Functional testing philosophies using neural networks

This paper describes the use of neural networks in combination with algorithmic test programs to aid in improving test efficiency and accuracy, especially in test situations where "bad actor" test programs exist that have difficulty in detecting and isolating Unit Under Test (UUT) failures. The paper will begin with a discussion of the theoretical basis for the use of neural networks as diagnostic aids. Specifically, as an electronic device or circuit is tested, the output of the Unit Under Test (UUT) may be considered as a function of the input. Through the use of multiple tests designed to exercise system capabilities in evaluating UUT performance, the characteristic behavior of the UUT can be established. Test results obtained from the knowledge of Automatic Test System (ATS) programmed stimulus and sensor readings can be used in conjunction with neural networks in classifying good and failed UUTs based upon this characteristic behavior. Indeed, failed UUT behavior can be further classified to distinguish faulty lower-level UUT assemblies and components.