Failure probability algorithm for test systems to reduce false alarms

The author describes a combination of two techniques for filtering an indicated fault to determine whether it is a true hard failure or a false alarm. The first technique is known as a Bayesian processor and the second is the standard M-out-of-N filter. The filter uses an approach developed previously for handling test and measurement errors. The approach is used for determining the probabilities of indicated failure false alarms and nondetects on the basis of the statistics of both the variable being measured and the test measurement. The combined technique is used as an extension of the M -out-of-N filter. The extended filter determines the probability of failure after each of the M tests, and a threshold can be set for this probability such that the proper value of M will announce the failure. When the accumulated probability goes above a certain threshold, a failure is announced. An approach to establishing the test measurement tolerance limits for various levels of maintenance, based on the relative cost of the two types of measurement errors, is presented. The approach leads to a procedure for minimizing the total cost of the errors associated with tests