基于GBDT的城轨塞拉门早期机械故障诊断

Early Mechanical Fault Diagnosis of Urban Rail Door Based on GBDT

为了提高城轨塞拉门系统的智能化水平,增强城轨塞拉门运行的安全性、可靠性,提出一种基于ReliefF优化梯度提升决策树的塞拉门早期故障诊断方法,对塞拉门故障隐患进行实时有效监测和排查.在不同早期故障类型下进行城轨塞拉门早期故障试验,利用传感器采集塞拉门电机转速、转矩以及电流,通过特征工程建立数据集;运用ReliefF多分类特征筛选并从中提取故障特征,与主成分分析法、线性判别分析相比,ReliefF特征筛选表现最优;使用梯度提升决策树算法进行模型训练与早期故障类型识别,采取网格调参与五折交叉验证法进行超参数寻优,缩短运算时间.采用集中学习投票机制,得出最终分类结果,模型精度达到90%以上,具有鲁棒性.

In order to improve the intelligent level of an urban rail door system and enhance the safety and reliability of urban rail door operation,it is necessary to effectively monitor and troubleshoot the potential faults of such doors in real time.An early fault diagnosis method for such doors based on ReliefF optimized gradient lifting decision tree is proposed.Early fault experiments of urban rail door are carried out under different early fault types.Motor speed,torque and current of the door are collected by different sensors,and a data set is established by using feature engineering.Compared with the other two feature fusion methods,PCA(principal component analysis)and LDA(linear discriminant analysis),ReliefF has the best feature selection performance.GBDT(gradient lifting decision tree)algorithm is used for model training and early fault type identification,and network pattern is used in five-fold cross verification for hyperparameter optimization,which improves the operation rate and greatly shortens the operation time.Due to the existence of three classifiers,a voting mechanism is adopted to obtain the final classification results.The model is robust with a test accuracy of more than 96%.