Fatigue life prediction of a supercritical steam turbine rotor based on neural networks

• Published in: Engineering failure analysis 2021-09, Vol.127, p.105435, Article 105435
• Main Authors: Zhao, Xiang, Ru, Dongheng, Wang, Peng, Gan, Lei, Wu, Hao, Zhong, Zheng
• Format: Article
• Language: English
• Subjects: Fatigue life; Finite element analysis; Neural networks; Steam turbine rotor
• ISSN: 1350-6307; 1873-1961
• DOI: 10.1016/j.engfailanal.2021.105435

•A data-driven life monitoring system for key components of steam turbine is proposed.•A FEM database is constructed for the neural network training.•The Neuber rule and trained network are combined for fatigue life prediction.•The framework provides new solutions for life damage monitoring of complex systems. The safety and stability of rotors are significantly important for smooth operations of steam turbines. To predict the fatigue life of a 350 MW supercritical steam turbine rotor online, a data-driven based neural network is proposed in this paper. Finite element analysis is employed to determine the danger zones of the whole rotor and then a large sample dataset consisted of temperatures and stresses is established for subsequent neural network training. Different from the traditional thermo-elasto-plastic or finite element methods, the proposed approach can effectively calculate temperatures and stresses at the danger zones by inputting measured parameters. The Neuber rule and Manson-Coffin equation are used to estimate the fatigue life of the rotor. It is shown that the proposed neural network-based method can assess the operating status of steam turbine during different cold startups and provide a feasible online health monitoring methodology for steam turbine rotor, without dealing with the quite challenging thermo-mechanical analysis.