This paper presents in-flight testing results of the implementation of an immunity-based approach for aircraft failure assessment on an unmanned aerial vehicle research system under upset conditions. The bio-inspired approach relies on the fault tolerance capabilities of the artificial immune system and a hierarchical multi self strategy combined with neural networks for feature definition. A low cost hardware and software system has been used to generate nominal and upset flight test data. The system includes an autopilot and sensor fusion solution that uses a non-linear Kalman Filter combined with a complementary filter to estimate the attitude states. Flight data were used to define a large set of 2-dimensional self/non-self projections as well as for the generation of antibodies designated for health assessment of aerial system under upset conditions. The methodology presented in this paper is assessed in terms of detection rates and false alarms.