Multidimensional Load Estimation Algorithms That Enable Performance Analysis of Industrial Gas Turbines Without A Priori Information

Abstract

Performance analysis and diagnosis for gas turbines usually assume the use of detailed design specifications or similar kinds of information for building and configuring engine models. This allows the nonlinearity of gas turbine performance characteristics to be taken into account. However, this approach tends to make it difficult for users of industrial gas turbines to analyze performance because (1) detailed design specifications are not necessarily supplied to the users, and (2) even if they were available, use of these kinds of information may often lead to complex procedures for model building and for making adjustments and configurations that all require high expertise. The purpose of this paper was to propose a direct modeling approach based only on operating data and not requiring a priori information like manufacturer-supplied specifications while preserving sufficient accuracy. The core element of this approach was the automatic identification and selection of base load operating data from various operating conditions. A set of load estimation algorithms was proposed. They were applied to 31,000h of operating data for two types of engines, which involved actual failure occurrences, and subsequent performance modeling and analysis were carried out. The following results were obtained. (1) The relative performance trends obtained revealed the quantitative extent of degradation during operation and of recovery by repair or engine change. (2) The performance trends gave a good account of actual failures. (3) The accuracy of the performance modeling measured by the 99th percentile of error was on the order of 1%. The proposed direct modeling approach offers sufficient accuracy to quantify the gradual degradation of performance and its recovery by maintenance. The performance trends obtained are useful for further fault diagnosis.