Research On Helicopter Rotor Fault Diagnosis Based On Blind Source Separation And Artificial Immune System

Helicopter health and usage monitoring system (HUMS) has been receiving increasing attentionsin the development of Helicopter Technology in recent years, which plays an important role inimproving the safety, reliability and reducing the maintenance cost of the helicopter. Helicopter rotorsystem, as the main lift and manipulating mechanism, is crucial to the helicopter flight safety. It isalso one of the main aspects covered by HUMS. Establishment of an efficient rotor system faultdiagnosis model is the development and update for helicopter HUMS. Due to complexity of the rotorsystem, it is difficult to directly diagnose the rotor fault. Using the fuselage vibration signal deducedby the rotor faults has paved a new way. This thesis studies a variety of helicopter rotor imbalancefault diagnoses only using the fuselage vibration signal, adopting blind source separation and artificialimmune system technologies which is novel in the helicopter fault diagnosis field. A new helicopterrotor imbalance fault diagnosis model is established.The thesis mainly consists of the follows:(1) The simulation experiments show that the blind source separation technology in the separation ofmultiple mixed signals and de-noising are in terms of feasibility and advantage. Under the conditionsof underdetermine and non-underdetermine, the efficiency of blind source separation of are compared.By adding virtual noise to the signals, the blind source separation de-noising effect is tested using theperformance comparison.(2) Direct feature extraction of the rotor single and compound imbalance based on ICA estimation iscarried out. The JADE and fast-ICA methods of blind source separation techniques are comparedwhich are commonly used.(3) In the feature extraction, BP neural network is applied to diagnosis of rotor single, compound faultand fault pattern recognition respectively which all show high accuracies of diagnosis. The diagnosisresults show that the feature extractions are successful. Then an artificial immune system diagnosticmodel is established and employed to diagnose the faults which shows higher diagnosis effectcompared with the BP neural network model.