Feature Extraction Algorithms For Weak Signals Showing Mechanical Incipient Faults In A Noisy Environment

In the complexity mechanical system (especially for the rotational machines), thereare kernel technology problmes waiting to be solved, such as the surface contactstiffness (the surfaces among the gear, bearing, rotor, pedestal and house et al), thesurface energy loss and the system nosie elimination et al, which lead to the lowsuccessful fault diagnosis ratio, especially for the mechanical incipient fault. Nowdays,to quickly, effectively and accurately separate the mechanical incipient fault weakfeature signal from the heavy noise environment, is the key technique to ensure thesucceed incipient fault diagnosis, and make sure the kernel equipments’ work conditionand defense the disaster accidents. It has been becoming the key technique for thecondition monitoring and fault diangonisis of the mechanical systems. So, it has thesignificance vaule to do research on the mechanical fault mechanism and algorithms forthe incipient fault feture signal separating from the heavy noise environment.The main aspects of the research are the mechanical incipient fault and couplingvibration signal mechanism, the non-stationary fault signal processing algorithms, andthe optimum adaptive nosie cancellation algorithms based on the evolutionary rules.The accuracy of the mechanism and algorithms are validiated by the simulation casesand experiments on the gearbox transmission system.①To study the gear meshing stiffness when there is an axle hole and tooth rootgrowth crack in the pinion, the precise gear tooth root crack growth model is presented.The relationship between the force and growth crack is built, and the correspondingtime-varying meshing stiffness calculation method is devirated based on the potentialenergy method. Based on the time-varying meshing stiffnesses and the16DOF dynamicmodel of the gearbox transimission, the relationship between the fault feature vibrationsignal and the growth crack is studied, and based on the signal analysis, the mechanicalfault vibration signal model are proposed, which proves the theoretic foundation of theresearch on the mechanical incipient fault weak feature signal detection from heavynoise environment.②The phase accumulation error and its performance are discussed when the signalare collected from the different varied shaft speed condition. For the varied shaft speedcan be estimated condition, based on the the fluctuation-period estimation from frequency domain, the modified Time Domain Averaging algorithm is put forward. Inthe algorithm, the phase accumulation error among the segments can be eliminated, andthe shortcoming of the period-based time domain averaing algorithm can be solved,when applied to process the gearbox vibration signal under the non-tachometer sesoravailable condition. It is a new method to detect the incipient gear fault of the gearboxtransmission system from the heavy noise environment.③Under the varied shaft speed can not be estimated and non-tachometer availablecondition, to solve phase accumulation error completely, the Ensemble Dynamic-TimeDomain Averaging algorithms are proposed, which is the coupling combination of theEmpirical Mode Decomposition, Dynamical Time Warpping and Time DomainAveraging algorithms. The algorithm realizes the phase shift estimation andcompensation directly from the time domain. And the feature signal detectionperformance of the algorithm is verified by the data sets from numerical simulation andthe helical gearbox test rig.④To realize the global optimum search when applying the adaptive noisecancelling method, the evolutionary rules are re-interpretation and the EvolutionaryAdaptive Noised Cancelling algorithm is proposed based on the rules. To evalutate thenoise cancellation performance of the algorithm, the indicator (Peak Ratio, PR) ispresented, which fixes the shortcoming of indicator (Signal-Noise-Ratio, SNR), and therelationship among the filter’s parameters (data length and evolutionary coefficients)and PR value are studied. To evaluate the filter parameters, the indicators (ConvergenceSpeed and Peak Ratio Height, CS and PRH) are proposed. The global optimum searchand steady convergence performance, and the noise cancelling properties are verifiedthrough the numerical simulation case, which proves the theoretic foundation for itsapplication in the condition monitoring of the complex mechanical transmission system.⑤To overcome the shortcoming (the slow global convergence speed) of theEvolutionary Adaptive Noised Cancelling algorithm, the beehive pattern evolutionaryrules are constructed and the Beehive Pattern Evolutionary Noise Cancelling algorithmis proposed base the rules. To improve the noise cancelling and weak feature signaldetection performance, the enhance filter algorithm is presented, which is coupled bythe wavelet de-noise and evolutionary adaptive algorithm. The properties of theproposed algorithms are valiated by the date sets from numerical simulations andgearbox transmission test rigs. It proves the evolutionary noise cancelling algorithm and its improved algorithms can be used to detect the fault feature signal and diagnose theincipient fault in the complex mechanical transmission system.