基于信息融合的柴油机故障诊断技术研究

发布时间：2018-12-29 08:58

【摘要】：目前,柴油机故障诊断已经成为国内外研究的焦点问题。柴油机故障诊断中,可以通过多个传感器得到不同的故障信息,为了得到较高的故障诊断的精度和可靠性,必须充分利用获得的故障信息对柴油机进行故障诊断。本文首先论述了柴油机故障诊断的国内外研究现状,以及将信息融合技术应用到柴油机故障诊断中的必要性和可行性,对柴油机振动信号进行了机理研究,为实验的顺利进行打下基础。为了能够从多方面反映柴油机的运行状态,首先采用神经网络方法对柴油机振动信号及压力信号进行特征级信息融合,研究了BP神经网络、RBF神经网络的原理及算法,并针对RBF神经网络的不足,提出了一种采用递阶遗传编码方案的混合递阶遗传算法,用该算法优化RBF神经网络的隐含层结构、隐含层中心值、基宽和输出的线性权值。分别采用这三种方法进行故障诊断,并在Matlab下进行仿真,将诊断结果进行比较,验证了遗传算法优化的RBF神经网络具有训练速度快、精度高的优越性。另外,为了综合利用多个传感器获得的诊断信息,将遗传算法优化的RBF神经网络和D-S证据理论相结合,针对振动信号和压力信号,将特征级融合结果构建mass函数,进行决策级融合的柴油机故障诊断,削弱了诊断结果的不确定度,诊断结果更加精确。由于D-S证据理论进行信息融合时,只是简单地对证据进行组合,本文采用加权证据合成法。当进行多个证据融合时,证据间的冲突程度用证据距离和冲突因子综合起来表示,之后通过得到的证据间的冲突程度的值确定权重系数,并修正证据基本信任分配函数。然后用D-S规则进行合成,故障诊断实验表明了此方法对冲突证据融合的有效性,充分验证了该方法能够有效提高故障诊断的准确性和可靠性。
[Abstract]:At present, diesel engine fault diagnosis has become the focus of research at home and abroad. In the fault diagnosis of diesel engine, different fault information can be obtained by multiple sensors. In order to obtain higher accuracy and reliability of fault diagnosis, it is necessary to make full use of the obtained fault information to diagnose the fault of diesel engine. In this paper, the present situation of diesel engine fault diagnosis at home and abroad, the necessity and feasibility of applying information fusion technology to diesel engine fault diagnosis are discussed, and the mechanism of diesel engine vibration signal is studied. To lay a foundation for the smooth progress of the experiment. In order to reflect the running state of diesel engine from many aspects, the characteristic level information fusion of diesel engine vibration signal and pressure signal is carried out by using neural network method, and the principle and algorithm of BP neural network and RBF neural network are studied. Aiming at the deficiency of RBF neural network, a hybrid hierarchical genetic algorithm using hierarchical genetic coding scheme is proposed. The algorithm is used to optimize the structure of hidden layer, the center value of hidden layer, the base width and the linear weight of output of RBF neural network. The three methods are used for fault diagnosis, and the simulation results are carried out under Matlab. The results show that the RBF neural network optimized by genetic algorithm has the advantages of fast training speed and high precision. In addition, in order to synthetically utilize the diagnosis information obtained by multiple sensors, the RBF neural network optimized by genetic algorithm and D-S evidence theory are combined to construct the mass function for vibration signal and pressure signal. The fault diagnosis of diesel engine based on decision level fusion weakens the uncertainty of the diagnosis result and makes the diagnosis more accurate. Since D-S evidence theory is only a simple combination of evidence, the weighted evidence synthesis method is used in this paper. When multiple evidences are fused, the degree of conflict between evidence is expressed by the distance of evidence and the conflict factor, and then the weight coefficient is determined by the value of the degree of conflict between the evidence, and the basic trust distribution function of evidence is modified. Then the D-S rule is used to synthesize and the fault diagnosis experiment shows that this method is effective for conflict evidence fusion and fully proves that the method can effectively improve the accuracy and reliability of fault diagnosis.
【学位授予单位】：大连海事大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP202;TK428

【相似文献】