Research On Fault Diagnosis Method For Diesel Engine Crankshaft Bearing Wear Based On Vibration Signal

The crankshaft bearing is the most important component in the engine.It bears the force transmitted by the connecting rod and converts it into torque,outputting and driving other accessories on the engine through the crankshaft for operation.It has a wide range of applications.Excessive axial clearance of crankshaft bearings can cause abnormal wear of the piston connecting rod group,crankshaft journal,and bearings.If it is severe,it can also cause the crankshaft to fracture due to excessive axial stress.When a diesel engine malfunctions,it can pose a safety hazard to personnel on board the vehicle.Traditional mechanical fault diagnosis methods require a large amount of manpower and material resources,requiring mechanical disassembly,and even causing unnecessary damage to components.The fault diagnosis method based on vibration signals can avoid such problems,achieve non disassembly detection,and save a lot of costs.Therefore,the research on fault diagnosis methods based on vibration signals has always been a hot topic both domestically and internationally.Firstly,establish a crankshaft bearing dataset based on the dataset collected from a certain unit,decompose the signal using Modified Ensemble Empirical Mode Decomposition,and draw the Fourier transform and AR spectra of the components for analysis.Combining Simulated Annealing to improve the stability of MEEMD and further improve the signal decomposition effect.Meanwhile,by comparing the decomposition and analysis effects of fast Fourier transform,Empirical Mode Decomposition,Ensemble Empirical Mode Decomposition and Comprehensive Ensemble Empirical Mode Decomposition,the reliability of MEEMD decomposition is further verified.Finally,the first five IMFs components obtained from the optimized MEEMD decomposition are used as the dataset for the neural network.Long Short Term Memory is used for training and prediction,and a training process diagram and a comparison chart between predicted and actual values are created to record the running time and recognition accuracy.The LSTM is optimized by using the improved Particle Swarm Optimization combined with Gaussian function to find the optimal parameters,and the final improved PSO-LSTM fault identification accuracy reaches 97.22%.At the same time,a control experimental group of MEEMD-SVD-BP and MEEMD-SVD-RNN using Singular Value Decomposition for feature extraction and fault diagnosis was established to verify the efficiency of LSTM.Establish SA-LSTM and PSO-LSTM based on linear decreasing weights to verify that the improved PSO-LSTM model can accurately and effectively identify faults in the wear level of crankshaft bearings...