Ind. Eng. Chem. Res. Prausnitz, J.; Anderson, T.; Grens, E.; Eckert, C.; Hsieh, S.; O'Connell, J. Computer Calculations for Multicomponent Vapor-Liquid and Liquid-Liquid Equilibria; Prentice Hall: Englewood Cliffs, NJ, 1980. Renon, H.; Prausnitz, J. M. AZChEHitherto the detection of gross errors has been focused on the u'se of three separate statistical tests. These tests sometimes give conflicting predictions, and none of them takes into consideration bound violations. In this paper we propose two new composite tests, DMT and EMT, which make use of more than one statistical test to reduce mispredictions and to account for bound violation. These

The measurement test proposed by Mah and Tamhane (1982) allows the gross error associated with a measurement to be directly identified without a separate procedure. In this paper a comprehensive evaluation of this test was carried out based on two different definitions of its power. The influence of constraints, network configuration, position of measurement, magnitudes of gross error and standard deviations, number of measurements, and other factors were summarized as rules and guidelines for the application of this test. The simulation procedure developed in this investigation may be used to design a gross error detection scheme for any specific application. C. IORDACHE, R. S. H. MAH and A. C. TAMHANENorthwestern Universlty Evanston, IL 60201 SCOPEProcess measurements are subject to two types of errors: random errors which are commonly assumed to be independently and normally distributed with zero mean, and gross errors which are caused by nonrandom events such as leaks, depositions, and inadequate accounting of departures from steady state operations as well as by measurement biases and malfunctioning instruments. In comparison with the random errors there should normally be a very small number of gross errors present in any given set of data. Nonetheless their presence invalidates the statistical basis of reconciliation procedures which are used to enhance the accuracy of process data burdened with random errors. It is therefore important to detect, identify, and remove gross errors before final data reconciliation.Three types of statistical tests have been proposed for gross error detection. Test statistics based on the residuals or imbalances of the constraints either individually (normal distribution test) or collectively (chi-square test) have been proposed by Reilly and Carpani (1963), Almasy and Sztano (1975) and Mah et al. (1976). However, in order to identify the sources or locations of the gross errors these tests must be supplemented by an identification algorithm or procedure. This requirement is obviated in the measurement test proposed by Mah and Tamhane (1982).In this paper we derive certain useful results concerning the partition of information for data reconciliation and gross error detection, and the uniqueness of the test statistics for the measurement test. We study the performance of this test as measured by its power. The power of the test is the probability of correctly detecting and identifying gross errors when they are present in the process data. Mah and Tamhane (1982) gave an upper bound on the power of the measurement test when exactly one gross error is present in the measurements. In this paper we continue to restrict our consideration to the presence of only a single gross error. We explore the influence of different parameters on the power of the test. Our aim is to evaluate the actual performance of the test and to provide guidelines on its applicability. Since the power of the measurement test is difficult to evaluate analytically, it is estimated under different conditions ...

No abstract

No abstract