【作者】 崔江； 王友仁；

【Author】 JIANG Cui YOUREN Wang Nanjing University of Aeronautics and Astronautics,College of Automation and Engineering,Nanjing 210016,China

【机构】 南京航空航天大学自动化学院；

【摘要】 针对模拟电路的故障诊断问题,讨论了一种基于有向无环图支持向量机分类器(DAGSVC)的故障字典新方法,并提出一种评估支持向量机分类器的测试复杂度指标。首先,对电路施加一定的测试激励,利用可测节点采集电路响应;其次,对采集的电路响应进行特征提取形成训练样本,并建立"l-v-l"SVMs进行训练,训练完毕并构建故障字典;最后,建立DAGSVC结构对未知样本加以测试。通过对一个模拟滤波器的实际测试和验证表明,本文方法性能要优于"l-v-r"SVC、"l-v-l"SVC和神经网络,适合模拟电路的故障分类和诊断。更多还原

【Abstract】 Focusing on the problem of diagnosing analog circuit,this paper discusses a new method of fault dictionary based on Directed Acyclic Graph SVMs classifier (DAGSVC),and a specification for estimating the test complexity of the Support Vector Classifier is also defined here.First,the Circuit Under Test (CUT) is stimulated by testing signal,and the response of the CUT is collected at the accessed nodes.Secondly,the collected responses are preprocessed with some feature extraction techniques to obtain training samples,and the"1-v-1"SVC are also established to train the samples;after training,the fault dictionary is constructed with the support vectors,etc..Finally,the DAGSVC is designed to test the unknown samples.An actual analog circuit is tested to validate the proposed method,whose performance is proven to be superior to the traditional methods,such as"I-v-r"SVC,"1-v-1"SVC,and Neural Networks classifier.更多还原