00038 ‰969714d1d462123041726de905b1d7

37

A Rule-Based Approach to Multiple Statistical Test Analysis

Ruleset 3—Based on the test value margins between the initial firing and the fifth additional firing. Since this set of rules measures the changes in the values over the additional runs, the advantage of using a Fast Initial Response (FIR) is alleviated.

Rule 14 - If (CUSUM2_m(+5) - CUSUM2_m(0)) < -0.5

==> False alarm    [30]

Rule 15 - If (SPRT_m(+5) - SPRT_m(0)) < -0.25

==> False alarm    [31]

Rule 16 - If (EWMA_m(+5) - EWMA_m(0)) < -0.6

==> False alarm    [32]

Rule 17 - If (EWMA_m(+5) - EWMA_m(0» > 0.1

=> Detection    [33]

Ruleset 4--Based on the voting on three criterion areas. The areas will cast a vote, either for detection or false alarm. Receiving two or three votes for detection will result in a detection classification. Since all incoming signals are classified in Ruleset 4, the ruleset could also be called the closure set. The three criterion areas are:

==> 1 detection vote	[34]
==> 1 detection vote	[35]
==> 1 detection vote	[36]

Rule 18 - If (CUSUM_m(+5) + CUSUM2_m(+5) + SPRT_m(+5) + EWMA_m(+5)) >2.80 Rule 19 — If sum of all test margins > -0.7 Rule 20 — If Sum of # fires > 1

Simulation Process

The rule-based system previously described is a perfect application for a production system. A production system is comprised of a number of rules that fire based on values of data elements in working memory. Production rules can be written which directly correspond to each of the twenty rules used to classify a signal. For this application, the production system used was OPS5.