基于不同工况下辅助数据集的齿轮箱故障诊断

摘要

针对变工况下齿轮箱监测数据重用性低,受复杂工况影响大和已训练模型经常失效的问题,提出基于不同工况下辅助数据集的迁移成分分析方法用于设备故障诊断.迁移成分分析(Transfer Component Analysis,TCA)通过核函数将训练样本与测试样本映射到潜在空间,进而减小训练样本与测试样本的分布差异性.重点对比分析训练数据中不同工况下辅助数据所占比例对迁移成分分析算法性能的影响,通过仿真分析和实验验证得出,迁移成分分析方法相比传统机器学习算法,明显地减小了训练样本与测试样本的分布差异,具有更高的监测数据重用率与更高的诊断准确率,有效提高了齿轮箱变工况故障诊断的准确率和可靠性.%In view of the low reusability of monitoring data,the influence of complex working conditions and the usual failure of trained model,the transfer component analysis (TCA) was introduced for the equipment fault diagnosis in variable working conditions based on auxiliary monitoring data of different working conditions.The TCA method decreases the distribution difference between training sample and test sample,by utilizing a kernel function to map the feature of the two samples to a latent space.Besides,the effect of different proportion of monitoring data under various working condition was compared and analyzed in TCA algorithm performance.According to the simulation analysis and experiment verification,compared with traditional machine learning methods,the TCA performs better in the reduction of distribution difference between the two kinds of samples and improves reusability of monitoring data under different operational conditions and gearbox diagnosis accuracy and reliability.