Using a Machine Learning Tool to Generate Adaptive Experiments

To locate a fault in a combinational circuit, a set of test vectors is often applied sequentially to the circuit and the set of corresponding output values of the circuit is observed. These values are compared with the circuit outputs in the fault free case. Each sequence of values is a clue to the existence of a specific fault in the circuit. However, many faults can be detected just by applying some subset of the test vectors. Adaptive experiments apply a test vector based on the results of applying the previous test vectors. Thus, they are more efficient. But, creation of adaptive experiments is done manually, and has not been automated so far. Thus, the process is time consuming, difficult, and error-prone. This paper presents a novel approach in which the problem of finding an adaptive experiment is cast as a classification problem. Then a decision tree learner, C4.5, is used to automatically generate a decision tree which represents the adaptive experiment