USING DEEP LEARNING BASED CLASSIFICATION ALGORITHM TO DETECT FAULTS IN TURBINE ENGINES

Authors : Ali Khalid Al-taıe

Pages : 121-140

Doi:10.53600/ajesa.1213047

View : 31 | Download : 9

Publication Date : 2025-06-30

Article Type : Research Paper

Abstract :In this paper We propose a comprehensive fault-domain-driven (FDD) approach for hydraulic systems to circumvent the constraints of supervised diagnostic tools in identifying atypical and beyond-label failures. This approach requires the inclusion of a categorization phase step prior to the diagnosis. Thus, the limits of supervised diagnostic procedures may be circumvented. In this part, we avoid the problem at hand by doing detection and diagnosis independently. Long Short-Term Memory (LSTM) autoencoders are used during the fault detection phase. In the subsequent phase, known as diagnostic, ML and DL classifiers are employed to identify the nature of the discovered defects. Even though there is evidence in the research pointing to the existence of this strategy, our work surpasses the previous art in the following respects: (1) The information collected from hydraulic test rigs has never been employed in conjunction with this specific schema. Two exhaustive trials demonstrated how this strategy may be used to resolve sensor and component difficulties. We used a unique LSTM autoencoder design in the third step, which was the detection phase. (4) During the autoencoder\\\'s detection phase, we devised a unique criterion for calculating the divergence between the anticipated signal and the input signal. It has been proved that this technique is superior to the conventional way for determining more exact diagnostic thresholds. (5) We gave a comprehensive examination of the performance of a wide variety of ML and DL classifiers that vary in their functionality and technique. These classifiers are proposed for usage during the classification\\\'s fault diagnosis phase. In addition, we analyzed the behavior of each machine learning and deep learning classifier using a range of time-domain feature selection techniques. This was done to aid future study by mapping each classifier to its most or least suitable time-domain feature in order to implement component or sensor FDD in hydraulic systems.
Keywords : FDD, LSTM, ML, DL.

ORIGINAL ARTICLE URL