Manifold Learning-based Fault Recognition Method Of Rotating Machinery

Rotating machinery plays a key role in the modern manufacture system. The whole mechanical system will break down due to domino effect if some part has a fault while the operator isn’t awareness of that. It also brings economic losses and injuries. Therefore, the fault recognition technique of rotating machinery obviously counts a lot. Generally fault recognition has three steps, acquiring the vibration signals, feature extraction, working condition recognition. The second step is so crucial that was deeply studied in this paper.The fundamental principle and algorithm of manifold learning was studied, and an improved algorithm was proposed to solve the problem that the number of observation samples sometimes was reduced in the analysis result. The effectiveness of manifold learning in revealing the underlying structure of data set was verified through UCI repository. High clustering of manifold learning in fault recognition was also discussed. Comparing to the canonical linear methods like PCA, our approach is capable of discovering the intrinsic structure that underlying complex natural observations.Feature information was extracted by means of manifold learning combined with intrinsic mode function(IMF) energy. Vibration signals were decomposed to some IMFs via a time-frequency distribution method, empirical mode decomposition(EMD). The feature vector took the energies of the chosen IMFs as its elements. Then manifold learning could extract the reduced features that still can preserve the fault information.Machinery fault simulator magnum was the experiment setup. The signal acquisition system was build based on Lab VIEW. Two practical cases, bearing and gear fault recognition, validate the merit of the proposed method. Results show that the manifold learning-based method was sensitive to fault severity but not sensitive to load and rotational speed. Finally manifold learning was introduced into wind turbine online monitoring and fault diagnosis system. The proposed method successfully diagnosed the drivetrain fault of the wind turbine. It is a meaningful attempt from technique theory to engineering practice.