On Adaptive System Diagnosis

In the theory of t-fault-diagnosable systems, one first chooses a set of diagnostic tests, then seeks the results of these tests, and finally proceeds to use the test results to identify the faulty units assuming that the number of faulty units does not exceed t. Nakajima was the first to suggest a departure from this practice. He proposed to adaptively choose the tests and to seek their results until one can identify a fault-free unit. This fault-free unit may then be used as a tester to identify all faulty units. In this paper, we exploit this idea fully and show that one needs the results of at most (n + 2t −2) adaptive tests to identify all faulty units in a t-fault-diagnosable system with n units. The impact of the applications of this idea to the various models and diagnosis algorithms is examined.