摘 要 乳化液泵站是矿山井下综合机械化采煤系统的重要组成部分，一旦发生故障给企业造成很大损失。所以，对乳化液泵站进行状态检测及故障诊断对于预防事故的发生，保证综采工作面的正常运行和安全生产有十分重要的意义。 振动信号蕴含的信息丰富，它的获取可用非接触式测量，便于现场实施。本文采用PC计算机、电涡传感器、声卡建立了振动数据采集系统。试验结果发现当排液阀弹簧损坏时，其振动信号的频谱在低频段的能量较低，而正常时则较高。进一步提出了在频域对信号提取故障特征量的方法，即取倍频程分别为31.5、63、125、250、1000Hz处六个频段的能量作为特征量。 人工神经网络模拟人的认知过程，通过一定的训练样本进行自学习、自调整，使网络自动形成训练样本中蕴含输入输出关系的映射。本文通过引入种群熵的概念把粒子群算法和神经网络的BP算法结合起来，提出了一种新的混合算法PSO-BP混合优化算法，从而充分利用了粒子群算法和神经网络的BP算法的各自具有的优点，并且把PSO-BP混合优化算法应用到故障诊断中去。最后用Visual Basic6.0将乳化液泵站故障诊断系统以友好的界面展现给用户。 结果表明，本文提出的特征提取方法及利用PSO-BP网络进行智能诊断的方法效果良好，并可以进一步的推广应用到其它往复机械的状态检测及故障诊断。

ABSTRACT The emulation pump plays an important role in integrated mechanical excavate coal in colliery. Once serious accident occurred, it would do great loss to the corporation. So it has great value to take condition monitoring and fault diagnosis to the emulation pump in order to prevent accident and make the machine to work smooth. The vibration signal has plenty of information of machine, it is noncontact to the machine when the vibration is being collected, and so it is convenient to pick up the vibration signal when the machine is running. With PC, eddy current sensor and sound card, a vibration signal collecting system was sacredly designed. After the research of the valve’s vibration, it was found that the fault signal’s power in the low frequency domain is lower than the normal signal, In this paper, the method of fault features extraction is pointed out. The valve’s vibration signal’s six-frequency section’s power in the low frequency domain is picked up as the fault feature. The artificial neural network can simulate the cause of human being’s perceiving. It can learn from the training samples and adjust the network by itself. After learning from the training set, the network can automatically form the mapping of the relationship between the network’s input and output. A new algorithm call PSO-BP optimal algorithm is presented in this paper by combining the particle swarm optimizer and network according to the nation of population entropy, so that people can make good use of the respective virtues. In order to solve the problem of the fault diagnosis of emulation pump, PSO-BP optimal algorithm is applied in this paper. The system of the fault diagnosis of emulation pump designed by visual basic 6.0 is displayed to users with friendly interface. The results indicate that the method used in feature extraction is effective; the PSO-BP optimal algorithm is capable of intelligent diagnosing. Both of them can be applied to the element’s condition monitoring and fault diagnosis in other reciprocating machine.