Multicopters and air cabs: "Virtual sensors" to prevent crashes
Scientists have developed a system that can immediately detect and react to faults in air cabs in flight. It works without sensors.
The NFFT system is already installed in the Autonomous Flying Ambulance model drone.
(Bild: Caltech)
If a rotor on a private drone fails and crashes, it costs money. In the case of VTOLs (Vertical Take-Off and Landing), also known as air cabs, such failures can cost lives. Scientists at the California Institute of Technology (Caltech) have developed a type of virtual sensor that is able to detect systemic errors that occur during flight. This works with the help of machine learning and adaptive control methods.
The safety of aircraft is the top priority. In the event of errors and malfunctions such as gusts of wind, crossing birds or failing rotors, they must not only remain in the air, but also be able to continue flying in a controlled manner. A team of scientists at Caltech has developed a control method based on machine learning, which the researchers describe in their paper "Learning-Based Minimally-Sensed Fault-Tolerant Adaptive Flight Control", published in IEEE Robotics and Automation Letters. The researchers call the system Neural-Fly for Fault Tolerance (NFFT).
"We have developed such a fault-tolerant system, which is crucial for safety-critical autonomous systems, and it introduces the idea of virtual sensors for detecting faults using machine learning and adaptive control methods," says Soon-Jo Chung, Professor of Control and Dynamical Systems at Caltech.
Empfohlener redaktioneller Inhalt
Mit Ihrer Zustimmmung wird hier ein externes YouTube-Video (Google Ireland Limited) geladen.
Ich bin damit einverstanden, dass mir externe Inhalte angezeigt werden. Damit können personenbezogene Daten an Drittplattformen (Google Ireland Limited) übermittelt werden. Mehr dazu in unserer Datenschutzerklärung.
The electrically powered VTOLs are equipped with several rotors, partly for redundancy reasons. If one of the rotors fails, the remaining ones can take over and compensate for its task. This keeps the VTOL in the air. To keep energy requirements as low as possible, most VTOLs also have wings. However, both rotors and wings offer plenty of potential for failure.
In an emergency, it must therefore be possible to detect very quickly if something is wrong with the aircraft. Sensors could be installed in each rotor for this purpose. But that is not enough, say the scientists. For example, a sensor would be needed to detect faults in the rotor structure, one to determine that the rotor is no longer running and another to detect a signal error. Such systems are expensive and difficult to control. It would also increase the weight of the aircraft. What's more, the sensors themselves could also fail.
Saving sensors through machine learning
The Caltech researchers' NFFT is based on a deep learning method. It can react to strong winds, for example, and at the same time determine whether a fault has occurred on board in any way. The on-board neural network is trained using real flight data. It learns in real time when individual parameters change and can then react accordingly. For example, the system can estimate how well each individual rotor is working at any given time.
Additional sensors are not required for this. Engineers can also save themselves the hardware for error detection and identification, Chung promises.
"We simply observe the behavior of the aircraft - its attitude and position over time. If the aircraft deviates from its desired position from point A to point B, NFFT can detect that something is wrong and use the information to compensate for the error."
These corrections are made very quickly in under a second. If a motor failed, you would hardly notice that it had failed. That's how quickly the system can compensate for the fault.
The researchers have tested their NFFT system on various small multicopters, including a model of an Autonomous Flying Ambulance, a hybrid electric vehicle and aircraft designed to transport injured people quickly to hospitals. The researchers have also tested the system on pure ground vehicles and want to extend the application of the NFFT to boats.
(olb)