AI-assisted future overhaul of aviation safety controls

AI-assisted future overhaul of aviation safety controls

The recent spate of accidents from the Boeing 737 Max 8 begs the question of whether our sensor-derived, computer-controlled systems have exceeded the two greatest safety control overrides in aviation—pilot brains and pilot skills/reflexes. Are we on a path to a pilotless commercial aircraft? And if so, will we survive?

As technology advances, our aviation flight control and power systems have also advanced. Advances in aircraft size accommodating ever larger loads have been accompanied by increased aerodynamic loads on the control surfaces, making it impossible for a pilot to control the aircraft directly.

Manufacturers began installing hydraulic-assisted flight controls to give the pilot the necessary muscle to manoeuvre the aircraft. Gradually, these systems moved from being minor assists to becoming ever more central components in the flight control system.

The pilot’s direct “muscle-mechanical linkage-to-control surface” was replaced by this hydraulic middleman, severing the direct pilot-to-control surface connection. Mechanical-hydraulic assisted flight control systems with direct linkages from pilot to the aircraft control surfaces have given way to light-weight computerised systems.

With this, control inputs are directed “by wire” through electronic control signals to a computer which calculates the correct inputs and sends them along to an actuating servo control motor which positions the aircraft control surfaces correctly to meet the pilot’s desired input.

“Fly-by-wire” aircraft systems have several advantages, such as weight savings through removal of mechanical cables, pulleys and hydraulic controls. These systems incorporate multiple safety features to make flying safer for everyone.

System redundancies and pilot override features have been key to modern aircraft safety. Redundancy in hydraulic pumps, power sources and systems reduce the chance of catastrophic failure. A single hydraulic system failure today is more of an abnormal event than an emergency. Complete flight control failure of all systems is extremely rare, if not impossible.

New or just new to us?

Fly-by-wire systems have been part of aviation for over 25 years now. They are standard on nearly all commercial aircraft.

The heart of the digital flight control system architecture is a series of highly sensitive inertial, temperature and air sensors providing feedback to the computer system. The digital flight control system assimilates information from the sensors and provides instantaneous information to safely, smoothly and adroitly control the aircraft.

Unfortunately, sensors sometimes fail. This can lead to erroneous data or no data at all being provided to the computer system. A hard sensor failure in a single sensor can be more easily recognised, and a back-up redundancy sensor activated.  However, a gradual degradation of a sensor group is a far more dangerous to overcome.

Insidious multiple sensor degradation over time may cause sensor errors to cancel each other giving an appearance of normal operations when in reality a serious problem exists. This situation can be amplified when the aircraft is in a critical flight phase such as takeoffs and landings. Add automatic operation of a flight control correction system and disaster can quickly occur.  This appears to be the case with the recent Boeing 737 Max 8.

Lion Air 610 investigators indicated a faulty sensor erroneously provided input indicating a stall as the aircraft nose was too far above the horizon. The false input automatically triggered the Manoeuvring Characteristics Augmentation System (MCAS) to make corrections lowering the aircraft’s nose to gain sufficient airspeed to fly through the stall.

The MCAS receives inputs from two sensors monitoring the plane’s nose relative to oncoming airflow to determine pitch attitude. This system was added as a safety feature to assist pilots in manually flying the aircraft during potential stall conditions. The MCAS system activates without pilot knowledge, sending a signal to the stabiliser to bring the aircraft nose down during steep turns and during flaps up flight at airspeeds approaching stall.

It can be easily but temporarily overridden using the electric trim control switch, or manual trim control, a hand-cranked wheel used by the pilot to neutralise control forces. However, the MCAS system will activate again within five seconds after the trim control switch is released if not shut off with main stabiliser trim cut out switch.

If the MCAS provides an unacceptable nose-down input, it is second nature for the pilot to pull back on the yoke to raise the aircraft’s nose. As the pilot does so, MCAS will continue to provide input further moving the nose position downward. This sets up an upward/downward wave action as the pilot fights against the MCAS inputs, leading to complete loss of aircraft control.

Uncontrolled software, sensor failure or pilot failure? The answer to this question is most likely sensor and pilot failure. The MCAS system is one of many complex systems incorporated in aircraft and dependent upon sensor inputs.

But sensors fail or give incorrect readings. Current research on data cleaning is dedicated to finding rapid methods for determining correct values of data in near real time. Researchers have already developed algorithms capable of detecting gradual degradation of sensors and distributing weighted factors for each sensor output appropriately (proportionate to degradation) to minimise effects of serious conditions.

However, the best immediate solution to the current problem may be as easy as training the pilots about the system and how to rapidly disable it.

New technologies may incorporate advanced artificial intelligence systems and machine learning “trained” to anticipate these failures and retain “soft skills” attributed to the pilot’s vast experience to analyse the situation and take the appropriate action, shutting down erroneous systems.

More research into human behaviours, real-time sensor inputs and the man-and-machine interface is required.

Research in the use of synthesised “on-the-fly” artificial intelligence may provide an answer by “melding” the pilot’s reactions with real-time inputs to the computer to automatically sense correct inputs in manoeuvring the aircraft. Sounds far-fetched? Think how far we’ve already come.

(Iyengar is Director, School of Computing and Information Sciences, Florida International University, Miami; Madni is Distinguished Adjunct Professor and Faculty Fellow, University of California Los Angeles; Miller is Associate Director Robotics and Wireless Systems at Discovery Lab, Florida International University)