An improved bistable stochastic resonance and its application on weak fault characteristic identification of centrifugal compressor blades
Large-scale centrifugal compressors play an important role in modern industry. As the core component of a centrifugal compressor, the blades are prone to fatigue failure due to the long-term operation in complex conditions. If the early stage blade failure cannot be found in time, catastrophes could be caused by this potential risk. However, since the blades work in a closed environment, there is always lack of effective monitoring techniques. Aiming at this challenge, pressure pulsation signal inside the compressor is studied in this paper for the condition monitoring and weak fault waring of blades indirectly. A big issue is that the fault characteristic induced by incipient blade crack is quite weak, which will be much weaker in pressure pulsation signal, and interfered with strong noise. Hence, appropriate feature extraction methods are urgently needed. An adaptive bistable stochastic resonance method combined with multi-scale noise tuning is proposed to improve this problem. As the classical stochastic resonance is just suitable for small parameter signal, normalized scale transformation is adopted to overcome this disadvantage. In addition, numerical stability analysis for the stochastic resonance system is conducted to ensure the convergence of system output and improve the characteristic enhancement performance of proposed method. Simulation signal is constructed to verify the effectiveness of the proposed method first. Then, the experimental pressure pulsation signal is analyzed by this method. Analysis results verify that the proposed diagnostic framework can effectively identify the weak characteristic frequency induced by blade crack and has potential for long-term condition monitoring and fault warning of large-scale centrifugal compressor blades.