Pattern Recognition And Fault Diagnosis Of Rolling Bearing Vibration Signal

With the development of the era and advancements in the transportation industry,higher demands have been placed on the safety of urban rail transit.Rolling bearings,as one of the fundamental and widely used components in rotating equipment,play an important role in ensuring the safe and smooth operation of the equipment and avoiding major incidents.However,due to the harsh operating conditions and susceptibility to noise interference,as well as the instability of vibration signals,appropriate signal feature extraction methods and effective fault diagnosis techniques are needed to ensure the reliability of fault detection.Therefore,it is a research topic of great significance to achieve pattern recognition and fault diagnosis considering the vibration signals of rolling bearings.The research findings of this thesis are as follows:1.Given that the vibration signals of bearings are nonlinear and non-stationary,there are often challenges such as noise interference and poor feature extraction performance when extracting fault information from these signals.In this regard,an improved method combining the Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(ICEEMDAN)and energy moment normalization is proposed.ICEEMDAN decomposes the acquired vibration signals of rolling bearings into multiple Intrinsic Mode Functions(IMFs),effectively eliminating problems such as mode mixing and end effects that may occur during the decomposition process.By applying this method,the measured signal is decomposed,and the first four IMF components are extracted.The feature vectors are obtained through energy moment normalization,enabling a deeper exploration of fault information within the signals and comprehensive feature extraction.2.To address the issue of poor accuracy in identifying fault types,a rolling bearing fault diagnosis method is proposed based on a Modified Fruit Fly Optimization Algorithm(MFOA)optimizing the Probabilistic Neural Network(PNN).To overcome the problem of MFOA often getting trapped in local optima during the optimization process,an improved version of the algorithm is introduced.By introducing a relaxation factor to increase the olfactory concentration judgment value,algorithmic premature convergence is avoided.PNN has extensive applications in pattern recognition and other fields,but the selection of its key parameters is still based on traditional sample clustering methods or empirical estimation,and the chosen values directly impact the overall performance of the network.To address this,MFOA is employed to optimize the smoothing factor,enhancing the network’s performance.Experimental results demonstrate that this method achieves a fault recognition rate of 99.5% for rolling bearings and enables effective fault diagnosis under different operating conditions.3.To further improve the diagnostic effectiveness and accuracy of rolling bearing faults,a fault diagnosis method based on the Harris Hawks Optimization(HHO)algorithm optimizing the Kernel Extreme Learning Machine(KELM)is proposed.The HHO algorithm is utilized to optimize the kernel parameters and regularization coefficients of KELM.The constructed feature vectors are used to train the KELM fault classification model,and different vibration test data from the rolling bearings under various operating states are employed for KELM classification.Experimental verification shows that the HHO-KELM-based method outperforms ELM and KELM methods in terms of accuracy,achieving a fault recognition rate of 99.875% for rolling bearings and yielding improved diagnostic effectiveness.