类间数据不均衡条件下基于平衡随机森林的轴向柱塞泵故障诊断方法

针对轴向柱塞泵实际故障诊断中采集到的故障类数据远少于正常类数据的情况,为提升故障分类精确率,提出了一种基于平衡随机森林(Balanced Random Forest,BRF)的轴向柱塞泵故障诊断方法。BRF算法是随机森林(Random Forest,RF)的改进算法,将欠采样方法与RF结合,强化了RF处理非均衡数据的能力。通过开源的UCI数据集对该算法的性能进行了测试,相较于RF以及合成少数类过采样(Synthetic Minority Over-sampling Technique,SMOTE)与RF的组合算法SMOTE-RF,BRF算法在少数类分类精确率方面有所提升。最后,将BRF算法应用于轴向柱塞泵的故障诊断中。结果表明,在类间数据不均衡的条件下,相较于RF及SMOTE-RF算法,BRF算法能够取得更高的故障分类精确率。

Fault Diagnosis Method of Axial Piston Pump Based on Balanced Random Forest Under Imbalanced Datasets

Aiming at the situation that fault data collected in actual fault diagnosis of axial piston pump is far less than normal data, an axial piston pump fault diagnosis method based on Balanced Random Forest (BRF) algorithm is proposed to improve the classification precision of faulty classes. BRF algorithm is an improved algorithm of Random Forest (RF) which combines under-sampling method with RF and strengthens the ability of RF to process imbalanced data. The algorithm performance is tested on the UCI open source datasets. Compared with RF and SMOTE-RF, BRF algorithm improves the precision of minority classes. Finally, BRF algorithm is applied to the axial piston pump fault diagnosis. The results show that BRF has higher classification precision of faulty classes than RF and SMOTE-RF under imbalanced datasets.