Gas Turbine Fault Detection by Non-linear Principle Component Analysis

In this paper, the application technique for the detection of faults in a gas turbine by using the non-linear principal component analysis (NLPCA) approach is examined as, the neural network-based nonlinear principal component analysis (NLPCA) has experienced a considerable revival of interest in recent years and has been widely used in the field of diagnosis. For this purpose, the five-layer neural network NLPCA approach is used for modeling the gas turbine process in normal fault-free operating condition, with the filtered squared prediction error index (SPE) for fault detection. Consequently, we determined the optimal number of principal components retained in the PCA model, and we validated the PCA model by checking the evolution of the measured and estimated variables. Furthermore, this study for this approach is validated by a practical gas turbine example, where we used real data of the operating of the gas turbine over a period of four years. The results obtained highlight that the NLPCA approach with the five-layer neural network from the chosen detection index is more efficient. Thus, the NLPCA method appears as an efficient tool for monitoring and diagnosing faults in a gas turbine. Finally, our contribution is to validate the performance of this method using data from a dynamic and nonlinear process such as the gas turbine.