Multi-state data-driven gas path analysis method

• Published in: Energy procedia Vol. 158; pp. 1565 - 1572
• Main Authors: Tang, Shanxuan, Tang, Hailong, Chen, Min
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.02.2019
• Subjects: artificial neural network; diagnosis; gas path analysis; topological data analysis; transfer learning; transfer learning; artificial neural network; diagnosis; gas path analysis; topological data analysis
• ISSN: 1876-6102; 1876-6102
• DOI: 10.1016/j.egypro.2019.01.367

Data-driven gas path analysis is a current research topic for energy fields for it can provide continuous monitoring to ensure safe and reliable operation of energy systems like gas turbine engines, wind turbines and distributed energy systems. The method utilizes information delivered by sensors to track equipment performance degradation during operation. In recent years, the rapid development of novel energy technologies promotes the progress of the method, but it also brings challenges. Energy systems are now becoming more and more complicated with strong nonlinear performance, highly coupled components and complex control laws. At present, an effective and universal diagnostic method to deal with energy systems with highly nonlinear dynamic performance has not been found. In this paper, a method called multi-state gas path analysis method is put forward to address the problem. The core idea of the method is to create sub-models to extend data available for modelling and diagnosing. The method integrates topological data analysis and transfer learning to construct a sub-model diagnostic network for nonlinear modelling. It allows designers to deal with highly nonlinear dynamic systems while preventing prohibitive computation effort. The method has been applied to an engine platform and verified. It achieves the same 92% diagnostic accuracy with only half data acquisition cost compared to a traditional data-driven gas path analysis method.