Research On Data Recovery Method For Operation And Maintenance Of Diesel Engine Lubricating Oil System

In order to understand the state of the ship,the staff and the controller need to get the accurate measurement data of the sensor.The accuracy of sensor measurement data ensures the safe navigation of the ship.However,compared with other parts of the ship,the probability of sensor failure is higher.If we can’t find the fault source in time and take the right strategy,it will lead to costly and unnecessary system shutdown,and even more serious safety accidents.Therefore,it is of great significance to study how to find the sensor fault in time and take faulttolerant measures after the fault occurs.The traditional method of ship is to improve the fault tolerance of the system through hardware redundancy,such as using two or more sensors to monitor the same parameter.In this paper,the method of software redundancy is adopted,that is,the analytical model is used to provide redundancy for monitoring parameters or variables.This method does not need additional hardware and can save the cost of large-scale system.Taking the marine diesel engine oil system as the research object,aiming at the problem of abnormal monitoring data caused by sensor failure,this paper mainly does the following two parts of research work:Firstly,after finding the abnormal monitoring data,Pearson correlation analysis method is used for fault identification.By comparing the correlation coefficient matrix of normal state and fault state,the fault of equipment or sensor can be distinguished.The results show that the method is feasible.Secondly,after the sensor failure is confirmed,the analytic redundancy method is used to recover the sensor data by using the correlation between variables.In this paper,the data recovery is the application of artificial intelligence theory and method to analyze and study the sensor data,using the best estimate to replace the fault data,to achieve short-term data recovery.The specific work is as follows:The vector autoregressive(VAR)model is used to recover the sensor abnormal data.The results show that although the error is very small,but because VAR model is only suitable for linear relationship mining,and in practice many cases are nonlinear,so this method has some defects.For the mining of nonlinear relationship,BP neural network is introduced.Aiming at the difficulty of selecting initial weights and thresholds,genetic algorithm is used to optimize them.The results show that the accuracy of BP neural network without genetic algorithm optimization is not as good as VAR model,and the network error after optimization is less than VAR model,so the effect is better.Due to the small amount of training data,the recovery effect of BP neural network is not good.Therefore,this paper studies a support vector regression(SVR)algorithm which is not only suitable for small samples,but also can fully mine the nonlinear relationship.The mean square error(MSE)of VAR model is 0.103;The MSE value of BP neural network is 0.1986;The MSE value of BP neural network optimized by genetic algorithm is0427;The MSE value of SVR model is 0.0107.The results show that the SVR algorithm has the highest recovery accuracy and the best stability,which meets the needs of engineering practice.