基于信息融合的发动机故障诊断研究

发布时间：2018-05-15 23:44

本文选题：故障诊断 + 特征提取　；参考：《哈尔滨理工大学》2017年硕士论文

【摘要】：发动机设备属于典型的多层次系统,各部件之间互相影响关系复杂,其工作状态往往受当前环境、运行条件等诸多不确定性因素的影响,这些复杂特性决定了发动机故障存在多样性和模糊性的特点,这使得传统的诊断方法故障识别率较低,对于复杂情况往往需拆机检查,进一步增加了停机造成的经济损失。因此,在发动机不解体的前提研究一种快速、准确地故障诊断方法变得十分必要。针对发动机故障检测的难点,本文将信息融合技术应用于发动机的故障诊断领域,给出了一种将人工神经网络和D-S证据理论相结合的决策级信息融合方法,该方法主要由故障信号的特征提取、故障的初级诊断和故障的决策级融合诊断等部分构成。对于故障信号的特征提取部分,本文提出了一种基于小波细节系数自相关分析的消噪方法,借助Mallat小波将原始信号进行噪声分离,更大程度的保留了故障信号中的有用信息,为故障诊断的时频特征提取奠定了基础。为了更全面准确诊断故障,本文还引入了发动机瞬时转速和气缸压力信号特征,为后续诊断提供了三种不同的特征矢量,以便于进行融合诊断。在故障的初级诊断阶段,分析了传统BP网络的局限性和不足,对其逐一进行改进优化,并通过发动机故障实例验证了改进后的BP神经网络在故障诊断中的有效性。为了进一步提高故障诊断准确度,在上述研究的基础上,本文采用D-S证据理论对局部诊断结果进行决策级融合,并针对融合过程中存在的证据冲突问题,提出了加权优化方法。最后,本文将改进BP网络与加权D-S证据理论相结合的数据融合方法应用于发动机的故障诊断中,实验结果表明,该方法能够对发动机燃油系统中的燃油泵失灵、喷油器损坏和燃油压力调节器损坏等故障进行有效诊断,得出的故障诊断结果与实际故障相吻合,充分验证了该方法在发动机故障诊断中的可行性。
[Abstract]:Engine equipment belongs to a typical multi-level system, and its working state is often affected by many uncertain factors, such as current environment, operating conditions and so on. These complex characteristics determine the diversity and fuzziness of engine faults, which make the fault identification rate of traditional diagnosis methods low. For complex cases, it is often necessary to remove the machine to check, which further increases the economic losses caused by shutdown. Therefore, it is necessary to study a fast and accurate fault diagnosis method on the premise that the engine does not disintegrate. Aiming at the difficulty of engine fault detection, this paper applies information fusion technology to the field of engine fault diagnosis, and presents a decision level information fusion method combining artificial neural network and D-S evidence theory. This method is mainly composed of fault signal feature extraction, fault primary diagnosis and fault decision level fusion diagnosis. For the feature extraction part of the fault signal, this paper proposes a denoising method based on the wavelet detail coefficient autocorrelation analysis. The original signal is separated by Mallat wavelet, and the useful information in the fault signal is retained to a greater extent. It lays a foundation for time-frequency feature extraction in fault diagnosis. In order to diagnose the fault more comprehensively and accurately, this paper also introduces the characteristics of engine instantaneous speed and cylinder pressure signal, which provides three different feature vectors for subsequent diagnosis to facilitate fusion diagnosis. In the primary fault diagnosis stage, the limitations and shortcomings of the traditional BP neural network are analyzed, and the improved BP neural network is improved and optimized one by one. The effectiveness of the improved BP neural network in fault diagnosis is verified by an engine fault example. In order to further improve the accuracy of fault diagnosis, based on the above research, the D-S evidence theory is used to fuse the local diagnosis results at decision level, and a weighted optimization method is proposed to solve the problem of evidence conflict in the process of fault diagnosis. Finally, this paper applies the improved BP network and the weighted D-S evidence theory to the engine fault diagnosis. The experimental results show that the method can deal with the failure of the fuel pump in the engine fuel system. The damage of injector and the damage of fuel pressure regulator are effectively diagnosed. The result of fault diagnosis is in agreement with the actual fault, which fully verifies the feasibility of this method in engine fault diagnosis.
【学位授予单位】：哈尔滨理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP202;TK407

【参考文献】