【作者】 白广来；

【导师】 孟宪尧；

【作者基本信息】 大连海事大学 ， 轮机工程， 2003， 博士

【摘要】 本文系统地研究了船舶柴油机和船舶机舱设备的故障诊断理论与应用技术问题。船舶柴油机是整个船舶动力的源泉,是船舶可靠运行的保障。采用分布式现场监控系统对柴油机的运行状态进行有效的监控,并利用柴油机的动态模型、灰色关联度法、神经网络、和信息集成故障诊断的方法对柴油机进行有效的故障诊断。 监测是进行船舶柴油机故障诊断的基础。本文首先讨论了柴油机和船舶机舱设备的监测系统结构和故障的检测方法。论述了在船舶机舱监控系统中采用现场总线系统的特点和优势,设计了一个基于现场总线的分布式机舱监控系统的结构模型。 本文讨论了基于柴油发动机瞬态转速波形和气压扭矩波形对柴油机气缸内故障进行诊断的方法。柴油机瞬态转速中包含有丰富的信息,体现了柴油机轴上的振动情况和各个气缸的作功能力。柴油机燃烧产生的气体压力是柴油机运转的原动力,是转速瞬态波动的原因。通过分析瞬态转速和气压扭矩的波形曲线,可以判断柴油机各缸的工作性能,并对故障进行诊断。本文建立柴油机故障诊断用的动态模型,根据模型和瞬态转速的测量进行柴油机气缸压力扭矩曲线的计算,并给出了根据瞬态转速和气压扭矩波形进行特征参数的提取和故障诊断的方法,对瞬态转速的测量和处理方法进行了说明。 灰色系统理论在模式识别和故障诊断领域中有着广阔的应用前景。本文讨论了灰色关联度法应用于柴油机故障诊断的可行性,并对灰色关联度用于故障诊断中的一些特殊问题进行了探讨,提出了柴油机故障诊断的改进的关联度算法。进行柴油机故障诊断时要选择能很好地表征系统特性的参数,并进行无量纲化处理。对B型关联度的使用,本文给出了在故障诊断中对B型关联度进行取舍的原则。为了更好地反映柴油机故障的不同程度,本文采用了不同程度故障参数的参数域,构造判断故障的基准模式,可以提高故障诊断的分类准确性。在很难建立典型故障特征模式,或不能预先设定典型故障时,灰色关联度故障诊断法受到限制。针对多缸柴油机的各种气缸故障,通过改进的优劣关联度的更多还原

【Abstract】 Fault diagnosis theory and technique applying to marine diesel engine and equipment in marine engine room are systemically investigated in this thesis. A marine diesel engine is the power supply of a ship, and its healthful running is the guarantee to the safe voyage of the ship. The operation states of a diesel engine can be effectively monitored through a distributed system on board. Some fault diagnosis based on engine dynamic models, gray relation method, neural network, and information fusion are studied in this thesis to make effective diagnosis of engine faults.Monitoring is the foundation of diesel engine fault diagnosis. The structure of the monitoring system of marine engine, and faults detecting methods are firstly discussed in this paper. The advantages of using fieldbus in marine engine room are described, and a structure model of distributed marine engine room monitoring and controlling system based on fieldbus is designed in this paper.A diagnostic method for diagnosing a diesel engine cylinder faults according the engine transient speed waveform and gas pressure torque waveform is presented in the thesis. There is plentiful information in the engine transient waveform, which represents the torsional vibration on engine shaft and individual cylinder’s working ability. The gas pressure due to engine combustion is the source power to drive the engine running, and the causation of the fluctuation of engine transient speed. Through the analyzing of the transient engine speed waveform and gas pressure torque waveform, the performance of each engine cylinder can be evaluated, and faults can be diagnosed. In the 3rd chapter, a diesel engine dynamic model for diagnosis use is established. Based on this model and the measurement of engine transient speed, the engine gas pressure torque waveform can be calculated. From the transient speed waveform and gas pressure torque waveform some characteristic parameters can be abstracted. The diagnosis method using these parameters and the measurement way of transient speed are described in detail.The gray theory has a wide range application in the fields of pattern cognition and fault diagnosis. In this thesis, the feasibility of applying gray relation degree method to diesel engine fault diagnosis is discussed; the speciality of applying gray relation to fault diagnosis is also studied; and an improved gray relation algorithm for diesel engine diagnosis is presented. It is essential to select proper parameters that can best represent system features in the engine diagnosis process. For the B-mode relation, the principle is given on how to use B-mode in fault diagnosis. To better representing different fault extent, a parameter field, which corresponds to different fault grade, is used to construct the standard fault model. This can improve the accuracy of fault classification. There is limitation for the gray relation fault diagnosis method in some case which typical fault model is difficult to establish. For cylinder faults in a multi-cylinder engine, an improved gray relation degree method, grey excellent and inferior relation method, can be use to analyze and evaluate thecylinder working sates, that is, how close the working states with the best states and worst states. And further an evaluation and fault diagnosis can be made.Neural network is an effect tool for fault diagnosis. In this thesis, several fault parameters, sampled from engine transient speed waveform and gas pressure torque waveform, are used to build a feed forward neural network, and the back-propagation (BP) algorithm is used to perform a case study of engine fault diagnosis. To overcome the disadvantageous of the BP algorithm, slow convergence speed and easy to fall into local small extremum, self-adjustable learning rate and momentum method are used to improve the performance of BP algorithm. The selection of learning rate and momentum coefficient are studied in detail. Using mathematical algorithm optimizing BP training can increase the network convergence speed. Case studies are made to compare these mathematical algorithms, and a guideline is given for the selection of neural network structure in fault diagnosis use. And, the application of Recurrent Network and Radius Basis Network to fault diagnosis is also primarily studied.Due to the complexity of a diesel engine and facilities in engine room, one monitoring or diagnosing method is sometime difficult to get an accurate result. Thus raise the demand for multi-sensor information fusion in fault detection, and integration of diagnostic method. The multi-sensor information theory, function model and structure are studied, and study case on diesel engine fault diagnosis using Bayesian Method and D-S Evidential Theory are given. To share the diagnosis information, and to increase the accuracy and reliability of fault detecting and diagnosis, a distributed multi-intelligent agent diagnostic structure can be used. In this paper, study work is mainly focus on the theory and structure of multi-intelligent agent diagnostic system, and multi-sensor information fusion theory and methods. Further should be done on the application of information fusion and integration.更多还原