涡轮泵实时故障检测的快速支持向量机算法  

Fast support vector machine algorithm for turbopump real-time fault detection

作　　者：洪涛[1] 黄志奇[2] 杨畅[3] 

机构地区：[1]电子科技大学空天科学技术研究院,成都611731 [2]电子科技大学自动化工程学院,成都611731 [3]电子科技大学英才实验学院,成都611731

出　　处：《仪器仪表学报》2012年第8期1786-1792,共7页Chinese Journal of Scientific Instrument

基　　金：载人航天预先研究计划资助项目

摘　　要：提出了一种基于边界样本的快速支持向量机(support vector machine,SVM)算法用于液体火箭发动机涡轮泵实时故障检测。算法按一定步长将涡轮泵振动信号分段,再将每个步长信号平分为多段且计算每段信号的均方根、裕度因子和峭度,并将之组合为3维向量作为故障特征,以每个步长信号中的故障样本点数目作为判断故障的依据;算法采用条件正定核函数计算原始样本集中正常样本与故障样本之间的距离,选择边界样本作为新的训练样本集,并以此计算支持向量并构造决策函数。用某型号涡轮泵振动加速度信号对算法进行验证,结果表明对包含5 600个故障样本和5 600个正常样本的原始训练样本集,算法的训练时间为0.68 s。对时长20.80 s的待检信号,算法检出故障时刻为20.43 s,比故障真实出现时刻晚0.42 s(在0.5 s之内)。该算法大幅度提高了训练速度与分类速度,具备良好的精确性与实时性。A fast support vector machine （SVM） algorithm based on boundary samples was proposed for liquid rock et engine （LRE） turbopump realtime fault detection. The algorithm divides the turbopump vibration signal into some segments with appropriate step, then divides every step of the signal into some average segments, and computes the root mean square（RMS） , margin factor and kurtosis of each segment, which are used to construct a 3 dimension al vector as the fault feature ; and the fault sample number in every step is used to judge the fault. In original training sample set, the algorithm computes the distance between each normal sample and each fault sample using the condi tionally positive definite kernel, chooses the boundary samples to construct a new training sample set, from which the support vectors and classifier are obtained. A part of the vibration acceleration signal of a certain type of turhopump was chosen as the test object to validate the algorithm. The test results show that for an original training sample set that includes 5 600 fault samples and 5 600 normal samples, the training time length is 0.68 s; for the test signal with a duration of 20.80 s, the algorithm detects the fault at 20.43 s, which is 0.42 s （ less than 0.5 s） later than the time when the fault really occurs. The algorithm improves the training speed and classification speed greatly, and has good accuracy and realtime performance.

关 键 词：涡轮泵 实时故障检测 快速支持向量机 边界样本 

分 类 号：TP206.3[自动化与计算机技术—检测技术与自动化装置]