FAILURE DIAGNOSIS SYSTEM USING ARTMAP NEURAL NETWORKS

A novel scheme for diagnosing failures in dynamic systems is presented. After a failure is detected by the state chi-square test, two ARTMAP neural networks are applied to recognize patterns formed by data of the state chi-square test in order to determine when the failure happened, how serious the failure is, and where the failure is located. The efficacy of the proposed scheme is demonstrated by simulation examples where soft failures in an integrated global positioning system/inertial navigation system are successfully diagnosed, An effective approach is also developed to overcome the difficulty of distinguishing gyro failures from accelerometer failures.