As technology has improved, the use of drones has become commonplace, operating in a variety of domains from defence to environmental conservation.  Among this are ‘wingman’ drones that are controlled by the pilot of an escorting vehicle. Their operational environment is often extreme and hence malfunctions can occur mid-flight which can vary from minor deviations in flight path to catastrophic failures. Dealing with these errors can lead to pilot overload and reduced situational awareness, especially within a complex domain such as congested or contested airspace. How does one predict the emergence of these faults and take action to mitigate them mid-flight autonomously?

Anomaly detection techniques allow for the identification of abnormal patterns within data and has been used for predictive maintenance. Generally, these techniques are trained on an ideal flight and sensor datasets which can be quite difficult to obtain. With the advent of the 4th industrial revolution, the boundaries between the physical and digital world have become blurred with technologies such as digital twins which can represent a physical system digitally.

This paper investigates using a drone digital twin and anomaly detection concurrently to predict and mitigate in-flight drone malfunctions. This will be demonstrated by the creation of a training dataset for anomaly detection techniques using a digital twin, the comparison of new anomalous data with past malfunction patterns and the use of the digital twin to inform mitigation strategies, both in training and operationally.

Keywords
AI,DATA,SYNTHETIC ENVIRONMENT

Additional Keywords
Anomaly Detection, Digital Twin