| In this dissertation, a methodology for diagnosing equipment partial failures in a process plant is described. A new model called F-curve model is developed to support the diagnosis. The particular advantages of using the F-curve model are as follows: (1) The proposed model can be easily applied to any probability-based diagnostic method such as the Bayesian method. This study adopts the Bayesian method because its construction is based on a rigorous probabilistic interpretation of data and expert judgment. In this study, the Bayesian method is supported by the F-curve model, and it is referred to as Improved Bayesian (IB) method. (2) In many cases, the model can improve the diagnostic accuracy in light of large uncertainties associated with the degree of belief to which a failure exists (i.e., strength of failure). Strength of failure, thus, depends upon the accuracy of the analyst's belief. If needed, the F-curve model can produce results with the same accuracy as the Bayesian method by means of special adjustments to the values of important parameters. (3) Sequential updating process can be applied to the F-curve model, similar to the Bayesian method, so that the values of important parameters of the model can be updated whenever new evidence becomes available. (4) When an equipment malfunctions due to a partial loss of its function--(i.e., partial failure), the F-curve model enables the diagnostic algorithm to determine the strength (i.e., the degree of belief the model estimates for the existence of a partial failure). This can be shown by a probability range along with a relevant graphic display of the results.;Some preliminary work before applying the proposed model is needed. Analysis of equipment partial failures, a priori, is essential for successful application of the model. The main purpose of this analysis is to prepare the prior information about partial failure strength distribution. The Analytical Hierarchy Process (AHP) (Satty, 1980) is employed to support this analysis. Application examples to illustrate the AHP process and to compare diagnostic accuracy between the Bayesian method and the Improved Bayesian method are provided. |
