Multiple-Failure Inference is used during fault isolation when diagnostics must be accurate when there are mutliple, simultaneous malfunctions at the time that diagnostics are performed. Multiple-Failure isolation attempts to identify an optimal set that is guaranteed to contain a failure. Because of this, there are times when a failed test does not result in additional fault isolation. For example, consider the following functional block diagram:
In this sample, tests T1 and T2 have both failed and the input to function A is assumed to be good. In this application of Multiple-Failure inference, suspicion has been restricted to those functions that could have caused test T1 to have failed (functions A, B & C). The other failed test (T2) does not change the suspect set, since that test may have failed for different reasons (i.e., a fault within either D or E). In this situation, T2 is considered to be a refinement test, since it only changes the suspected set of functions when it passes.
Now, for this same scenario, Multiple-Failure inference could have generated the following alternative diagnosis:
Here, functions A, B, D and E have been called into suspicion since the failuire of test T2 could only have been caused by a failure to one of these functions. T1 is now a refinement test, since it may have failed due to a problem with function C.
A multiple-failure diagnostic strategy must provide a way of evaluating these two alternative inferences to select the optimal isolation. Although the first alternative may have isolated to fewer functions (ABC, as opposed to ABDE), these functions may be associated with more replaceable components (this would be the case, for example, if function C were to be associated with a different component than the other four functions). The diagnostic strategy may also wish to consider replacement cost & time (or other Attributes) when determining which inference to apply.
One disadvantage of Multiple-Fault diagnosis is that it results in larger isolated fault groups than does Common Cause diagnosis. Furthermore, contracted testability requirements are often based on the single-fault assumption and therefore can be satisfied using the more-attractive Fault Isolation metrics that result from Common Cause diagnostics. On the other hand, if the contracted requirements are truly supposed to represent predictions of diagnostic performance in the field, then the Analyst should consider using a Multiple-Fault diagnostic strategy, coupled with prioritized replacement for selected fault groups. This approach will offer accurate fault isolation, yet reduce the impact of multiple fault inference upon isolated ambiguity.
