基于离散粒子群优化算法的汽车发动机故障特征选择

Vehicle engine fault feature extraction based on discrete particle swarm optimization algorithm

最优特征选择属于组合优化范畴,针对汽车发动机机械故障特征选择问题,分析了冗余特征的存在对于故障分类器分类性能的影响,选择最优特征组合可以提高故障分类的正确率,提出基于离散粒子群算法的特征优化组合算法,利用BP神经网络评价特征优化的性能,并将其应用到汽车发动机曲轴轴承磨损故障诊断中.实验结果表明,与遗传算法相比,基于离散粒子群算法的特征优化算法优化效率较高,分类正确率较高,优化后的特征集可以显著地提高故障分类器的分类性能.

The optimal feature extraction is subordinated to combined optimization.Specifically for the vehicle engine fault extraction,the impact of redundant features on the performance of fault classifiers is analyzed.Accordingly,the selection of optimal feature combinations can enhance fault classification accuracy.Based on this notion,an optimization algorithm for feature combination is proposed based on the discrete particle swarm.By applying the BP neural network for performance evaluation on optimal features,the crankshaft bearing wear is detected in a vehicle engine.From experimental results,it is found that,compared with GA,the feature optimization algorithm via discrete particle swarm possesses higher efficiency and accuracy rate so that the optimized feature set can significantly promote the classification performance of fault classifiers.