【摘要】 针对传统旋转机械故障诊断方法中单一信息不能准确地描述复合故障类型及程度，以及单一机器学习模型面对复合故障出现的诊断精度低，泛化能力差且性能提升有限等问题，提出基于多规则轮询式(MCRM)改进的均匀流行逼近与投影算法(UMAP)的异构集成学习网络(MCRM-UMAP-Staking)的故障诊断方法。该方法首先对多源数据进行数据挖掘以克服信息冗余和冲突，然后利用异构集成机器学习模型对齿轮箱进行故障诊断，最后通过试验评估该故障诊断方法的可行性。实验结果表明，所提方法在不同程度的复合故障下的故障诊断精度高于98%,所提特征工程措施使模型训练时间减半，性能明显优于其他方法。更多还原

【Abstract】 The method of fault diagnosis based on Multi-Criteria Modified Round-Robin Uniform Flow Approximation and Projection Algorithm(MCRM-UMAP-Stacking) was proposed to address the issues encountered in traditional rotating machinery fault diagnosis methods, such as the inability of single information to accurately describe compound fault types and severity, as well as the low diagnostic accuracy, poor generalization ability, and limited performance improvement of single machine learning models in dealing with compound faults. Firstly, this method conducted data mining on multi-source data to overcome information redundancy and conflicts. Then, heterogeneous ensemble machine learning models were utilized for gear fault diagnosis. Finally, the feasibility of the proposed fault diagnosis method was evaluated through experiments. Experimental results demonstrated that the proposed method achieves a fault diagnosis accuracy of over 98% under compound faults at different levels. Moreover, the feature engineering measures proposed reduce the model training time by half, and the performance is significantly superior to other methods.更多还原