Neural Network False Alarm Filter. Volume 1.

Abstract : This effort identified, developed and demonstrated a set of approaches for applying neural network learning techniques to the development of a real-time built-in test (BIT) capability to filter out false-alarms from the BIT output. Following a state-of-the-art assessment, a decision space of 19 neural network models, 9 fault report causes and 12 common groups of BIT techniques was identified. From this space, 4 unique, high-potential combinations were selected for further investigation. These techniques were subsequently simulated for application to a MILSATCOM system. Detailed analyses of their strengths and weaknesses were performed along with cost/ benefit analyses. This study concluded that the best candidates for neural network insert ion are new systems where neural network requirements can be included in the initial system design and that a major challenge is the availability or real data for training of the networks. Volume I of this report documents the activities and findings of the effort, including an extensive, annotated bibliography. Volume II contains a tutorial overview of the neural networks, BIT techniques and false alarm causes utilized in the final phases of this study.