一种基于多维特征提取的知识化制造系统故障监测方法  

An approach to fault detection in knowledgeable manufacturing systems based on multi-dimension feature extration

作　　者：施文武[1] 严洪森[1] 汪峥[1] 熊福力[1] 

机构地区：[1]东南大学自动化研究所,南京210096

出　　处：《制造业自动化》2006年第2期33-38,共6页Manufacturing Automation

基　　金：国家自然科学基金资助项目(60574062;50475075);江苏省高校"青蓝工程"第二期省级中青年学术带头人培养计划

摘　　要：为了减少对知识化制造系统生产故障的误诊和漏诊,本文首先提出了一种新的基于多维特征提取的故障监测方法,该方法先对采集的信号进行多维特征提取,再通过归纳学习建立设备的正常状态空间,并以此来判断设备的故障状态。然后,提出一种故障误判概率控制方法。通过多维特征提取和误判概率控制,可以很好地减少对故障的漏判和误判。仿真实验证实了该故障监测方法的有效性。In order to minimize the probability of false diagnosis in knowledgeable manufacturing systems, a new approach based on inductive learning to fault detection is proposed in this paper. Firstly, multi-dimension features are extracted from sampled signals and compose a normal state space of equipment by inductive learning, according to which an abnormal signal can be detected. A method of controlling the probability of false diagnosis is then proposed to greatly decrease it.The effectiveness of the approach is verified by simulation examples.

关 键 词：知识化制造 归纳学习 多维特征提取 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]