Multiple Fault: Half-Split Failure Probs. (refinement postponed)
What is the purpose of this algorithm?
The Multiple Fault: Half-Split Failure Probs. (refinement postponed) fault isolation algorithm effects what is essentially a compromise between the size of the generated test sequence and the facility with which that sequence is able to isolate. In order to reduce the total number of nodes in the generated test sequence, this Multiple-Failure algorithm always performs tests that are guaranteed to reduce the size of the suspect set (regardless of whether the test passes or fails) before tests that only reduce the suspect set if that test passes or if it fails (one or the other). What is sacrificed is the fact that some of the refinement tests may be much more easily performed than some of the isolation tests that this algorithm forces to precede them (for example, most signature tests and all operational tests defined at untested terminal output flags would generally be treated as refinement tests). Nevertheless, for many systems, the diagnostic test sequences that would result if refinement tests were allowed to precede isolation tests would be prohibitively large. This algorithm helps reduce the size of that sequence. One of the quirks of this algorithm is that it will perform output flags or net functions as isolation tests before it uses a test set test for refinement (this is the only predefined algorithm that allows a generic test to precede a test set test). This is because the algorithm assumes that the only time that output flags or net functions are included as tests in diagnostics would be when the Analyst is attempting to determine how to improve on the isolation that can be achieved prior to refinement, since isolation tests are more efficient than refinement tests. Remember, when the final diagnostics are generated, all testing should be performed using test set tests. The weightings for this algorithm favor tests that come close to half-splitting the suspect set when they pass or fail.
How was this algorithm implemented?
The full set of test selection criteria for this algorithm are as follows:
1. Test Candidate Grouping 1 of 4: Test Set Isolation
a) Candidate Test Types (1)
b) Weightings: uses algorithm defaults
c) Cutoffs: uses algorithm defaults
2. Test Candidate Grouping 2 of 4: Internal Isolation
a) Candidate Test Types (2)
2) All Input Flags
b) Weightings: uses algorithm defaults
c) Cutoffs: uses algorithm defaults
3. Test Candidate Grouping 3 of 4: Test Set Refinement
a) Candidate Test Types (1)
b) Weightings: uses algorithm defaults
c) Cutoffs: uses algorithm defaults
4. Test Candidate Grouping 4 of 4: Other Refinement
a) Candidate Test Types (2)
b) Weightings: uses algorithm defaults
c) Cutoffs: uses algorithm defaults
1. Test Weighting 1 of 6: Sum Failure Probability
a) Priority: 50
b) Entity: Failure Probability
c) Type: Sum
d) Domain: Suspect Functions Detected
e) Best Equals: Half-Split
2. Test Weighting 2 of 6: Sum Failure Probability
a) Priority: 50
b) Entity: Failure Probability
c) Type: Sum
d) Domain: Suspect Functions Proven
e) Best Equals: Half-Split
3. Test Weighting 3 of 6: Sum Failure Probability
a) Priority: 40
b) Entity: Failure Probability
c) Type: Sum
d) Domain: Suspect Failure Modes Detected
e) Best Equals: Half-Split
4. Test Weighting 4 of 6: Sum Failure Probability
a) Priority: 40
b) Entity: Failure Probability
c) Type: Sum
d) Domain: Suspect Failure Modes Proven
e) Best Equals: Half-Split
5. Test Weighting 5 of 6: Count Number of Items
a) Priority: 20
b) Entity: Number of Items
c) Type: Count
d) Domain: Suspect Functions Detected
e) Best Equals: Half-Split
6. Test Weighting 6 of 6: Count Number of Items
a) Priority: 20
b) Entity: Number of Items
c) Type: Count
d) Domain: Suspect Functions Proven
e) Best Equals: Half-Split
C. Default Test Cutoffs (2)
1. Cutoff 1 of 2: Count Number of Tests
a) Entity: Number of Tests
b) Type: Count
c) Domain: Isolation Path
d) Modifier: Test Usage
e) Target: Refinement
f) Condition: >= 3
g) Action: Ignore in Sequence
2. Cutoff 2 of 2: Count Number of Items
a) Entity: Number of Items
b) Type: Count
c) Domain: Suspected Items
d) Modifier: <none>
e) Target: <none>
f) Condition: <= 1
g) Action: Terminate Sequence
