基于大规模训练集SVM的发动机故障诊断

Fault diagnosis method for aero-engine based on SVM with large-scale training set

提出了一种新的学习策略,用于解决发动机故障诊断中大规模支持向量机(SVM)的训练问题.通过保留初始SVM分类器支持向量超平面附近的样本以及错分样本,使最终得到的约减集规模明显缩小,从而可在保持较高分类精度的前提下使训练时间明显缩短;同时,由于支持向量的数量减小,分类时间也相应缩短.探讨了序贯最小优化(SMO)算法的参数选择和实现过程中的关键问题,为这种极具潜力的算法在发动机故障诊断中的实际应用奠定了坚实的基础.仿真实例表明,这种基于大规模训练集SVM的发动机故障诊断方法有效、可靠,容易实现,可以作为工程应用的基础.

A learning strategy was presented to solve the large-scale support vector machines（SVM） training problem of aero-engine fault diagnosis.According to the strategy,only those samples were retained for final training,which were near the support vector hyperplane of the original small-scale SVM classifier or classified mistakenly by the original classifier.The final pruning set was reduced evidently and the training time was shortened obviously on the condition of high classification accuracy.Meanwhile,the classification time was shortened correspondingly because of the decrease of the support vector number.Something about the parameter selection of the sequential minimal optimization（SMO） method and how to use this potential algorithm in aero-engine fault diagnosis were also discussed.Simulation examples show that this fault diagnosis method proposed for aero-engine,based on SVM with large-scale training set,is effective,reliable and easy to be implemented for engineering application.