No abstract

Analog in-circuit program generator (APG) technology has improved to a state where the quality of tests written by APGs can hardly be met by tests written by humans [Crook 901. One of the best examples of high-quality APGs can be found in Hewlett-Packard's HP 3070 board tester.The primary reason why automatically generated tests are non-optimal is that the input to the APG, the data that describes the board under test, contains inaccuracies. It requires a long debug process to catch these inaccuracies. We propose a new debug methodology for tests, in which the root cause for tests that fail during debug is diagnosed and repaired.In order to aid the test developer and to cut debug time we have developed a debug environment that supports this methodology. An expert system, named ROSALIND, analyses the results of a debug testplan and is able to take additional measurements on the testhead in order to diagnose failing tests.A research prototype has been implemented and tested. It cuts diagnosis time, facilitates higher test qualities and makes the implementation of revision changes to the board under test much easier.