A rule-based approach to multiple statistical test analysis of binary data

Computer-Integrated Manufacturing (CIM) and related techniques provide more data faster and more often than was poss/ible in the past. Also, high-quality manufacturing requires methods capable of detecting process changes even when the process is producing in the defects per million range. Past research has shown that the best method for analyzing binary data for most situations was the CUSUM procedure. Today's manufacturing environment requires even faster, more accurate detection of process changes than the CUSUM offers. This paper describes a study in developing a system based on using multiple statistical tests together to attain the extra speed and accuracy needed. The results of the research show that the system performs very well when measured against the baseline method: a CUSUM procedure with an h 1 value of 4.2. The system outperformed the baseline handily in speed of detection. At some points, the system would have correctly classified the equivalent of over 10% of the total received signals before the baseline. For processes measured in defects per million, the actual time differential would be even more pronounced. In addition, the system performed at least as well in accuracy, providing a composite performance over the test scenarios of 95.02% compared with 94.57% for the baseline.