Evaluation and Comparison of the Principal Component Analysis (PCA) and Isometric Feature Mapping (Isomap) Techniques on Gas Turbine Engine Data

Abstract

This paper performs a comparative analysis of the results of PCA and ISOMAP for the purpose of reducing or eliminating erratic failure of the Gas Turbine Engine (GTE) system. We employ Nearest-neighbour classification for GTE fault diagnosis and M-fold cross validation to test the performance of our models. Comparison evaluation of performance indicates that, with PCA, 80% of good GTE is classified as good GTE, 77% of the average GTE is classified as average GTE and 67.6% of bad GTE is classified as bad GTE. With ISOMAP, 67% of good GTE is
classified as good GTE, 70.8% of the average GTE is classified as average GTE and 81% of bad GTE is classified as bad GTE. PCA produces 26% error rate with nearest neighbour classification and 17% error rate with M-fold cross validation. While ISOMAP produces 35% error rate with nearest neighbour classification, and 26.5% error rate with M-fold cross
validation. Results indicate that PCA is more effective in analyzing the GTE data set, giving the best classification for fault diagnosis. This enhances the reliability of the turbine engine during wear out phase, through predictive maintenance strategies.