Research On Remaining Useful Life Prediction Of Dual-fuel Engine Based On CBAM Attention Mechanism

The evolution of the whole life cycle of Marine engine is an asymmetric process,and the characteristics of its internal components are different,with highly nonlinear and complex characteristics.The traditional prediction model is difficult to apply to the medium and long term prediction,and often ignores the influence of time-varying information on the prediction effect.In this regard,the main research of this thesis is as follows:First,a simulation model calibrated under different working conditions is established to ensure that the built model can reach the required simulation accuracy.After analyzing the fault mechanism,some parameters of the model are modified to realize the simulation of the failure of the engine.The simulation data of the simulation model under normal running state and the actual ship data obtained were cross-connected to form the engine running data under healthy state,and then connected with the data of the fault degree from light to serious to build a complete data set of the whole life cycle of the dual-fuel engine.Secondly,the method and experimental process of predicting Remaining Useful Life(RUL)based on Convolutional Block Attention Module(CBAM)are proposed.Convolutional Neural Network(CNN or Conv Net)is used to extract local features,and Gated Recurrent Unit(GRU)network is used to capture long-term dependency relationships.The channel and spatial concerns of CBAM are used to reduce the impact of different data lengths on the final prediction.Hybrid networks can capture information diversity in the data.Then,data distribution analysis and correlation analysis were carried out on the turbofan engine data set generated by Commercial Modular Aero-Propulsion System Simulation(C-MAPSS)published by NASA and the data set of dual fuel engine built in this thesis.Relevant variables that could reflect the engine performance decline trend were selected to reduce unnecessary data dimensions,and then data standardization was carried out.The training sample of host RUL prediction model is established by sliding time window.Finally,the above prediction method was verified on different engine data sets,and the internal time and space dependence between multiple sensor data was mined effectively,which improved the precision of RUL prediction.The Root Mean Square Error(RMSE)and Score evaluate the predictive performance of the model.In the RUL prediction experiment of dual fuel engine with four different degradation modes,RMSE values were 41.29,43.43,58.02 and 56.27,Score values were 43.27,28.43,11.99 and204.40,respectively.It is proved that the method has good applicability and effectiveness.Accurate and effective RUL prediction of engines can provide scientific basis for engine maintenance and warn of possible failures,significantly reduce maintenance costs and prevent shipwrecks.