Voting Techniques for Fault Diagnosis from Frequency-Domain Test-Data

The basic voting technique, which estimates the posterior fault-probability from prior fault-probability plus test data, is developed from first principles and given a pattern-recognition interpretation. This leads to the philosophy that the technique is one of a family suitable for fault diagnosis, if necessary down to component level, using only input-output frequency-domain test-data. A simulation of 39 200 faulty 7-component passive circuits is described, which shows that the two voting techniques considered in detail are superior to previously reported template-matching methods of fault diagnosis by factors of 5:1 or more in situations where nonfaulty components are permitted to have standard deviations of 3% nominal value, which is not untypical of much current practice. The circuit model is similar to modules making up larger systems of several hundred components. Because voting techniques require relatively little storage of data they are well suited to quality control schemes based on automatic test equipment. In comparison, data storage for template-matching methods are often prohibitive. Best use of voting techniques is possible when clearly defined fault cases are established which should ideally be considered at the system design stage.