Planetary Gear Fault Diagnosis Based On EMD And Fish Swarm Algorithm Optimizing Neural Network

Planetary gearbox has been widely used in aerospace,energy and heavy industry,such as helicopters,wind turbines and heavy trucks,due to their high transmission ratio and strong carrying capacity.The transmission structure of planetary gearbox is complex and has unique working characteristics.The vibrations of the gears of different phases are coupled with each other in operation.There are multiple vibration transmission paths from the gear mesh point to the sensor on the gearbox shell,and the load on the gearbox also increases the influence of nonlinear transmission path.These factors make it difficult to extract the fault features of the planetary gearbox and increase the difficulty of its fault diagnosis.This paper takes the planetary gearbox as the research object,uses CEEMDAN(Complete Ensemble Empirical Mode Decomposition with Adaptive Noise)as the signal analysis method,introduces the sample entropy algorithm as the feature extraction method.This paper uses ALNAFSA(Adaptive Local Neighborhood Artificial Fish Swarm Algorithm)to optimize the BP neural network(ALNAFSA-BP)and identifies the planetary gear failure model.This paper first introduces the research background and significance of this subject,introduces the domestic and foreign research status of planetary gearbox fault diagnosis technology and the research status of artificial fish swarm algorithm.Subsequently,the EMD(Empirical Mode Decomposition)and three noise-assisted EMD methods are described and verified by simulation experiments.Simulation results show that the CEEMDAN algorithm can effectively reduce the modal aliasing phenomenon in the EMD algorithm,and the reconstruction error value of the method is small.Then the sample entropy algorithm is introduced,and a feature extraction method based on CEEMDAN and sample entropy is designed.Experimental results show that the method is effective.After that,AFSA(Artificial Fish Swarm Algorithm)and its improved mode are discussed,including the parameter settingand optimization behavior in the algorithm.After that,a planetary gearbox test bench was built,and the vibration signals under five working conditions were collected,and the signals were compared and analyzed using the CEEMDAN method.Finally,ALNAFSA and BP neural network are combined to optimize the initial weights and thresholds of BP neural network,and ALNAFSA-BP model is constructed;the vibration signal of planetary gearbox is extracted by the method of CEEMDAN and sample entropy;the wear fault,crack fault and composite fault of planetary gear are identified and diagnosed by ALNAFSA-BP model.The results show that ALNAFSA-BP has achieved good results in the fault diagnosis of planetary gearbox and improved the accuracy of recognition.