Feature-Based Analysis for Fault Diagnosis of Gas Turbine using Machine Learning and Genetic Algorithms

Ahn, Byung Hyun; Yu, Hyeon Tak; Choi, Byeong-Keun

doi:10.7736/kspe.2018.35.2.163

Cited by 6 publications

(2 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Due to this problem, a data-driven (artificial intelligence) method has emerged [51,52], which is also called a performance-based health monitoring approach, which success-fully detects anticipated faults. Data-driven methods include Artificial Neural Network (ANN) [50,[53][54][55][56], Fuzzy Logic (FL) [57][58][59][60], Bayesian Belief Network (BBN) [59,[61][62][63], Deep Learning (DL) [64][65][66][67], Support Vector Machine (SVM) [39,[68][69][70][71], K-Nearest Neighbor (KNN) [72][73][74] and Genetic Algorithm (GA) [75][76][77]. In the data-driven approaches, the data collected from the engine will be utilized to develop a diagnostic model.…”

Section: Gas Turbine Diagnostics Approachesmentioning

confidence: 99%

Simultaneous Fault Diagnostics for Three-Shaft Industrial Gas Turbine

et al. 2023

View full text Add to dashboard Cite

The study focused on the development of -gas turbine full- and part-load operation diagnostics. The gas turbine performance model was developed using commercial software and validated using the engine manufacturer data. Upon the validation, fouling, erosion, and variable inlet guide vane drift were simulated to generate faulty data for the diagnostics development. Because the data from the model was noise-free, sensor noise was added to each of the diagnostic set parameters to reflect the actual scenario of the field operation. The data was normalized. In total, 13 single, and 61 double, classes, including 1 clean class, were prepared and used as input. The number of observations for single faults diagnostics were 1092, which was 84 for each class, and 20,496 for double faults diagnostics, which was 336 for each class. Twenty-eight machine learning techniques were investigated to select the one which outperformed the others, and further investigations were conducted with it. The diagnostics results show that the neural network group exhibited better diagnostic accuracy at both full- and part-load operations. The test results and its comparison with literature results demonstrated that the proposed method has a satisfactory and reliable accuracy in diagnosing the considered fault scenarios. The results are discussed, following the plots.

show abstract

Section: Gas Turbine Diagnostics Approachesmentioning

confidence: 99%

Simultaneous Fault Diagnostics for Three-Shaft Industrial Gas Turbine

et al. 2023

View full text Add to dashboard Cite

show abstract

“…Yu [14] combined autocorrelation function and Hilbert transform to extract the characteristics of this fault. Ahn [15] used the machine learning and genetic algorithms to identify blade rubbing. Prosvirin [16] applied a deep learning-based observation technique for blade rubbing fault identification, and developed a deep neural network approach to diagnose the blade rub-impact faults of different severity levels [17].…”

Section: Introductionmentioning

confidence: 99%

Rub-impact fault identification of a bladed rotor based on chaotic features

Kou

Yue

Lee

et al. 2023

Preprint

View full text Add to dashboard Cite

The bladed rotor is an important part in turbine machines. Timely detection of its blade rubbing fault may avoid serious accidents. This paper developed a rub-impact fault identification approach for the bladed rotor based on chaotic features such as average period, time delay, embedded dimension, and largest Lyapunov exponent. A dynamic analytical model of a rotating variable thickness blade with the rub-impact fault is established. It is verified that the blade rubbing model could obtain the relationship between the chaotic behaviors and the rub-impact fault for a rotating blade. Through the nonlinear characteristic analysis, we could establish that different blade rubbing states have a certain corresponding relationship with the chaotic characteristics, which provides the theoretical basis for the blade rubbing identification using the chaotic features. In particular, a rub-impact fault identification approach based on chaotic features is further studied. And the blade rubbing fault can be easily identified by combining four chaotic features, which is proven by experiments. Comparing with the time domain and frequency domain analysis methods, the proposed approach provides a new way to identify such blade rubbing fault.

show abstract