Research On Fault Diagnosis And Remaining Useful Life Prediction Methods For Marine Machinery Bearings

Fault identification and remaining useful life prediction are important elements of prognostics and health management(PHM)technology.By monitoring the operating status of equipment,it is possible to analyze the deep state information of the equipment and predict the remaining useful life(RUL),so that scientific maintenance recommendations can be given to improve the safe operation time of equipment and reduce maintenance costs.Bearings are widely used in marine machinery and equipment.Early identification of faults and prediction of remaining useful life,timely detection of potential defects and prediction of deterioration trends,and the adoption of efficient maintenance measures are of great significance in building a ship PHM assurance system and improving the reliability of marine equipment.In this thesis,I focus on four aspects: early fault identification method with very sparse and weak fault characteristics,small sample fault diagnosis method for non-smooth and non-linear vibration signals,remaining useful life prediction method and noise reduction performance optimization of prediction model.The research proposes an early fault identification method based on parameter-optimized variational mode decomposition(VMD),a fault diagnosis model based on multi-kernel relevance vector machine(MRVM),a remaining useful life prediction method based on improved temporal convolutional network(TCN),and the optimization of the prediction model in terms of noise immunity under noisy interference environment.The main work is as follows.(1)For the problem of weak early fault characteristics of bearings,which are difficult to identify effectively,an early fault identification method combining analytical energy operator(AEO)and optimal variational modal decomposition(VMD)is proposed.First,the original vibration signal is processed using the AEO to enhance the shock component of the signal;then,the optimal values of the key parameters and of the VMD are searched for using the equilibrium optimizer(EO)algorithm;then,the IMF components are obtained by decomposing the signal using the optimal VMD,and the IMF with the largest cliff value is selected for envelope spectrum analysis.The case study shows that the method based on the AEO operator and the optimal VMD can effectively extract the bearing early fault characteristic frequencies.(2)For the small sample fault diagnosis problem of non-linear and non-smooth vibration signals,a multi-kernel relevance vector machine(MRVM)fault diagnosis method based on complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN)and sine cosine algorithm(SCA)optimization is proposed.Firstly,the CEEMDAN method is used to decompose the original vibration signal and extract the proportion of the energy of IMF to the total energy of the signal and the energy entropy as fault characteristics;then the hybrid kernel function is introduced and the SCA algorithm is used to optimize the weight parameters and kernel parameters to construct the MRVM model to realize fault diagnosis of rolling bearings.The results of the case study show that the model has a high diagnostic accuracy.(3)For the remaining useful life prediction problem of bearings,a method of remaining life prediction based on improved TCN is proposed on the basic of temporal convolutional network(TCN)which has the advantage of processing time series data.The basic building blocks of the TCN are first introduced,and then some improvements including the use of the PRe LU activation function,the design of asymmetric residual units and the introduction of an attention mechanism are incorporated.Simulation experiments using a bearing degradation dataset show that the improved TCN model has high prediction accuracy for RUL prediction.(4)To address the problem that the noisy environment can have an error effect on the RUL prediction results,deep residual shrinkage network(DRSN)is used to process the noise-laden signals,and an improved TCN remaining useful life prediction method based on the DRSN is proposed.The good denoising performance of DRSN is analysis by using noisy signals with different degradation states and different signal-to-noise ratios.The good noise immunity and good prediction accuracy of the proposed method under noise interference are demonstrated by model prediction examples of noisy signals.