Nature adapts to faults in a system using various techniques. One example of this is pain, which causes you to walk more carefully after, for example, spraining your ankle. Researchers at Delft University of Technology and Wageningen University & Research want to apply this concept to detect faults in technical systems such as drones and autonomous vehicles at an early stage.
The research, published in the Proceedings of the National Academy of Sciences (PNAS), utilises early-warning indicators based on a phenomenon known as ‘critical slowing down’. This approach could play a key role in enhancing the safety and reliability of modern technology, as problems are often detected before any serious damage actually occurs.
Tipping point approaching
In nature, a critical delay occurs when a system becomes less resilient and takes longer to recover from disturbances. This is often a sign that the system is approaching a tipping point. A healthy forest, for example, can recover quickly after a dry spell. But after successive periods of drought, that recovery takes increasingly longer. Ultimately, a relatively minor extreme climate-related event can lead to large-scale forest dieback.
Scientists can monitor recovery times to determine whether an ecosystem is indeed approaching a critical tipping point. By recognising similar patterns in technical systems, researchers hope to identify problems earlier and prevent potential failures.
Drones, aeroplanes and autonomous robots
The method has long been used in ecology and climate science. Researchers at Delft University of Technology and Wageningen University & Research have now applied the concept to actively controlled systems, such as drones, aeroplanes and autonomous robots. This shows that the early warning signals from ecology also reliably indicate when instability is imminent in controlled systems.
To test the approach, the research team carried out experiments at the CyberZoo. This is a unique drone research facility run by the Faculty of Aerospace Engineering in Delft. Here, scientists can safely test drones to their absolute limits, cause controlled damage to them and thus collect data that helps to understand how faults develop.
The researchers have combined data from simulations, analyses of flight data and extensive experimental tests. This enabled them to determine which combinations of damage, flight conditions and manoeuvres are most likely to lead to a loss of control. This provides a clearer picture of the moments at which a system can still be corrected before a fault becomes irreparable.
Adjusting behaviour in real time
These indicators not only make it possible to detect instability, but also to adapt a system’s behaviour in real time in response to it. For example, to enable a drone to land safely despite damage to a wing.
According to the researchers, a major advantage is that the method does not rely on detailed physical models of a drone. Instead, data from low-cost on-board sensors can be used to detect subtle changes in system behaviour.
Widely applicable
This makes the system applicable to a wide range of technical systems. Possible applications include monitoring critical infrastructure, carrying out predictive maintenance on aircraft and other vehicles, improving quality control during production processes, and increasing the reliability of autonomous systems such as self-driving cars. However, the drone sector is also worth considering, where the technology can help prevent accidents.
Furthermore, this development could contribute to a future in which machines not only respond to faults, but also learn to predict them independently. This will enable technical systems to operate more safely, efficiently and sustainably. Particularly in situations where human control is limited, such as with spacecraft, inspection drones or autonomous vehicles, the early detection of problems can make a significant difference.
