Generalized likelihood ratio method for gross error identification

Abstract

A new method for detecting, identifying, and estimating gross errors in steady state processes is described in this paper. The generalized likelihood ratio method is based on the likelihood ratio statistical test and provides a general framework for identifying any type of gross error that can be modeled. The procedure is illustrated with gross errors caused by measurement biases and leaks. One significant advantage of the method is that the identification of gross errors is not confounded by departure from steady state conditions, which may now be accounted for by "leaks." Also proposed is a new strategy for identifying multiple gross errors using serial compensation of gross errors, which may be applied to all types of gross errors including leaks and biases and which requires less computing time than the existing strategies.

Accurate Identification of Biased Measurements Under Serial Correlation

“…It has been used in a number of methods under the assumption of white noise (Iordache et al 13 , Heenan and Serth 14 , Rosenberg et al 15 , Narasimhan and Mah 16 ). To the authors' knowledge, Kao et al's application of the MT, is the only GED method dealing with serial correlation that has been formally evaluated and thus, should be used as the medium for comparing the performance of the new methods stated here.…”

“…Hence, this indicator does not provide suf®cient information regarding the false identi®cation of non-biased variables when bias exists. To give a better measure of performance for unbiased variables when bias exists, P(type I error) was replaced with the AVTI (averaged type I error) (Narasimhan and Mah 16 ) and the OPF (overall performance) (Rollins and Davis 1 ) added to the study. These measures are de®ned as…”

Intelligent state estimation for fault tolerant nonlinear predictive control

“…When an actuator or a sensor fails abruptly, then the models for failure modes have to be developed in a different manner [17]. For example, if jth actuator is stuck abruptly at instant t, then plant input uðkÞ subsequent to the failure (denoted as u u j ðkÞ) can be represented as u u j ðkÞ ¼ mðkÞ þ ½b u j À e T u j mðkÞ e u j rðk À tÞ ð 15Þ…”

Several techniques to detect and identify systematic biases when process constraints are bilinear

“…Another popular strategy is called the serial compensation strategy (Narasimhan and Mah, 1987). As in the serial elimination strategy, measurements are serially removed based on the largest value of the test statistic that exceeds the critical value.…”

“…Narasimhan and Mah (1987) have shown that both of these serial strategies can still have high type I errors in the presence of multiple biases. This is not surprising since these techniques are based on the MT.…”