What happens if some of the 20 billion sensors projected to be in place in a few years throw out some bad data? If that sees you buying Cheerios instead of Frosted Flakes, I suppose we can all live with that. But what if a faulty sensor kills people?
The two deadly crashes of the new Boeing 737 Max 8 are a case in point. A preliminary report about the causes of the Lion Air crash pointed to erroneous data from a sensor causing the aircraft’s new automated stabilizer system to push the jet’s nose down. The pilots struggled to pull the plane up, and it crashed into the sea. Now it turns out the same faulty sensor may have caused the Ethiopian crash too.
One report faults an AOA sensor, meaning, Angle of Attack which detects the nose of the airplane with respect to the wind. But in another report, an old culprit popped up, the airspeed indicator, which has been cited in quite a few fatal crashes.
When a sensor reports the wrong information, the software that consumes it can do aberrant things. On a personal note, I drive a Ford SVT Raptor truck. A faulty sensor, the constant speed control, caused my transmission to almost catastrophically drop into first gear when I was traveling about sixty miles an hour, causing the rear wheels to lock up and I almost lost control of the vehicle. This symptom caused a recall and Ford flashed a new version of the transmission control system and the problem was resolved.
Unfortunately, these sensor failures don’t always end up so well. There have been many fatal plane crashes due to the failure of something called a pitot-static system. A pitot-static system is a system of pressure-sensitive instruments used to determine an aircraft's airspeed, Mach number, altitude, and altitude trend. Errors in pitot-static system readings can be extremely dangerous.
Several commercial airline incidents and accidents have been traced to a failure of the pitot-static system. NB - I haven’t been able to determine if the AOA and airspeed sensors are one and the same.
- Austral Líneas Aéreas Flight 2553, all 74 passengers and crew killed. According to an investigation by both the Argentine and Uruguayan Air Forces, the pitot tube—the primary instrument for measuring aircraft airspeeds—froze when the aircraft passed through a 15,000-metre (49,000 ft) high cumulonimbus cloud, blocking the instrument and causing it to give a false reading. Compounding this problem was the absence of the alarm designed to report such a malfunction (raising serious questions about inspection irregularities by the Argentine Air Force).
- Birgenair Flight 301 176 passengers and 13 crew killed. Investigations later showed that the plane was actually travelling at 220 knots (410 km/h; 250 mph) at the time of the accident. The investigation concluded that one of the three pitot tubes, used to measure airspeed, was blocked and had a fatal pitot tube failure which investigators suspected was due to insects creating a nest inside the pitot tube; the prime suspect is the black and yellow mud dauber wasp.
- Air France Flight 447 216 passengers and 12 crew killed. The French air safety authority BEA said that pitot tube icing was a contributing factor in the crash of into the Atlantic Ocean.. The BEA's final report concluded that the aircraft crashed after temporary inconsistencies between the airspeed measurements—likely due to the aircraft's pitot tubes being obstructed by ice crystals—caused the autopilot to disconnect, after which the crew reacted incorrectly and ultimately caused the aircraft to enter an aerodynamic stall, from which it did not recover.
But what about the more common sensors that drive the Industrial IOT? Most of these are rather simple devices, and they exhibit a failure curve known as the bathtub curve. In other words, there is a high incidence of failure at the beginning, due to a number of factors such a manufacturing defects, improper installation, improper application. Beyond that, they perform for a long time before they reach a normal lifespan. But there are still failures throughout that lifespan.
There are two ways to know if a sensor is failing. First, in the best case, it will tell you. Not many sensors are currently equipped for this explicitly. The second way is noticing some unexpected results from your models and analytics. In the case of the aircraft, that’s not soon enough. We will research and write more about this shortly, including some innovative tools for dealing for sensor fatigue and failure. In the meantime, the hype about IoT and Edge analytics is extreme, but there is little written about this topic.