| Diesel engine is the power core of many important equipments and plays an important role in the fields of generator set,national defense equipment,ship power and construction machinery.However,the harsh operating environment and complex system structure make the diesel engine failure occur frequently,which not only affects the use of diesel engines,but also may cause serious economic losses and even threaten the personal safety of the workers.At present,the widely used online fault monitoring method has limited effect and relatively backward maintenance strategy,which has been increasingly unable to meet the actual needs of modern equipment production.At the same time,according to the requirements of the diesel engine under variable operating conditions,it is necessary to constantly switch the diesel engine running in various stable rotating speed and load states,which will cause great difficulties in online monitoring and diagnosis of faults.Therefore,based on signal processing,machine learning and deep learning technology to carry out the research and application on fault early warning,diagnosis and maintenance decision optimization method for diesel engine under variable operating conditions,which is of great significance and value for diesel engine to provide equipment optimized maintenance decision based on equipment status,and enhance the security,availability,and economic benefit.Taking diesel engine as the research object,this thesis aims to realize the fault on-line monitoring and maintenance decision optimization under variable operating conditions.Starting from the study on the extraction and selection of signal characteristics,the methods for operating condition recognition are proposed,and then the online fault monitoring and maintenance decision optimization methods under variable operating conditions are studied as well,and all the methods are verified by using experimental data and engineering cases.The main research contents of this thesis are as follows:Firstly,the methods of feature extraction and selection are studied based on the characteristics of vibration signal of diesel engine.The multi-domain feature extraction of nonlinear and non-stationary vibration signals is carried out,and a feature selection method based on multi-objective optimization is proposed by constructing the importance,dimension and redundancy indexes of feature combination,which lays a foundation for the extraction and selection of condition and fault features.Secondly,based on the characteristics of vibration signal under the variable operating conditions,the methods for operating condition recognition are studied.In the case of only a few samples,the signal decomposition and multi-domain signal feature extraction are carried out based on variational modal decomposition,and the condition recognition method which combining multi-domain features and linear discriminant analysis is presented.Moreover,the local feature extraction of vibration signal based on one-dimensional convolutional neural network is studied to optimize the parameters of condition recognition model,and the method of adaptive dropout is studied to prune the network structure,and the long short term memory network is used to describe the temporal relationship between local features,in the case of a large number of samples,the condition recognition method based on one-dimensional convolutional long short term memory network is proposed to realize high efficiency and high precision condition recognition end-to-end.The methods have been verified by experiments and engineering cases,and has achieved good application.Then,on the basis of condition recognition and in the absence of fault data,the method of abnormal early warning for diesel engine is studied.The data enhancement is carried out by combining with the image transformation of vibration signal,and then the potential space model corresponding to the normal status of diesel engine based on generative adversarial network,and the model that can map vibration signal to potential space is trained based on autoencoder,and the sample anomaly degree can be evaluated based on the features about potential space and discriminator,the anomaly detection method is proposed based on generative adversarial network and autoencoder finally and verified through experiment and engineering cases.Moreover,based on the condition recognition model,the fault diagnosis method under variable operating condition is studied.According to the operating condition recognition results of cylinders,the operating condition of diesel engine is determined,and then an adaptive diagnosis method for misfire fault is proposed.Aiming at the abnormal valve clearance fault,the accurate representation of fault information is realized by the fusion of multi-domain features.In combination with condition recognition model of diesel engine,the fault diagnosis method suitable for variable operating conditions is proposed based on soft interval support vector machine.As for the problem of data unbalance caused by the scarcity of fault data,a fault diagnosis method based on improved SMOTE is proposed to improve the generalization ability of fault diagnosis model with unbalanced data set.Finally,based on the vibration signal characteristics and performance degradation of diesel engine,the maintenance decision optimization method based on vibration condition monitoring is studied.The performance degradation model for diesel engine is modeled based on the historical maintenance times,load and the degradation feature extracted from vibration signal,and the Weibull distribution is further used to describe the change process of failure rate of diesel engine,and the weibull proportional hazards model is established to evaluate the real time failure rate of diesel engine.Taking the maximum availability as the objective function,the maintenance decision is optimized,and the maintenance decision optimization method based on vibration condition monitoring is proposed finally. |
PDF Full Text Request