针对涡轮泵,提出了基于2维能量特征的快速SVM实时故障检测算法。以70为步长,以时域能量和能量变化率为故障特征,计算原始训练样本集。算法采用条件正定核函数计算原始训练样本集中正常与故障样本间的距离,筛选得到边界训练样本集,以此计算支持向量并构造决策函数。该算法对各1000个正常和故障样本进行训练,仅用时0.42s。对于检测数据,提前关机时刻3.02s报警。该算法提高了训练与分类速度,具有良好的实时性与准确性。

A fast support vector machine (SVM) based on two-dimensional energy features were proposed for turbo-pump. With step length 70, the algorithm computed energy and change rate to constitute original training samples set. The algorithm computed the distance between each normal sample and each fault sample by using the conditionally positive definite kernel, chose the boundary samples to construct a new training samples set, and got the support vectors and classifier by training. The algorithm trained 1 000 normal samples and 1 000 fault samples using only 0.42 s. For the detection data, the algorithm alarmed 3.02 s ahead of the shutdown time. The algorithm improves training speed and classification speed, and it has good accuracy and real-time performance.