Research On Marine Diesel Engine Fault Diagnosis Method Based On Extreme Learning Machine

With the development of computer information technology and the advent of artificial intelligence and Era of Big Data,the concept of Shipbuilding is undergoing a transformation from automation to intellectualization.As the core equipment of intelligent engine room,the importance of marine diesel engine is self-evident.The traditional malfunction diagnosis methods based on monitor alarming and human inspect can't meet the need of the intelligent ship.As a potential killer of safe and reliable operation of ships,marine diesel engine fault diagnosis needs the escort of intelligent fault diagnosis methods and theories.On the basis of summarizing and drawing on the previous research results,With the wide conclusion and absorption of predecessor research,this paper takes the R6105 AZLD marine diesel engine as research object.It takes the vibration signal as the basis of fault diagnosis,using the improved empirical mode decomposition(EMD)combined with threshold denoising method to denoise the vibration signal,makes uses of manifold learning algorithm and Extreme Learning Machine(ELM)to diagnose the diesel engine fault.The main research contents and achievements include the following aspects:(1)Firstly,the EMD decomposition method is introduced,and an improved threshold denoising method based on EMD is proposed for marine diesel engine vibration signals,taking advantage of the advantages of EMD in dealing with nonstationary and non-linear signals.Among the EMD,EEMD and CEMMDAN,the two improved algorithms based on EMD,by comparing,the results show that CEEMDAN has obvious superiority.On this basis,the improved threshold denoising processing of each component obtained by decomposition is carried out,and the results show that the method has a good denoising effect.(2)Secondly,manifold learning algorithm is studied.In order to Fault Feature Extractionfrom diesel engine vibration signals more comprehensively,combining the advantages of feature fusion and manifold learning algorithm when they deal with nonlinear data,a new feature extraction method is proposed.At first,the features are extracted from three dimensions which are the features of Statistical parameters in time domain and frequency domain?CEEMDAN-SVD method and CEEMDAN-IMF method,The features of the three dimensions are combined by Serial feature fusion technique,Secondly,manifold learning method is used to extract secondary features in order to remove redundant features.According to the method,the features are extracted for the diesel vibrant signals,and the method is testified from the view of 3D visualization.(3)Finally,on the basis of systematic research on the structure and algorithm principle of ELM network,aiming at the low value of the robustness and poor generalization performance of the ELM,the ELM model are improved with using Particle swarm optimization and L2 regularization According to the experimental data and the test results,the superiority of the method is testified.When the sample of the diesel malfunction data is sufficient,this method is more efficient.For the diesel malfunction sample data being obtained not easily,the Online Sequential ELM is used for the diesel malfunction diagnosis when the data sample is insufficient.The results indicate that this method can guarantee that it can realize online malfunction diagnosis under the condition which the precision of classification is very high.