Generation of Artificial Immune System Antibodies Using Raw Data and Cluster Set Union

Within the artificial immune system paradigm for system abnormal condition detection, two approaches can be used for the generation of antibodies/detectors from multiple sets of experimental data defining the normal system operation, or “self”. The raw data set union-based approach consists of collecting all experimental data in one file, which is then used for antibody generation. The cluster set union-based approach consists of clustering each set of experimental data, collecting all the clusters in one set, and then generating the antibodies. In this paper, the two approaches are described, discussed, and their advantages and disadvantages analyzed based on examples obtained during the development of a comprehensive artificial immune system for aircraft sub-system abnormal condition detection. Data from a motion based six degrees-of-freedom flight simulator are used. The detection performance in terms of percentage detection rate and false alarms of the two sets of detectors has been compared for a sub-set of relevant projections under several types of failure and resulted, in general, to be similar. The raw data set union method requires large computer memory, but the total computation time is lower. The cluster set union method can be implemented on lower memory computers; however, the overall computation time increases, unless parallel computation is used.