| A new approach for detecting, identifying and estimating gross errors is developed and evaluated in this work. The significant advantage of this method over existing methods is that it provides a general framework for identifying any type of gross error which can be modeled. It also makes use of information regarding the effect of a gross error on the process, provided by a gross error model. The identification of gross errors is carried out by using a likelihood ratio statistical test. We illustrate the application of the generalized likelihood ratio (GLR) method developed in this study, for identifying leaks, measurement biases, controller biases and failure of controller in steady state and dynamic processes.; We develop a methodology for applying the GLR method to a general steady state model which includes unmeasured variables and non-diagonal measurement matrices, by making use of constraint residuals. In dynamic processes, the innovations (or measurement residuals) are used to identify gross errors. In this case, we also account for the effect of control action, which has been hitherto ignored. Simulation studies on selected processes are performed to evaluate the performance of the GLR method and also to aid in its design for dynamic processes.; We also propose strategies for enhancing the performance of the GLR method. A new strategy based on serial compensation of gross errors is developed, which is useful for identifying multiple gross errors of any type and is more efficient than existing strategies in the literature. In dynamic processes, the time of occurrence of the gross error also has to be estimated. We make use of a chi-square test for this purpose and recommend guidelines for the appropriate design of this test through simulation studies of a level control process.; The conclusion obtained from this study is that the GLR method with its enhancements is preferable for identifying gross errors in steady state or dynamic processes, for its ability to differentiate and diagnose different types of gross errors. |
