Fault diagnosis method for gas turbine based on t-SNE algorithm with time factor

Abstract

In fault diagnosis of gas turbines, it is necessary to determine the exact time when the fault occurs, so as to obtain the operating and environmental conditions of the turbine. However, the health and fault data of the turbine are mixed and hard to be distinguished because of the noise disturbance. In order to eliminate the influence of noise on clustering results, the time factor was introduced based on the traditional t-SNE algorithm. Then, a two-dimensional grid clustering method was used to process the dimension-reduced data of the t-SNE algorithm, so as to rapidly distinguish the health status of different data to determine the exact time of failure. Besides, a simulation dataset for the performance attenuation of compressors was created with the performance attenuation mechanism model of a compressor established in Matlab Simulink, and based on that, the t-SNE algorithm with a time factor was verified. The results indicate that the new algorithm is superior to other processing methods. The fault diagnosis method of gas turbines based on the t-SNE algorithm with time factor could rapidly and accurately determine the exact time when the fault of the turbine occurs, thus providing technical support to the health condition monitoring and fault diagnosis of gas turbines.

Keywords

• pipeline
• gas turbine
• fault diagnosis
• time factor
• machine learning
• clustering
• dimensionality reduction