ПОВЫШЕНИЕ ЭФФЕКТИВНОСТИ СТАТИСТИЧЕСКОГО КОНТРОЛЯ МНОГОМЕРНОГО РАССЕЯНИЯ ПРОЦЕССА

Abstract

Some techniques of statistical process control can be applied in investigating an object for likely defects. In this regard, the multidimensional scattering is monitored with the use of a generalized variance algorithm. At fixed time intervals samples of observations are recorded, according to which the values of the determinant of covariance matrix (the generalized variance) are calculated. The corresponding point, falling outside the borders of control chart, indicates the process failure. The test results based on vibration data obtained from the hydraulic unit at the Krasnopolianskaia Hydroelectric Power Station showed that the algorithm of the generalized variance has an insufficient quick response relating to both gradual and spasmodic changes in the scattering level. To make the monitoring process more efficient, the following techniques have been suggested. The first one detects defects by analyzing special kind structures on the generalized variance chart. The second technique applies the warning borders on the chart and the third one implies an algorithm of exponentially weighted moving averages for generalized variance. All the tests revealed that the search for nonrandom structures on the generalized variance chart is most effective for the early detection of changes in the scattering process.