Comprehensive fatigue estimation and fault diagnosis based on Refined Generalized Multi-Scale Entropy method of centrifugal fan blades

Abstract

Fan is widely used in industry as a very important component of pressure transmission and system cooling. Fan blade is the key part which decides the service life of fan and the whole system. In this paper, the research focuses on the centrifugal fan of high-speed train cooling system and the blades on it. The experimental system and research method of fatigue analysis and fault diagnosis are built. The fan can work in both steady and unsteady state conditions controlled and driven by the test system. At the same time, the signals of strain and acceleration at different positions can be obtained accurately. Based on the static tension and dynamic fatigue test of the material, the accurate simulation of the fan is carried out. By simulation analysis, steady-state test, long-time start-stop test and metallographic test, the prediction of fatigue life of fan blades is conducted with in-depth analysis and comprehensive evaluation. The results show that fan meets the requirements of actual working life. As the actual working state is more diverse and severe than the above test process, some phenomena of damage also appear in the blades after service. Based on real situation, the blade faults are classified. After determination of measuring point position, the vibration signals of four fault states of blade under rated speed are collected. This paper proposes a new fault feature extraction method – the Refined Generalized Multi-Scale Entropy (RMSEσ2) combined with Support Vector Machine (SVM). According to the results of the empirical mode decomposition (EMD), the effective intrinsic mode function (IMF) components are selected by energy distribution and correlation coefficient. The comparative analysis of multi-scale entropy (MSE), generalized multi-scale entropy (MSEσ2) and RMSEσ2 between the original signal and 1–8 IMF components are carried out. Results show the RMSEσ2 of IMF1 component of Z direction of point 1 is the best choice. And it is imported into SVM for pattern recognition with scale factor 20. The optimal SVM model is obtained by choosing kernel function and parameters that are optimized by Particle Swarm Optimization (PSO) method. The results show the faults of centrifugal fan blades can be classified accurately. A set of research flow and method is set up, which provides important reference for fatigue and fault pattern recognition of fan blades in the future.