Research On Dynamical Feature Extraction For A Breathing Cracked Rotor And Prognositcs Methodology

Rotating machinery is indispensable in the modern industrial time which has significant influence on the national economy and key fields of national defense such as navigation and aerospace fields. Once sudden failures happen, the maintenance cost and the lost caused by unscheduled downtime are always huge, and sometimes the unexpected breakdowns can lead to serious social problems such as human death or environmental damage. Fatigue crack is one of the main causes of these dangerous damages in rotating machinery. Therefore it is necessary to make an intensive study of the generating and propagating process of the crack fault and perform the remaining useful life (RUL) prediction so that the prediction maintenance can be taken to reduce the downtimes by balancing the risk of failure and achievable profiles, which guarantees the system works with high efficiency, safety and low cost.This dissertation presents researches on the key technologies of prognostics including dynamics modeling, fault feature extraction, performance estimation and RUL prediction for a Jeffcott rotor with a transverse breathing crack. A new prognostics approach is proposed by combining the dynamics modeling, fracture mechanism and data-driven based prediction model.The breathing process of a cracked rotor is represented by closed and independent mathematical functions. By adopting these functions, the governing equation for the cracked Jeffcott rotor can be built and typical dynamics behaviors are analyzed which include the vibration amplitudes and whirling orbits around the1/3and1/2sub-critical rotating speeds. Floquet theory is used to study the influence of system parameters including rotating speed, crack depth and damping on the stability of the rotor system. Empirical mode decomposition (EMD) is employed to decompose the vibration signals and extract the high-order frequency component as the fault feature, which is helpful to understand the variation law of the crack and provides necessary foundations for performance estimation. An experiment has been set up in a rotor testbed to verify the theoretical analysis. The whirling orbits and the variations of high-order frequency components during the passage through the sub-critical rotating speeds show highly agreement with the theoretical results, which indicates that the modeling of breathing functions and the EMD based feature extraction method are correct.After studying of rotordynamics and feature extraction, performance estimation based on back-propagation neural network (BPNN) is implemented with consideration of information obtained by the dynamics and fracture mechanism theory. The generation and propagation of cracks are studied by utilizing fracture mechanism theory, and the concept of stress intensity factor (SIF) is introduced in order to accomplish the fatigue life estimation. A finite element model (FEM) is constructed by using the ANSYS software for the breathing cracked rotor in consideration of the contact problem when crack breaths. Vibration displacements obtained from the dynamics model is used as the input of the FEM and dynamic SIFs at the crack tip for one rotating periodic are calculated. The Paris law is used to estimate the fatigue life for the cracked rotor at given crack depths as the calibration basis for performance estimation. By using the features extracted with the EMD method as the inputs and the performance based on the crack depth and the fatigue life as the outputs, a BPNN is trained which accomplishes the nonlinear mapping between the vibration features and the performance representing the health state of the cracked rotor.Finally, based on these estimated performance samples and the idea of dynamic prediction, both an improved Markov model and a multi-BPNN model are constructed to dynamically predict the RUL for the cracked rotor. By taking account of the prediction effect, a combined RUL prediction strategy is proposed by using both these two prediction models. Simulation results show the proposed method has high dynamic prediction accuracy which is an effective RUL prediction method.In this dissertation, based on the cracked rotordynamics and fracture mechanism theory, the proposed prognostic strategy investigates the mechanism of the generation, propagation and representing features of the crack fault, and accomplishes the performance estimation and RUL prediction by using intelligent models, which introduces the model based analysis of degradation process to prognostic field, providing theoretical foundation for the engineering application of data mining based prognostics.