控制系统传感器故障的两次预测诊断方法  

TWICE PREDICTIVE METHOD OF CONTROL SYSTEM SENSOR FAILURE DIAGNOSES

作　　者：房方[1] 牛玉广[1] 孙万云[1] 魏乐[1] 

机构地区：[1]华北电力大学动力系,河北保定071003

出　　处：《中国电机工程学报》2001年第11期14-18,23,共6页Proceedings of the CSEE

摘　　要：为了克服建模误差对诊断结果的影响 ,增强故障检测方法的鲁棒性 ,同时充分利用生产过程中存在的大量冗余信息 ,文中提出了一种基于径向基函数网络 (RBFnet)的传感器故障诊断的新方法。针对多传感器系统的特点 ,利用神经网络的非线性拟合能力 ,将相关传感器的输出数据综合 ,对待诊断传感器的输出进行两次预测 :第一次预测用于故障的识别 ;第二次预测可以实现故障传感器的定位 ,并利用第一次预测的输出数据对故障信号进行恢复。仿真实验表明该诊断方法对于传感器的几种不同故障形式均能够进行识别和恢复 ,对于系统工况的变化具有一定的适应性。由于径向基函数网络具有很好的收敛性 ,同时在使用过程中采用离线训练、在线使用的方式 ,因此该方法具有较好的实时性。In order to depress the effect on diagnostic result caused by the model error, improve the robustness of failure diagnosis method, and make the best of a great deal of redundant information existed in process, this paper presents a new method of sensor failure diagnosis based on the radial basis function networ ks(RBFnet). To the characteristic of multi-sensor system, using the nonlinear a pproach ability of Neural Networks, synthesizing the output data of relational s ensors, makes the twice prediction outputs of the sensor that will be diagnosed. The first time prediction is used to identify the failure;The second time predi ction is used to locate the fault sensor and makes use of the first time predict i ve output data to resume the fault signal. Simulation tests show that this diagn osis m ethod can identify, locate and resume several kinds of failure forms, and it has adaptability to the change of system operating conditions. What's more, because of the well astringency of RBFnet, and its offline-training and online-using, this method has better real time quality.

关 键 词：控制系统 传感器 故障诊断 人工神经网络 

分 类 号：TM921.5[电气工程—电力电子与电力传动] TP212[自动化与计算机技术—检测技术与自动化装置]