Projecte llegit
Títol: Probabilistic modelling of drone operations for collision safety assessments
Estudiants que han llegit aquest projecte:
VALERO GIL, MARIO (data lectura: 30-10-2026)- Cerca aquest projecte a Bibliotècnica
VALERO GIL, MARIO (data lectura: 30-10-2026)Director/a: PASTOR LLORENS, ENRIC
Departament: DAC
Títol: Probabilistic modelling of drone operations for collision safety assessments
Data inici oferta: 04-02-2026 Data finalització oferta: 04-10-2026
Estudis d'assignació del projecte:
GR ENG SIS TELECOMUN
GR ENG SIST AEROESP
| Tipus: Individual | |
| Lloc de realització: EETAC | |
| Paraules clau: | |
| Risk models, drones, collision analysis | |
| Descripció del contingut i pla d'activitats: | |
| Development of a probabilistic risk model for drones, computing the dynamic behaviour of such vehicles from real historical data. The models would be tailored to different types of dynamics: flight-plan-based drone operations and manually piloted drone operations.
The models are intended to be employed to create uncorrelated collision models between drones and between drones and manned VFR aircraft. |
|
| Overview (resum en anglès): | |
| An increase in the use of unmanned aircraft in urban and semi-urban environments is expected, making Detect and Avoid (DAA) systems increasingly necessary for beyond visual line of sight (BVLOS) operations. These systems have traditionally been designed and tested using simulations based on statistical models. However, the lack of sufficiently large operational datasets for small drones means that the Drone Encounter Generation (DEG) model relies on expert opinion to populate its statistical tables.
This work aims to populate the DEG model tables using statistical information derived from real-world flight data. The dataset consists of telemetry from drone operations conducted by the Fire Department of the Government of Catalonia between 2023 and 2025. It initially comprises 1,461 flights and was provided by the Universitat Politècnica de Catalunya. The data are first cleaned and filtered, yielding 669 valid flights. The high proportion of telemetry points with zero horizontal velocity is then analysed, and potential errors are identified and discarded. Subsequently, three detectors segment the flights according to horizontal motion, vertical motion, and speed. The resulting segment distributions are stored in variable-width histograms, obtained by reducing the number of bins using Knuth's method. This approach achieves a median reduction factor of 7.7 while almost entirely eliminating bins containing very few data points. The main outcome is a set of probability tables covering all 13 variables of the DEG model. Two additional analyses are conducted to complement the study: the development of a flight-plan detection algorithm, which identifies 198 flights with a flight plan, and an investigation into the conditional probability of a segment type given the flight characteristics of the preceding segment. Flights with a flight plan exhibit differences primarily in segment durations, while segment type is strongly influenced by the characteristics of neighbouring segments. The resulting tables replace expert-based estimates with statistics derived from real operational data. The main limitations of the study are the reliance on a single data source and the limited sample sizes available for flights with a flight plan and for conditional dependencies. Future work should focus on expanding the dataset and validating the tables through the generation of simulated trajectories. |
|