Decoupling And Intelligent Diagnosis Of Marine Diesel Engine Multiple Faults

The characteristics of high power,high efficiency and good mobility of diesel engine have led to their important application in many fields,marine transportation and industrial production,to name a few.The malfunction of diesel engine will not only bring huge economic losses,but also cause serious safety accidents with personal casualty.It is an urgent need to timely and accurately diagnose the potential faults to make sure the safe and reliable operation of diesel engine.Fault detection and diagnosis technique of diesel engine is an effective way to make sure the safe and high-efficiency operation of diesel engine in its life cycle.It has made great advances after a long time development.A marine diesel engine has a complex structure.A set of faults may be considered as various options when the diesel engine breakdowns,and these multiple faults usually have strong correlations:a fault may unstably lead to various symptoms if occurrence and a detected symptom may be caused by many different faluts,which brings considerable difficulties to locate the real root causes of diesel engine.Consequently,pinpointing the real root cause from various options,so-called multi-fault diagnosis,has become an urgently need to be solved technological problem in the research of engine fault diagnosis.This thesis carries out a series of studies focus on engine multi-fault diagnosis from four aspects:analysing the coupling relationships of engine multiple faults,optimally selecting the operating characteristics for fault isolation,fault detection and abnormal parameters mining,decoupling and diagnosis of the multiple faults.The main research work can be listed as follows:(1)Analysing the coupling relationships of engine multiple faults:an engine multiple faults coupling analysis method is proposed based on bond graph and temporal causal graph.This approach firstly constructs a bond graph model of diesel engine according to its structure and working principle.The temporal causal graph model is then constructed via the transformation of bond graph.The function relationships of engine parameters are well described using temporal causal graph in the way of directed gragh and the response characteristics of parameters under different failures are then obtained.The causal relationships of faults and symptoms of engine lubrication system are educed using the proposed approach.The experiments based on Beta 14 marine diesel engine test cell verify the effectiveness of the proposed approach.(2)Optimally selecting the operating characteristics for fault isolation:two novel approaches are proposed in this section to optimally select the engine operating characteristics for fault isolation.The purpose of this research is to reduce the number of parameters need to be detected without decreasing the ability of fault isolation.The first approach views the problem of operating characteristics selection as a set division problem.The abnormalities of parameters are used as the basis for fault isolation,and all feasible sensor placement solutions are then obtained according to the fault classification result based on abnormalities of parameters.The second approach makes use of conditional entropy as an indicator to evaluate the capability of a subset of parameters to classify pairs of faults.A minimum subset of parameters which is enough to distinguish engine faults is then obtained using heuristic search method.The demonstration of engine lubrication system shows that the quantity of parameters needed for engine fault isolation is reduced from 8 to 4 after using the proposed approaches,which decreases the difficulty to apply engine lubrication multi-fault diagnosis techniques in practice.(3)Fault detection and abnormal parameters mining:in order to solve the shortcoming of univariate statistics regarding detecting the early faults,a fault detection approach based on adaptive kernel density estimation combined with multivariate statistics is proposed in this section.This approach firstly uses principal component analysis to project the engine observations into principal component subspace(PCS)and residual subspace(RS).Two statistics,i.e.Hotelling's T~2 and Q statistics,are then introduced to detect deviations in the PCS and RS,respectively.The control limits of these two statistics are determined by adaptive kernel density estimation.The faults can then be effectively detected if any statistic exceeds its control limit.The proposed approach is verified by the experimental measurements from a MTU8V396 marine diesel engine.The comparisons with conventional univariate statistics fault detection method shows that the proposed approach improves the fault detection rate of oil filter clogging from 65.75%to 94.50%;and from48.75%to 91.50%for oil leakage;and from 67.75%to 94.00%for bypass valve leakage,at maximum,and the deterioration of parameters are well described by their contribution values to the statistics.(4)Decoupling and diagnosis of the multiple faults:after finding the abnormal operation parameters,a Bayesian network-based fault isolation method is proposed to locate the real root cause by blending the experts'knowledge with the abnormalities.An additional nodes expert is introduced into the conventional Bayesian network to capture the experts'knowledge on the prior probabilities of engine faults and eliminate the bias of experts'knowledge by fusing the different opinions.In order to solve the problem that establishing the diagnosis model always calls for massive prior knowledge,this paper introduces the noisy-MAX model into conventional Bayesian network.The coupling effects of multi-fault to the same symptom are decoupled to the ones under a single failure,and the quantity of the conditional probability needed by the model is simplified into the linear form of the failure quantity.The results show that the proposed approach can effectively find out the real root causes of an abnormality and therefore provides a feasible way for engine multi-fault diagnosis and fault location.The thesis aims to explore an effective way to decouple the complex correlation relationships between diesel engine faults and symptoms,and distinguish pairs of faults from each other.The research results will provide theoretical guidance and technical support to improve the maintenance support level of marine diesel engines.