This post is not an article summary but an observation I have made on the challenges DJI may be facing in bringing a new product to market (Phantom 3) perhaps at a demand rate it was not prepared for. As you will see in my analysis below, statistically, now that drones are being produced at higher quantities, could we have 20 Phantom 3s fall out of the sky this year, with potentially no warning? It may be more likely than you think.
To be clear I am not making any claims that DJI or any other drone manufacture are having significant issues, but I understand there are significant challenges when it comes to producing items in large quantities (millions) and that even very small margins of error can have significant consequences.
It could be happening now
There is a thread on the forum PhantomPilots.com in which a pilot of a new Phantom 3 Pro discusses and shares a video on how his drone suddenly went out of control and fell out of the sky. The thread goes on for 10 pages and can be found here http://www.phantompilots.com/threads/phantom-3-malfunction-and-crash-flight-record-video.42904/. There is some debate on whether the issue was caused by pilot error, a collision with something, or a hardware failure. In the end, the conclusion the group makes is the crash was the result of a motor failure and it seems like a plausible explanation. The pilot had experienced a “Jello” affect with his video several flights before the crash that became progressively worse. The “Jello” affect is an indication of extreme vibration or turbulent winds that the gimbal cannot compensate for. Another pilot posted that he had a “Jello” issue and after testing his motors with a tachometer it was determined he had a bearing failure in one of his motors that has now been replaced by DJI. If you put those two data points together it seems reasonable to believe the pilot who originally started the thread had a failing motor that eventually gave out.
The statistical analysis
This thread got me thinking about my days as an operations manager at Lockheed Martin and what I learned about Statistical Process Control in regards to manufacturing. The goal for major manufactures who put a strong emphasis on quality or lean manufacturing is to develop processes that perform at six sigma or better. What that means is statistically you can produce parts/features within a normal distribution that are +/- 6 standard deviations (sigma) within spec. What that boils down to (factoring in sigma shift) is that your 6sigma process will yield a success rate of 99.99966% or 3.4 defects per million opportunities (DPMO). Now that seems pretty good, but hypothetically, if you figure there are 12 critical components on a Phantom 3 (4 motors, 4 props, 4 electrical components) and if DJI produces say 500,000 Phantom 3 drones this year with a respectable 6sigma process, it could still create a potential for 20 (3.4 / 2 x 12 = ~20) brand new Phantom 3s to fail and fall out of the sky potentially without warning. Although that is a relatively small number it is still scary and could lead to someone getting seriously hurt.
An even scarier thought is that what if a company like DJI who is producing a new drone product, is still coming up the manufacturing learning curve and has processes that are only operating at +/- 3 sigma. In theory, this is certainly reasonable and you couldn’t fault a manufacturer for only being at that level in that short amount of time. That process on average would result in a success rate of 93.3% or 66,807 DPMO and if you apply that to the example of 500,000 Phantom 3s above, you are now looking at a potential of 400,842 (66.8K / 2 X 12) drones falling out of the sky without warning. Now that seems highly unlikely and suggests that a company like DJI would certainly make sure they have a very robust and high performing manufacturing process. But this example illustrates how important quality is and how sensitive the performance of the manufacturing process is to safety of drones in large quantities (millions).
What is the solution?
I think fail safes like automatic parachute systems will need to become a requirement in the future. I think drone companies need to implement more self diagnostics into their systems. For military systems they were called Built-in-Test or BIT and were used to check the health of the system before each use. Also, I think the FAA needs to consider implementing a validation system that can verify drone manufactures are producing with high performing and controlled processes before they are able to sell their product in the US in large quantities.
What’s your take?