Uncertainties in gas-path diagnosis of gas turbines: Representation and impact analysis

Gas-path diagnosis is of great efficiency and economic benefit to gas turbines, whose algorithms are generally developed and tested by simulation. However, the existing simulation methods take insufficient consideration of a battery of uncertainties compared with the physical system. This shortcoming results in the poor performance of well-trained algorithms in the real system. A systematic representation scheme that covers all major uncertainties is urgently needed to narrow the gap between simulation and reality. This paper shows a representation scheme comprised of all major uncertainties. Various uncertainty ingredients are considered to fit the real system. The different impacts of uncertainties are monitored via a benchmark gas-path diagnosis method based on convolutional neural networks. Simulation results show the feasibility of uncertainty impact monitoring through a benchmark diagnosis method and verify the consistency between the proposed scheme and the reality. The fatal impact of the uncertainty with a slow frequency is discovered. And the evident sensitivity of the fault diagnosis to performance deterioration is identified in the end. The proposed representation scheme provides a platform where gas-path diagnosis algorithms can be compared under the unified and realistic benchmark.