Research On Fault Diagnosis Of Gearbox Based On Multi-sensor Information Fusion

As the core component of rotating machinery,gearboxes are widely used in major machinery industries and are one of the key components of mechanical equipment.With the development of society and the progress of science,modern machinery and equipment continue to develop towards high-speed,heavy-duty,and high-efficiency.Gearbox is the main component of rotating machinery and equipment.Once the gearbox fails in operation,it will cause the chain response of the entire system.Not only will it increase production costs and cause economic losses,even cause serious staff casualties.Therefore,it is particularly necessary to monitor and predict the working status of the gearbox.However,the information collected by a single sensor is one-sided,which is not conducive to identification and diagnosis.Therefore,it is necessary to combine the information of multiple sensors to analyze the fault state of the gearbox.The main content of this paper includes:(1)Based on the gearbox model,after discussing the current commonly used test bench structure,an open structure was chosen,matched with variable frequency motors,speed torque sensors,planetary gearboxes,magnetic powder brakes,couplings and other transmission devices.The accelerometer,signal acquisition card,industrial computer and other measuring devices are selected and matched with the signal acquisition system,which can realize multi-channel simultaneous acquisition of signals and simple data processing.(2)Aiming at external interference in the original data collected by the gearbox,a Wavelet Packet Transform Method is proposed to decompose and reconstruct the original signal.The Wavelet Packet Energy Extraction Method is used to calculate the energy ratio of each component as the fault feature of the gearbox,and the Neural Network is used for classification and diagnosis.Aiming at the slow convergence speed of Neural Network and the existence of local optimization,a Chaos Particle Swarm Optimization Algorithm(CLSPSO)is proposed to optimize the neural network.The experimental data verifies that the optimized network can effectively improve the accuracy of gearbox fault diagnosis.And the time-frequency domain characteristics of the signal was extract as input,which was imported into the built CLSPSO-BP diagnosis model to get the diagnosis result.(3)Aiming at the incompleteness of the fault information collection,a fault diagnosis method combining multi-sensor information is proposed.Three sensors were set at different positions of the box body to collect the vibration signal of the gear box.The original signal is decomposed by Variational Mode Decomposition,and the Wavelet Threshold Denoising Method is used to denoise the high-frequency noisy components in the decomposed signal.The multi-sensor information fusion diagnosis results are fused by Dempster/Shafer theory,which are compared with the single-sensor training model.Experiments showed that multisensor information fusion can effectively improve the efficiency of gear fault diagnosis.(4)Design the gearbox fault diagnosis system according to the completed content.The system includes: signal noise reduction module,feature extraction module and pattern recognition module.Noise reduction modules mainly include: Wavelet packet decomposition and reconstruction,Wavelet Noise Reduction,Empirical Mode Decomposition.The feature extraction module includes a Wavelet Energy Feature Extraction Module and a signal Timefrequency Domain Feature Module.The pattern recognition module includes Chaos Particle Swarm-Back Propagation Network,Back Propagation Network and Information Fusion Diagnosis Model.