Research On Fault Monitoring Method Of Gas Turbine Rotor

As a large-scale advanced power machinery,gas turbines are the hallmark of modern industrial technology.Gas turbines have many types of faults due to their complicated structure and harsh working environment.Once a fault occurs,it may cause equipment performance degradation or mechanical damage.If it damages the entire system,it may even cause casualties and cause irreparable damage.Studies have shown that effective monitoring of equipment operations and timely measures can avoid most accidents.In fact,it is very difficult to directly monitor the fault of the gas turbine.Most of the faults are related to the rotor.Therefore,this paper studies the fault monitoring method of the gas turbine rotor.This paper first analyzes the basic form of gas turbine rotor vibration,and analyzes the basic characteristics of rotor vibration based on the simplified dynamic model.Then,several typical rotor fault types are studied in depth,and the causes,vibration mechanism and vibration signal of rotor fault are analyzed in detail.Features,and a brief description of the impact of failures and preventive and maintenance measures.When the gas turbine is in operation,not only the rotor will vibrate,but also the mechanical components will vibrate.The collected signal is not a simple rotor vibration signal.In addition,due to the sensor’s own characteristics,the signal will also have a certain degree of distortion.Therefore,the actual acquired rotor vibration signal data is very confusing,and has non-stationary and nonlinear characteristics,which cannot be directly used to establish a fault monitoring model for the rotor operating state.Based on this,this paper proposes a variational mode decomposition algorithm based on morphological singular values to process the rotor vibration signal.Firstly,the algorithm uses the morphological singular value decomposition algorithm to filter the random noise and abnormal mutation components in the vibration signal,and then uses the variational mode decomposition algorithm to decompose the signal into modal components with different center frequencies,and calculate each component.Arranging the entropy value as a feature vector provides a reliable input space for establishing a gas turbine rotor fault monitoring model.For the precise mechanical structure of the gas turbine rotor,the characteristics of each fault state are complex,and the recognition accuracy is not ideal.This paper proposes the AdaBoost-ELM algorithm.The algorithm takes the permutation entropy of each component of the signal processed by the variational modal decomposition algorithm based on the morphological singular value as the input space,and the extreme learning machine based on the improved mexh wavelet function as the weak classifier,adjusting the data set through multiple iterations.The probability distribution trains multiple weak classifiers and linearly combines them into a strong classifier as the final gas turbine rotor fault diagnosis model.Finally,the effectiveness of the proposed method is verified by the typical fault vibration data collected on the Entek rotor simulation test bench.