Fault Diagnosis of Plunger Pump in Truck Crane Based on Relevance Vector Machine with Particle Swarm Optimization Algorithm
Fault Diagnosis of Plunger Pump in Truck Crane Based on Relevance Vector Machine with Particle Swarm Optimization Algorithm
复制标题
DOI：
10.3233/sav-130784
复制
发表时间：
2013
影响因子：
1.6
通讯作者：
Wenliao Du;Ansheng Li;Pengfei Ye;Chengliang Liu
中科院分区：
文献类型：
--
作者：
Wenliao Du;Ansheng Li;Pengfei Ye;Chengliang Liu
快速准确地处理设备故障对提高设备可靠性、减少停机时间具有重要意义。提出了一种基于粒子群优化算法的相关向量机（RVM）的汽车起重机柱塞泵故障诊断新方法PSO-RVM。利用粒子群优化算法确定RVM中核函数的核宽度参数，并训练5个二叉树结构的两类RVM进行机构状态识别。将该方法应用于汽车起重机柱塞泵故障诊断中。采用正常状态、轴承内圈故障、轴承滚子故障、柱塞磨损故障、推力盘磨损故障和斜盘磨损故障6种状态，对PSO-RVM模型的分类性能进行了测试，并与BP神经网络、蚁群神经网络、RVM、和支持向量机与粒子群优化算法（PSO- SVM）。实验结果表明，PSO-RVM上级前三种经典模型，与PSO-SVM具有相当的性能，相应的诊断准确率分别达到99.17%和99.58%。但PSO-RVM模型中相关向量的数量远少于支持向量，约为支持向量的1/12-1/3，更适合于低复杂度、实时监控的应用。
Promptly and accurately dealing with the equipment breakdown is very important in terms of enhancing reliability and decreasing downtime. A novel fault diagnosis method PSO-RVM based on relevance vector machines (RVM) with particle swarm optimization (PSO) algorithm for plunger pump in truck crane is proposed. The particle swarm optimization algorithm is utilized to determine the kernel width parameter of the kernel function in RVM, and the five two-class RVMs with binary tree architecture are trained to recognize the condition of mechanism. The proposed method is employed in the diagnosis of plunger pump in truck crane. The six states, including normal state, bearing inner race fault, bearing roller fault, plunger wear fault, thrust plate wear fault, and swash plate wear fault, are used to test the classification performance of the proposed PSO-RVM model, which compared with the classical models, such as back-propagation artificial neural network (BP-ANN), ant colony optimization artificial neural network (ANT-ANN), RVM, and support vector machines with particle swarm optimization (PSO- SVM), respectively. The experimental results show that the PSO-RVM is superior to the first three classical models, and has a comparative performance to the PSO-SVM, the corresponding diagnostic accuracy achieving as high as 99.17% and 99.58%, respectively. But the number of relevance vectors is far fewer than that of support vectors, and the former is about 1/12-1/3 of the latter, which indicates that the proposed PSO-RVM model is more suitable for applications that require low complexity and real-time monitoring.