An Efficient Approach for Graph-Based Fault Diagnosis in UAVs

In this work, we tackle the problem of systematic population of a bank of residual generators for model-based fault diagnosis in Unmanned Aerial Vehicles (UAVs). Intended for detailed, large and non-linear system models, Structural Analysis (SA) is applied to produce a graph-based abstraction of the problem in the form of a bipartite graph. The Branch and Bound Integer Linear Programming (BBILP) algorithm is employed, properly adapted to seek a solution for the constrained graph matching problem. Appropriate causality constraints are formulated, which link the structure of the system graph with the analytical form of the residual generators and certify that all resulting residual generators can be implemented automatically using numerical processes. An extensive performance investigation of the proposed approach is carried out, which is shown to be more efficient than other similar algorithms. Benchmarks of UAV models taken from the literature are presented and a simulated response of the diagnostic system against a fault in the roll-rate sensor is showcased.