Research On Fault Monitoring And Diagnosis Of Marine Fuel Oil System Based On Optimized KPCA-SVM

The complexity 'and automation of modern large ships are increasing day by day,and the fault monitoring and diagnosis technology of marine diesel engines has become one of the important ways to improve the safety and reliability of ships.The analysis of shutdown failures provided by the British Diesel Engine Engineers and Users Association shows that the fuel system failures account for 27%of the various causes of diesel engine shutdown consequences,the largest proportion.Due to the complicated structure of the marine diesel engine fuel oil system,there is a complicated nonlinear coupling relationship between the various components,resulting in the sampled data with typical nonlinear multi-dimensional characteristics,so the traditional linear method has a greater limitation.Based on this background,this paper has carried out the research of diesel engine fuel oil system fault monitoring and diagnosis technology based on Kernel Principal Component Analysis(KPCA)and multi-class support vector machine.In view of the problem of strong nonlinearity and noise interference in the sample data of the marine fuel oil system,firstly,it is proposed to use KPCA to extract the non-linear features of the sample data,extract the high-dimensional information of the sample data features,and then build T2 and SPE statistics in the feature space Quantity model,and finally realized the real-time monitoring of the failure of the marine fuel oil system by monitoring the change of sample data statistics.Since the performance of KPCA is affected by the internal kernel function parameters,a method for selecting kernel function parameters based on particle swarm optimization is proposed.By establishing a kernel function parameter optimization model,the optimization of the KPCA kernel function parameters is realized,and the inertia factor in the particle swarm optimization algorithm is improved,to a certain extent,the particle swarm optimization algorithm is prevented from falling into local optimization during the optimization process.The fault monitoring experiment on the marine fuel oil system shows that this method can effectively reduce the number of linear principal components and improve the accuracy of fault monitoring;The method of fault diagnosis based on multi-class support vector machine is studied,and two major factors that limit the performance of multi-class support vector machine fault diagnosis are summarized:1.Sample data quality;2.The selection of Multi-class support vector machine kernel function parameters and penalties.Based on this research,a fault diagnosis method based on particle swarm optimization KPCA-SVM is proposed,which effectively solves the problems of strong nonlinearity of fault characteristics of marine fuel oil system,low classification accuracy of small samples and difficult selection of support vector machine parameters,and realize the complementary advantages of the three algorithms.In order to improve the accuracy of marine diesel fuel oil system.fault monitoring and diagnosis,this paper proposes to combine KPCA-based feature extraction method and multi-class support vector fault diagnosis method,and adopts improved particle swarm optimization algorithm to optimized the kernel function parameters and penalty factors.A fault monitoring model based on particle swarm optimization KPCA and a fault diagnosis model based on particle swarm optimization KPCA-SVM are established.The results of simulation experiments of marine fuel oil system fault monitoring and diagnosis show that the accuracy of the fault monitoring and diagnosis model proposed in this paper is significantly improved,which verifies the effectiveness of the method.