Research On Fault Diagnosis Method Of Marine Diesel Engine Based On Probabilistic Neural Network

The proposal of intelligent engine room and unmanned ship indicates that the ship has entered the era of intellectualization,which puts forward higher requirements for the fault diagnosis technology of diesel engine.As the power source of the whole ship system,marine diesel engine provides power output for the whole ship.Its reliability is directly related to the navigation safety of the ship.However,the traditional fault diagnosis relies on manual maintenance after the event,which obviously can not meet the requirements of intellectualization.Therefore,it is of great significance to study the intelligent fault diagnosis of marine diesel engine.In this paper,R6105 AZLD diesel engine is taken as the research object.In order to improve the accuracy of fault diagnosis,signal processing and fault diagnosis are studied respectively.The whole research process is mainly divided into signal denoising,feature extraction and fault pattern recognition.(1)Signal noise removal.This paper studies the signal pretreatment methods of partial least squares method and five-point approximation method.The de-linear trend term and deburring pretreatment of diesel engine vibration signal are carried out.And introduces the principle of wavelet threshold denoising,proposes a new wavelet threshold denoising method based on a new threshold function,which resolves the shortcomings of the traditional threshold function,and uses the simulation experiment to verify the superiority of the new threshold function.Finally,the wavelet denoising method based on the new threshold function is adopted to compare the wavelet basis,threshold value and different decomposition layers with the SNR and the mean square error as evaluation indexes,and obtained the denoising scheme with the best denoising effect.The results show that the wavelet denoising based on improved threshold function can better remove diesel engine noise signal,and it can also be used to denoise diesel engine noise signal.Retain the active component of the signal.(2)feature extraction.The feature extraction methods of time domain,frequency domain and wavelet domain are described,and the feature parameters extracted by the three methods are fused to obtain a rich set of fault features.Then,the stack automatic encoder is used to extract the deep features of the fusion features.And studied the influence of different network layers and network nodes on feature extraction,obtained the optimal scheme of network feature extraction.The results show that the feature extraction method based on stack automatic encoder has less overlap and obvious difference,and the clustering effect is better.(3)Fault pattern recognition.This paper introduces the theoretical knowledge of probabilistic neural network.According to the difficulty in selecting parameters of probabilistic neural network,an artificial fish swarm algorithm is proposed to optimize parameters.In order to save training time and improve network efficiency,PCA is used to reduce the dimension of initial data.Finally,the data of seven working conditions of diesel engine are diagnosed by using the model.The results show that the probabilistic neural network based on artificial fish swarm optimization has higher diagnostic accuracy and shorter training time after PCA processing.