Study On Intelligent Diagnosis Of Wear Fault For Marine Diesel Engines Based On Evidential Reasoning

Marine diesel engine is the main power source in most ships all over the world,and its reliability is extremely crucial for the safe and economical operation of ships.As marine diesel engine is composed of various friction pairs,wear fault is one of the main types of main engine faults,taking up almost 50% of faults in main engine.Therefore,it is necessary to study the wear fault diagnosis of marine diesel engines including wear fault location and wear fault identification to extend the service life of components,promote engines efficiency,and improve engines security and reliability.Intelligent ships have attracted wide attention in the global shipping industry,and intelligent engine room is one of the most important aspect in intelligent ships.In this dissertation,wear friction pairs of marine diesel engines are used as research objects to investigate problems in current wear fault diagnosis for marine diesel engine and study intelligent wear fault diagnosis,especially the development of diagnostic model and intelligent information process,which includes integrated utilization of multi-categories of information(including quantitative information and qualitative information),processing of information uncertainty,complexity and fault tolerance control of intelligent diagnostic models,and model update.The dissertation uses belief rule based(BRB)inference methodology and evidential reasoning(ER)rules to study concurrent wear fault diagnosis,wear mode identification,diagnostic model optimization and update,and multi-decision-making sub-systems fusion.The main research contents and innovations are listed as follows:1.Oil condition monitoring systems are designed according to the different lubrication methods of marine diesel engines(including dry sump lubrication and wet sump lubrication).By analyzing lubricating oil samples collected from diesel engines,elements concentrations in lubricating oil,and two-dimensional(2-D)and three-dimensional(3-D)characteristics of wear particles are extracted,which are used as the diagnostic features for locating wear faults type and identifying wear modes respectively.Defects existing in the historical datasets such as small dataset,imbalanced samples and missing values have been recovered,and these datasets are used in the development of wear fault diagnostic models for marine diesel engines.2.A fault diagnostic model in a parallel structure is designed which consists of several parallel binary classification sub-models.The newly designed diagnostic model can well used in single and concurrent wear fault diagnosis.Based on the BRB inference methodology,qualitative information(i.e.domain expert experience)and quantitative information(i.e.historical data)are integrated to describe the non-liner relationship between fault features and fault modes.The fuzzy uncertainty,probabilistic uncertainty and ignorance of information are all expressed by belief degree in the belief rule base.Specifically,the optimal cut-point on the receiver operating characteristic(ROC)curve is used as the middle referential point of every antecedent attribute in the BRB model so that the data distribution of every fault feature can be well represented.The wear fault diagnostic model based on BRB is used for single and concurrent wear fault diagnosis of a marine diesel engine.In the five-fold cross-validation,the performance of the BRB-based diagnostic model is compared with that of other commonly used diagnostic models,including artificial neural network(ANN)model,support vector machine(SVM)model,and binary logistic regression model.The results show that the fault diagnostic model with several parallel binary classifiers can detect concurrent faults effectively,and the model structure can be expanded when a new wear fault is considered.Additionally,the BRB-based diagnostic model can well deal with the information uncertainty and ignorance,and it can diagnose wear faults accurately and stably.The inference process is transparent and can be well explained.3.By using 2-D characteristics and 3-D characteristics of wear particles as the antecedent attributes,a bi-level BRB(BBRB)model is developed to determine the causes of wear faults in marine diesel engines intelligently.Referential points for every antecedent attribute are determined by the silhouette value,and the numerical input is transformed into belief degree by using fuzzy c-clustering algorithm.With these methods,the combination explosion of belief rules can be avoided.The parameters of the BBRB model is optimized by genetic algorithm(GA).A dynamic BRB-based classification model is developed to make the model be improved and updated in the future application and increase the model feasibility in large scale classification problems.In the dynamic model development,the incompleteness of the BRB model is analyzed and strategies are proposed to recover the incomplete BRB models.According to the results,it can be found that the 3-D characteristics of wear particles is essential to distinguish wear particles.The bi-level structure and the method to determine referential points based on silhouette value can reduce the complexity of the BRB model significantly,and the performance of the BBRB model can be greatly improved after being optimized by GA.The dynamic BRB-based classification model can well keep a balance between model accuracy and model complexity.Strategies applied in the dissertation including adding new referential points,adding consequent attributes,and adding new rules can improve the accuracy of the identification model and increase the model feasibility for new samples.4.A wear mode identification system is developed by using the ER rule algorithm to solve the problem that the other systems fail to make a decision for the samples with incomplete input features.Quantitative referential points are used to determine evidence belief distributions instead of dividing the evidence into several intervals so that the evidence can still be activated even though an input value is beyond the value range of its corresponding evidence.In the five-fold cross-validation,the ER rule-based model is compared with other different models,and the model sensitivity to the reliability factor of each evidence(i.e.characteristic of wear particles)is analyzed.It can be concluded that the ER rule-based wear mode identification model can make a decision for the incomplete input samples,and each piece of evidence can be applied in a wider range of input.The reliability of the 3-D characteristics of wear particles has a stronger influence on the accuracy of the sub-system than the 2-D characteristics.The performance of the ER rule-based wear mode identification model is improved by increasing the reliability of every wear particle characteristic which is used as every piece of evidence.5.The fusion of multi-identification systems is studied in the decision level to enhance the fault-tolerant capability of the wear mode identification model.The reliability and importance of every system are considered when the output given by every system is fused by ER rules.A new method which considers accuracy and stability of every model together is proposed to determine the reliability of every piece of evidence.The corresponding relations among categories of wear particles,wear modes and probable fault causes are summarized to identify the wear modes and wear fault causes on the basis of the wear particle information.It can be concluded that by fusing the different identification systems in the decision level,wear modes of marine diesel engines can be distinguished more accurately,and the fault-tolerant capability is strengthened remarkably.Moreover,the fusion of every subsystem can make the distinction between the real wear mode and the rest wear modes more obvious by giving a higher belief degree to the real wear mode and a lower belief to the other wear modes,which makes the identification result is clearer and more credible.This dissertation explores the intelligent wear fault diagnosis of marine diesel engine on the basis of evidential reasoning method to enhance the diagnostic model’s ability of multi-source information utilization,to decrease the model complexity,and to increase the updating and fault-tolerant capability of the model.Additionally,the intelligent fault diagnostic model can provide technical support for workers on evaluating the health condition of marine diesel engines.