| A number of data sets are used by computers in modern chemical processes to determine the optimum operating conditions of the processes. But the measurements of these data sets are subject to random errors and also to systematic errors that result from instrument bias or failure. The application of Studentized multiple outlier detection techniques to identify gross errors in steam-metering systems is studied by means of computer simulation. Marasinghe's statistic for two or more outliers is combined with Lund's or Grubbs' statistics for a single outlier to obtain a procedure valid for any number of outliers. The performances of these outlier detection methods are evaluated by means of stochastic testing, and the results are compared with those obtained by the MIMT and maximum Studentized residual algorithms. The effect on MIMT algorithm performance of using an estimated standard deviation is also studied. |
