999 the Titan IVB rocket launcher failed to put a billion dollar military communications satellite into orbit because a flight engineer entered one parameter wrongly. He entered -0.1992476 instead of the correct -1.992476.
One place where analytics and aeronautics can have a very happy marriage is in wind tunnel testing. In a wind tunnel we mimic the flying environment by placing a scaled model of an aircraft, or a rocket, inside an enclosed chamber and subjecting it to a calibrated flow of air. The model is instrumented to capture data – that allow us to measure the loads experienced by the model as it faces this barrage of air. This process is expensive; and not entirely accurate because the sensors and probes on the model tend to affect air flow, and therefore data quality.
Today we can do wind tunnel tests more elegantly. We can get rid of the instrumentation on the model, and instead ‘paint’ the model with pressure sensitive paint. As the model faces the air stream it colors up to show different pressure loads at different locations of the model! And, one day, the physical wind tunnel might itself disappear; they are already talking of Java virtual wind tunnels!
Or think of conjoint analysis – a statistical tool often used in marketing exercises to determine consumer preferences. I have often wondered why we haven’t used this tool in aeronautics (perhaps we do, and I don’t know of it!). Some 15 years ago, I strayed into NAL’s simulator facility. I saw a group of Indian Air Force (IAF) pilots testing out the various handling qualities of the proposed light combat aircraft. For every chosen maneuver, the pilots were asked to rate the handling quality. I wondered how they would analyze their responses but I guessed that they weren’t looking beyond simple averages. This bothered me. Surely we could make the feedback from India’s best pilots more informative? Especially because we had to model for the fact that on-ground pilot responses would always be slightly different from in-flight pilot responses … after all, the pilot knows that he won’t crash even if he makes a mistake!
Another of my favorite applications of analytics in aeronautics relates to aircraft wake vortices. Let me start by informally explaining a ‘wake’ and a ‘vortex’. Imagine a big ship sailing in the high seas. When the ship moves forward, it leaves behind a kind of water separation that’s easy to see. That’s the ‘wake’. Now imagine that you are in front of your wash basin. Close the basin cork and open the tap to let the water fill up. When the basin is practically full … uncork! Water will start draining, and soon you will notice that the draining water looks like a conical whirl that makes angry gurgling noises. This whirl is the ‘vortex’.
When a huge aircraft like the Boeing 747 (or a similarly monstrous Russian aircraft) lands, its wake contains a large number of vortices … that you can’t see, but are known to be deadly killers! If a tiny trainer aircraft were to attempt a landing just after the big plane, it would be caught in the big plane’s wake and the powerful vortices within would either cause the plane to come to a thudding crash, or flip and go into an uncontrollable spin. It is widely believed that the first man in space, Yuri Gagarin, died in a plane crash caused by these killer vortices.
The only way to overcome the vortex threat is to maintain a sufficient time between two aircraft landings (or, equivalently, a sufficient separation distance between two successive landing aircraft). But that’s exactly where busy airports feel the pinch – every landing brings in money and an unnecessarily large separation distance diminishes revenue! The analytical challenge is to optimize this separation distance using a suitable wake vortex model. In tomorrow’s world of big data, one expects the separation distances to be controlled … in the clouds via cloud computing!
Let us next talk about metal fatigue; most laymen don’t understand the expression (a familiar reaction is: “We understand ‘human fatigue’, but what’s ‘metal fatigue’?”). I answer this question by talking of a metallic clip (also called a gem clip; it is now the icon to attach files in computer software). I ask my audience to open out the metallic clip and try to break it. Soon everyone discovers that there’s only one way to break the clip: cyclically bend the clip one way and then the other, and repeat the process 3-4 times till the clip snaps.
Now imagine an aircraft flying, with either wing attached to the fuselage (the plane’s ‘body’). The plane is continually experiencing a cyclic force that makes the wing keep going very slightly up and then very slightly down. One day, just like the gem clip, the wing will snap leading to a catastrophic crash. We therefore need to find a way to measure the fatigue life of an aircraft wing.
The classical engineering way to measure fatigue life is to hold the aircraft wing on a rig, stick actuators at different locations of the wing and apply loads that the wings would have experienced in actual service life. At a certain point the wing would crack … that’s the estimate of the true fatigue life.
Back in the 1970s, the IAF had a curious problem. All IAF fighter jets were of Russian (then, ‘Soviet’) make; most were versions of the MiG aircraft. The Russians aircraft manufacturers said that their rated service life was 2400 hours. This meant that most squadrons would have to retire their aircraft in 1975 (year fictitious). Now imagine that the delivery of the next generation MiGs was to happen a year later in 1976. How could IAF ensure military preparedness for that one year, since it would be unduly expensive to buy replacements just for a year? NAL ran accelerated fatigue life tests for IAF and came up with the reassuring verdict that the true service life could be as high as 4000 hours.
How would we handle the fatigue life question in 2013? I’m guessing we’d do it far more elegantly. We would embed fatigue crack monitoring sensors at locations on the wing known to experience the toughest loads … and every time the aircraft is airborne we would be obtaining live data of the fatigue crack propagation data.
In a FBW aircraft there is a computer introduced between the pilot and the control surface. So, if the pilot wants to yaw sharply to the left, his ‘request’ is first submitted to the computer, and it is the computer that tells the rudder to carry out the maneuver.
FBW technology makes sense because it allows the pilot to delegate many of his more routine or mundane tasks to the computer while being free to manage the more important task of safely flying the aircraft. For example, landing in poor weather or in zero visibility becomes a lot simpler.
As readers can guess, the FBW technology first originated in the design of fighter aircraft. A fighter aircraft is as different from a passenger aircraft as chalk is from cheese. A fighter aircraft is all about maneuverability, speed and control, while a passenger aircraft is really about stability, safety and comfort. When the pilot just has a few seconds to successfully complete his mission, the computer on board can give him that crucial winning advantage.
But why would Airbus want to bring in FBW technology in passenger aircraft? The decision was both a surprise and a worry. Airbus’s rival Boeing reacted in the early days by saying it wouldn’t attempt FBW technology in civil aircraft. The introduction of Airbus A320 proved to be indeed difficult. There was an early crash in France, and then the more tragic crash in India.
One happy consequence of FBW technology was that civil aircraft started churning out much more data; something that you would expect when you insert a computer on-board. This raw data of course required mediation, processing and management; but it also made it so much easier to measure flying performance, analyze accidents and design better aircraft. Today big data is poised to change the future of aviation.
Here’s an interesting Airbus A320 story to explain how things began changing. Many years ago an impatient pilot was waiting to take off his Airbus A320 aircraft from the old airport in Hyderabad, India, but the traffic controller kept asking him to wait. The angry and agitated pilot wanted to know why.
“I see a flock of gulls at the other end of the runway, and you must wait till they fly away”, the traffic controller told him. The pilot looked as far out as he could see, but saw no gulls. So he decided to take off anyway.
What the pilot briefly forgot was that there was a slight hump in the middle of the runway that obstructed his visibility. He started his run, speeded up, attained a speed of 110 knots (after which you must take off) and then was horrified to find that there was indeed a flock of gulls.
This was big trouble, and the pilot had to make a premature take-off before reaching the prescribed take-off speed of about 140 knots. Fortunately, the plane took off safely avoiding bird hits, but its tail brushed ever so slightly with the runway in the process.
When a maintenance engineer asked why the tail was damaged, the pilot pretended that he didn’t know. But a simple plot mapping altitude to take-off speed nailed his lie: data from the onboard computer successfully established that the plane took off well before attaining the required take-off speed.
Modern aircraft are now so safe that pilots can afford to be careless, and they often are. Most passengers recognize a hard landing, but there are a dozen other ways in which pilots abuse aircraft. This abuse is not immediately noticeable but it eventually begins to show up with a higher frequency of maintenance visits. This, in turn, translates to significantly higher insurance costs.
Indeed this used to be a major dilemma for airline managers: how do they tell their prima donna pilots that they are flying their planes badly? Big data and stylish 3D visualization and animation now make this possible. Every flight path (especially during landing and take-off) can be visualized, and the pilot’s performance can be continually monitored. These visuals are so splendidly ‘textured’, and have such good refresh rates, that they are almost perfect replicas of the actual flying path.
I believe that big data analytics will also completely change our ideas and processes involved in aircraft maintenance and aircraft accident investigation. Even today aircraft maintenance processes tend to be very structured. There are rules that say that a certain part must be checked after ‘x’ flying hours, and another after ‘y’ flying hours. Why ‘x’? Does this ‘x’ depend on the age of the aircraft, or on the terrain in which the aircraft flies? How did we determine the value of ‘x’ in the first place?
I suspect that a lot of these parametric values are coming out of ancient flying manuals, or from old models that determined a certain probability distribution for part failure. With the influx of big data we can play the game very differently. We could, for instance, determine maintenance schedules based on simulated sample flight paths and correlations that seem to recur.
Aircraft accident investigation too continues to happen in the traditional Sherlock Holmes style: try to put together possible hypotheses based on recorded cockpit conversations, correlate these hypotheses with the data recorded in the digital flight data recorder (sometimes called a ‘black box’ although it is actually bright orange in color), and undertake metallurgical investigation of the wreckage. The mind goes back to the Air India Kanishka crash, or the probe into the attempted Makalu sabotage caused by a disgruntled left-handed employee of Air India. Big data analytics now offers the opportunity to move from the qualitative to the quantitative: e.g., image processing of damaged surfaces and creation of big event databases.
There’s much more we can talk of. Oren Etzioni’s big data analysis of aircraft ticket pricing (described in the admirable new book on Big Data by Viktor Mayer-Schoenberger and Kenneth Cukier), the emerging idea of ‘free flight’ (instead of flying only in well-defined air ‘corridors’) that will suddenly create an immense new expanse of flying space in the skies and significantly reduce fuel costs, variants of ticket pricing models, etc., etc.
Let me end off by a comment made by Capt G R Gopinath, one of India’s aviation entrepreneurs: “It is all rather simple. Just make sure that your planes are flying in the skies. Planes earn money when they fly, but lose money when they are on ground!”