Research On Fault Diagnosis Method Of Locomotive Gear Based On POS Optimized MCKD

Gears are critical components in rotating machinery systems to transmit power and change speed.Under the impact of cyclic loads,gears often fail because of lubrication,temperature and vibration.If these failures are not resolved,it will cause the transmission system and even the entire mechanical system stopping work,and causing major safety accidents and economic losses.Health assessment of gear systems based on vibration signals continues to be one of the most effective solutions.However,due to the harsh working environment of the gear and the complicated transmission path of the vibration signal,the collected signal is severely disturbed and it is difficult to accurately extract the fault characteristics.Aiming at this problem,a gear diagnosis method based on particle swarm optimization to optimize MCKD(maximum Correlated Kurtosis Deconvolution)is proposed in this thesis.This method has a dazzling ability to adaptively extract gear fault features and can accurately locate the location of the fault.At the same time,a probabilistic neural network algorithm is designed to effectively improve the accuracy of fault classification and recognition.The main research content in this thesis is shown as follows.(1)The formation of the vibration signal and the time-frequency domain signal characteristics of the gear under different fault modes are analyzed,and the time-domain simulation signal of the gear is constructed based on the time-domain signal characteristics of the gear under local fault conditions.Several common methods of signal analysis are presented.Through preprocessing,feature enhancement and demodulation of the constructed signals,their ability to improve the signal-to-noise ratio of gear fault simulation signals and extract fault features is compared and analyzed.(2)A gear fault diagnosis method based on particle swarm optimization and maximum correlation kurtosis deconvolution is proposed.This method effectively overcomes the original algorithm's tendency to fall into a narrow optimal solution by improving the initial particle distribution and selection of inertial parameters of the particle swarm algorithm.An extreme value differential evaluation function is created founded on the characteristics of the fault signal.The evaluation function can adaptively extract the filtered signal containing the most fault characteristics.The algorithm is employed to the extraction of pulse features and the diagnosis of gear faults in wheel-on running experiments.Simulation and experiments show that this method can effectively overcome the limitation that the maximum correlation kurtosis deconvolution depends on prior knowledge,and it can fault features are well extracted,and the location and form of the fault are determined.(3)A classification model combining particle swarm optimization maximum correlation kurtosis deconvolution algorithm and intelligent classification and recognition algorithm of probabilistic neural network is proposed,and the comprehensive test bench for power transmission fault diagnosis proves that the model can effectively improve the probabilistic neural network.Fault classification recognition accuracy of the algorithm.