Research On Fault Modeling And Diagnosis Of Marine Engine Based On Machine Learning Algorithm

Marine engine is the core component of high technology ships,and the power,reliability and economy of the ship during navigation are all related to its engine.In the process of marine engine fault diagnosis,it is easy to have problems such as difficulty in fault data collection and poor diagnosis effect,so this thesis conducts a research related to the typical fault simulation and fault diagnosis method of marine engine.This research introduces the basic theory of the working process of the marine engine thermal system,establishes the simulation model of the engine thermal system based on AVL BOOST software,and compares and analyzes the simulation data with the experimental data,the results show that the errors of the relevant thermal parameters are under 1%,which proves the accuracy of the simulation model.By studying the typical failure mechanism of 11 thermal systems and setting the relevant parameters to simulate each typical failure thermal system state,and analyzing the changes and magnitude of 14 thermal parameters under each failure state,as well as the changes of each engine host parameter.Finally,a typical fault sample data set of marine engine thermal system was obtained,which laid the foundation for the subsequent research of thermal system fault diagnosis.The dynamics model of the turbocharger rotor system is established,and the calculation results of the first-and second-order critical speed are compared with the classical Riccati transfer matrix method and ANSYS finite element calculation results,and the error is within3%,which proves the accuracy of the proposed dynamics model.And on this basis,by analyzing the mechanism of each typical failure of the rotor system,establishing the kinetic models of rotor system unbalance failure,friction failure and crack failure and carrying out simulation calculation and result analysis,the dynamics characteristics of turbocharger rotor system under different failure conditions are studied,which provides a reference for subsequent fault diagnosis research of the turbocharger rotor system.The optimization of BP neural network,SVM model and KELM model using bat algorithm,grasshopper optimization algorithm and searcher algorithm,respectively,is proposed and input to the acquired typical fault dataset of thermal system for diagnostic study.Their diagnosis results show that the BA-BP optimization model,GOA-SVM optimization model and SOA-KELM optimization model all achieve significant improvement in fault diagnosis rate,with the SOA-KELM model having the highest diagnosis accuracy(99.67%).Using Hu invariant distance theory and Gaussian noise addition theory,feature extraction and data enhancement are performed on the rotor system simulation data to build a typical fault expansion data sample of the rotor system.Based on the optimization of two mainstream deep learning models,XGBoost algorithm and CNN neural network,SSA-XGBoost model and CNN-LSTM model are designed and input to the acquired typical fault dataset of rotor system for diagnostic study.Their diagnosis results show that both the SSA-XGBoost model and the CNN-LSTM model achieve significant optimization improvement effects in terms of fault diagnosis rate and model performance,with the CNN-LSTM model having the highest diagnosis accuracy(100%).