基于联想记忆神经网络的故障识别

发布时间：2018-05-11 14:21

本文选题：离散Hopfield网络 + 联想记忆　；参考：《燕山大学》2012年硕士论文

【摘要】：工业生产、工程机械、航天、船舶中因某一关键设备发生故障，常常造成巨大的经济损失甚至灾难性事故，故障识别技术可以分析故障，防患于未然，减少损失。液压泵是液压系统的一个重要组成部分，其性能好坏直接影响着液压系统工作的可靠性和稳定性，所以对其进行故障识别的研究具有很重要的现实意义。本文对具有联想记忆功能的神经网络技术进行研究，并对轴向柱塞泵的故障进行识别。首先，研究了离散Hopfield神经网络，对结构和有关收敛稳定性的理论进行了详细的探讨；另一方面对联想记忆的功能实现和概念进行了具体描述，在应用中探讨了输入信号必须是二值型的弊端，从而结合BP网络强非线性处理的优点，对网络结构进行合理的拟合，在实际中可以得到普遍应用。其次，针对联想记忆神经网络易陷入局部极小值的特点，引入粒子群算法对网络权值进行了优化，得到了收敛性能较高的网络。在联想记忆神经网络故障识别之前，对网络中参数：层数，隐层神经元个数，学习速率，惯性权重，加速因子等进行了试探性的确定。最后应用本文提出的联想记忆神经网络BP-HNN和BP-HNN-PSO对液压泵的各种故障进行识别和分析比较，验证了方法的有效性，，并发现加入粒子群算法的联想记忆神经网络的识别结果值较高，识别率较高，较可靠。为了减少故障识别中的“错分”现象，利用艾宾浩斯记忆遗忘曲线对学习样本进行交叉循环安排，提高了学习的记忆效果，一定程度上达到了减少“错分”的目的。
[Abstract]:In industrial production, construction machinery, spaceflight and ship, the failure of a certain key equipment often results in huge economic loss or even catastrophic accident. The fault identification technology can analyze the fault, prevent the trouble from happening, and reduce the loss. Hydraulic pump is an important part of hydraulic system. Its performance directly affects the reliability and stability of hydraulic system. In this paper, the neural network technology with associative memory is studied, and the fault of axial piston pump is identified. Firstly, the discrete Hopfield neural network is studied, the structure and the theory of convergence stability are discussed in detail, on the other hand, the functional realization and concept of associative memory are described in detail. This paper discusses the disadvantage that the input signal must be a binary type in application, thus combining the advantages of strong nonlinear processing of BP network, the network structure is fitted reasonably, which can be widely used in practice. Secondly, aiming at the characteristic that associative memory neural networks are prone to fall into local minima, particle swarm optimization algorithm is introduced to optimize the weights of the networks, and a network with high convergence performance is obtained. Before the fault identification of associative memory neural network, the parameters of the network, such as the number of layers, the number of hidden layer neurons, the learning rate, the inertia weight, the acceleration factor and so on, are determined tentatively. Finally, using the associative memory neural network BP-HNN and BP-HNN-PSO presented in this paper to identify and compare the various faults of hydraulic pump, the validity of the method is verified, and it is found that the recognition result of associative memory neural network with particle swarm optimization algorithm is higher. The recognition rate is high and reliable. In order to reduce the phenomenon of "wrong points" in fault identification, the learning samples are arranged by using the Albinhaus memory forgetting curve, which improves the memory effect of learning and to some extent achieves the purpose of reducing the "wrong scores".
【学位授予单位】：燕山大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TH165.3;TP183

【参考文献】