彭继慎,董 晶

摘 要:针对现行研究中压铸机实时检测与控制系统中相关传感器的常见故障问题,通过对人工神经网络理论与方法的学习,建立了一种基于径向量基函数神经网络RBFNN的控制系统传感器故障诊断观测器模型。通过来自压铸机的实测参数进行模型训练,采用模糊K均值聚类算法选取聚类中心,利用该观测器确定传感器输出值与传感器实际输出值之间的残差,以此判断传感器是否发生故障。仿真结果表明,RBFNN观测器具有较强的非线性处理和任意函数逼近的能力,预测精度高,学习时间短,网络运算速度快,性能稳定,可满足传感器故障诊断的要求。

关键词:压铸机;RBFNN;故障诊断;模糊K均值聚类算法

中图分类号:TP212文献标识码:A

文章编号:1004-373X(2009)12-179-04

Method of Control System Sensor Fault Dignosis Based on RBF Neural Network

PENG Jishen,DONG Jing

(Faculty of Electrical and Engineering Control,Liaoning Technical University,Huludao,125105,China)

Abstract:For the problems of related sensor common fault in the die casting machine real-time detection and control system,through the study of artificial neural network′s theory and methods,sensor fault diagnosis observer model of control system based on a radial basis function neural network control system RBFNN is established.The model is trained by the measured parameters of die-casting machine,adopting fuzzy K means clustering algorithm to select clustering centre,using the observer to forecast the residuals between the sensor output value and the sensor actual output value to diagnose sensor fault.The simulated results show that RBFNN observer has strong capacity of non-linear process and arbitrary function approximation,and has many advantages such as high-precision,learning time short,rapid network computing and stable performance to meet the requirements of sensor fault diagnosis.

Keywords:die casting machine;RBFNN;fault dignosis;fuzzy K means clustering algorithm

0 引 言

传感器是现行研究的压铸机实时检测与控制系统[1]的关键部件,系统利用传感器对压铸机的各重要电控参数(如:合型力、油压、压射速度、模具温度等)进行检测,并进行准确控制。这一过程中,各传感器输出信号的质量尤为重要,其优劣程度直接影响压铸机控制系统分析、处理数据的准确性,最终影响压铸件产品质量的优劣。由于大型压铸机生产环境较为恶劣,长期的高温、高压、高粉尘及来自周边器械的电磁干扰等因素的存在,不可避免地会造成传感器软硬故障的发生,有故障的传感器所发出的错误信号,会使整个压铸机控制系统分析、处理和控制功能紊乱,造成系统无法正常运行,带来无法估计的生产安全隐患及严重的后果。……