Faults Detecting of High-Dimension Gas Turbine by Stacking DNN and LLM

Increasing the input dimension cause curse of dimensions which make the inefficiency of methods, especially in analysis and interpretation practically. In this paper, a new two-part structure for fault diagnosing and identifying (FDI) of high-dimension systems has been presented. The first part of which consists of Auto-Encoder (AE) as Deep Neural Networks (DNNs) to produce features engineering process and summarize the features, and the second part Local Models Networks (LMNs) with LOcal LInear MOdel Tree (LOLIMOT) algorithm to model outputs. Then the residual which generated by comparing output of system and obtained models in each condition is used to alarm faults. Standard laboratory Gas Turbine data is the case study for this structure. Finally, by comparing the simulated results with the several reliable works, the effectiveness of this proposed structure is well illustrated.