Abstract:Aiming at the rubbing phenomenon caused by the transient imbalance of the turbo-shaft engine between the rotor transition state and the steady state, an improved genetic algorithm (GA) optimized extreme learning machine (ELM) diagnosis model was proposed. Based on the vibration signal envelope curve of a turbo-shaft engine turbine casing, four working conditions, the normal state of the turbo-shaft engine, the rubbing state of the gas turbine rotor, the rubbing rotation state of the power turbine rotor and the rubbing rotation state of the gas and power turbine rotors were simulated; the frequency spectrum of these vibration signals were analyzed, and the characteristic parameters of the vibration signal were extracted to construct a fault sample data set; an improved genetic algorithm was used to optimize the extreme learning machine, and it was used in rubbing fault diagnosis.The results show that the average diagnosis rate of the training set is 96.8%, the fluctuation amplitude is 2.82%; the average diagnosis rate of the test set is 95.43%, the fluctuation amplitude is 0.93%, and the convergence error reaches 0.22.The method proposed has high diagnosis rate, small fluctuation amplitude and low error, which is suitable for rubbing fault diagnosis.