How can an aircraft keep navigating safely when its sensors or GPS fail?
A fault-tolerant navigation design that spots failing sensors with machine learning and fuses the rest, so pilots keep accurate position, speed and attitude when GPS is lost.
Key findings
- Controlled flight into terrain and runway excursions top the list of navigation-sensor crashes
- The three airline pilots we interviewed wanted earlier warning of failing systems
- In our risk table, an accelerometer bias shift scored highest, with a risk priority of 45
An independent proposal, written in response to a competition brief. The organisation it addresses did not commission this work, and we do not suggest it has acted on our recommendations.
The challenge
Aircraft can lose their navigation data in GPS-denied situations: over oceans, in urban canyons, near jamming, or when a sensor ices over, in every phase from taxiing to approach. Losses like these have contributed to controlled flight into terrain and runway excursions. The brief asked how an aircraft can keep navigating safely when critical instruments fail, which redundancy and fusion techniques give fault-tolerant navigation, and how such a system could be certified.
Our approach
We interviewed three airline pilots, two Airbus A320 first officers and a Boeing 787 captain, and studied accidents in which faulty sensor data played a part, including Air France 447 and Ethiopian Airlines 302. We mapped the sensors each phase of flight relies on, compared them on altitude, attitude and angle of attack and on how they cope with icing, bad weather and poor light, and compared fusion algorithms (Kalman filters, particle filters and neural networks) on accuracy and computing load.
How we found the answer
Every sensor has a weakness, but rarely the same one: a radio altimeter struggles in fog where satellite navigation does not. So the answer is to fuse them, and to catch the faulty one fast. A standard Kalman filter assumes straight-line behaviour that aircraft don't follow, so we chose an error-state extended Kalman filter, with state scaling and sequential processing to keep it stable and light enough to run in real time.
The pilots' main ask, earlier warning when a system misbehaves, shaped how failures are graded and reported.
What we proposed
- Machine-learning error detection: KNN regression predicts what a failed sensor should read, and each failure is graded. Level 0 is removed automatically, level 1 is given less weight in the fusion, and level 2 isolates the sensor and switches to a backup.
- Three navigation systems: GPS and inertial fused in a tightly coupled closed loop by default; a loosely coupled open loop of GPS, inertial, altimeter, angle of attack and airspeed as the backup; and inertial, angle of attack and ILS as the last resort.
- The pilot is told at once, and can override the system or switch to another.
- A failure mode and effects (FMECA) table that ranks each sensor's failures by severity, occurrence and detection, and sets automatic, semi-automatic or manual responses.
- A path to certification under DO-254, DO-178C (Level A) and ARP4754A, with crew training and manuals.
Potential impact
The design aims to keep position, speed and attitude data continuous when sensors or GPS fail, with fewer deviations from the flight path. It is planned over 48 months from research to certification, and its safety would have to be shown in simulation, ground and flight tests.
Modelled in our analysis. These are estimates, not results measured in operation.
Recognition
The full strategy
13 pages, view only. Every page carries the reader's email address as a watermark.
Sign in to read the full strategy
Our decks are shared with readers who sign in with Google. Every page you open carries your email address as a watermark.
When you sign in we record your name and email address (from Google), which deck you open and when, your approximate city and country, and your browser and device. Privacy policy
More of our work
Case studies and other strategy proposals.
Why is NIT Raipur's NIRF rank falling and what would bring it back within three years?
NIT Raipur fell from 64th to 86th in NIRF Engineering between 2021 and 2025, although its score rose. We traced the lost ground to six fixable causes and found seven measures, each already working at a peer NIT worth about 4.2 points.
How can Chhattisgarh’s 3,000 rice mills stay profitable when the paddy runs out mid-year?
Chhattisgarh’s 3,000 rice mills sit at the centre of India’s rice supply, yet earn about ₹500 a tonne, wait months to be paid and run short of paddy by mid-year. We traced where the strain comes from, using public data.
Coffee costs are up 50 to 60%. How can a growing café chain stay profitable while opening about 120 new cafés a year?
Leading Coffee Chain
Get in touch
Tell us what you’re deciding and by when. We’ll reply with a few questions or a time to talk.