Research On Rotor Crack Fault And Its Classification Identification Based On Super-Harmonic Resonance Characteristics

The extensive use of rotating machinery not only provides guarantee for the livelihood of society,but also reflects the importance of national science and economic development.However,with the continuous development of rotating machinery in the direction of heavy load and high efficiency,the rotor-bearing system as its core component is "overwhelmed" and mechanical failures of various scales are inevitable.As one of the most common failures in the process of metal fatigue,the safety hazards caused by cracking failures are incalculable.In order to ensure the production safety of the power station and the personal safety of the staff,the fatigue cracks are monitored and diagnosed in a timely manner.In this paper,by establishing the dynamic model of the cracked rotor and designing the test platform related to the cracked rotor,the detailed numerical calculation and experimental analysis of the fault characteristics of cracks at various angles at different depths are carried out.And based on the study of crack fault mechanism,a classification diagnosis method between crack fault and other faults is proposed.The main research content and results are as follows:(1)The fracture mechanics method is used to establish a crack stiffness model.and a detailed force analysis of the cracked rotor is carried out.By studying the influence of the crack depth and angle on the time-varying stiffness of the rotating shaft,its influence on the vibration characteristics of the cracked rotor system is explored.The conclusions show that when the depth of the crack increases,the duration of fully open and fully closed states during the breathing process shortens continuously,and at the same time,the time-varying stiffness trend of the crack in the horizontal direction changes greatly.Among them,when the crack depth is shallow(a/R=0～0.7),the nonlinear characteristics of cracks at all angles are relatively similar.When the crack depth ratio increases to 0.8,0.9,1,the critical speed of the cracked rotor decreases significantly with the increase of the crack angle.The resonant peaks of 2x and 3x frequency also appear more complex changes,especially the abnormal peaks of 2x frequency at ωn,1/2ωn and 3x frequency at 1/2ωn of cracks with an angle exceeding 45°.For the stiffness anomaly point such as the 45°crack at a/R=0.9,phenomena such as the 3x frequency reduction at 1/3ωn appear.(2)To further confirm the correctness of the model calculations in this paper and to enhance the efficiency of the superharmonic resonance characteristics for crack fault diagnosis,a multifault rotor test bench was designed for experimental verification and comparison of some common faults.The conclusion shows that for the rotor-bearing system with an slant crack with a crack depth of a/R=1 and an angle of 45。,the superharmonic resonance appears near 1/3 ωn and there is a lag in the horizontal resonance peak,which is basically consistent with the model results.The misalignment and bearing faults do not cause those resonance peaks in the 2x and 3x frequencies in the subcritical speed region,and their impact on the system is relatively small.(3)After the mechanism research of the crack fault of the rotor system and the comparison of the results of different faults,the time and frequency domain feature extraction is carried out for various faults,and the classification of different faults is realized by using various neural networks such as BP,KELM,and LSTM;and also the three models were optimized by Sparrow Search Algorithm(SSA).The conclusion shows that:KELM and LSTM have strong fault classification capabilities,and the network learning time consumed is relatively short.Especially,the high efficiency of KELM is more prominent.The efficiency of BP neural network before optimization still has a great advantage over LSTM,but its correct rate of fault classification is still slightly inferior After using SSA to optimize the corresponding parameters,the fault classification results of the three networks all reached a very high standard,and there was no obvious difference between them.At this time,KELM still shows a great advantage in efficiency,compared with LSTM,which takes the most time.