Damage classification based on machine learning applications for an unmanned aerial vehicle
Document typeConference lecture
Rights accessRestricted access - publisher's policy
Unmanned aerial vehicles (UAVs) are well-known by its advantages in several applications such as surveillance and monitoring for instance in agricultural applications or fire control among others. These missions can be associated to the robotics area due to the smart applications and tasks that can be performed by these systems autonomously. Although its designs and developments are in most of the cases joined to the applications, currently it is possible to design or acquire a UAV for specific applications by defining features about the task to perform. One of the problems with the use of UAV is attached to the variations of the operational conditions which can produce some damages during operation, landing and de-landing tasks. Since several damages can affect the structural state of these vehicles, the use of a Structural Health Monitoring system is a necessity to provide an automatic monitoring system. This work includes a description of a preliminary damage detection and classification system for a UAV. The system includes the description of the data analysis from a piezoelectric sensor network with independent component analysis and machine learning approaches. Some tests are available to validate the system with data from a wing of the UAV called VANT Solvendus from the Fundación Universitaria Los Libertadores. Tests and the application of the methodology for detecting and classifying damage are performed to a part of the UAV wing skin and results show the advantage of the methodology.
CitationAnaya, M., Ceron, H., Vitola, J., Tibaduiza, D.A., Pozo, F. Damage classification based on machine learning applications for an unmanned aerial vehicle. A: International Workshop on Structural Health Monitoring. "IWSHM 2017: 11th International Workshop on Structural Health Monitoring: Stanford, California: September 12-14, 2017: proceedings book". Stanford: 2017, p. 2042-2049.
|2042.pdf||Article principal||10,11Mb||Restricted access|