基于集成深度学习的转辙机故障诊断研究

Research on Switch Machine Fault Diagnosis Based on Integrated Deep Learning

[目的]为了能够充分利用故障日志数据诊断转辙机故障,提出了基于集成学习算法的道岔转辙机故障诊断方法。[方法]通过分析转辙机故障文本数据,并结合专家经验,建立了两级故障诊断思路;将故障文本数据预处理为机器能够识别的数据,作为故障诊断模型输入数据;介绍了基于AdaBoost集成学习法的CNN(卷积神经网络)-LSTM(长短期记忆网络)故障诊断模型的原理和方法。[结果及结论]试验结果表明,在数据类别不平衡或者样本数量有限的情况下,采用CNN-LSTM模型能够有效提高故障诊断的准确率;与其他故障诊断模型相比,CNN-LSTM模型性能更好;所提出方法具有有效性,能够满足应用场景准确率要求。

[Objective]To make full use of fault log data for diagnosing switch machine faults,a fault diagnosis method based on integrated deep learning is proposed.[Method]By analyzing the textual data of switch machine faults and combining expert experiences,a two-level fault diagnosis approach is established.The fault text data is preprocessed into machine-readable data,serving as input data for the fault diagnosis model.The principle and method of the CNN-LSTM fault diagnosis model based on the AdaBoost integrated deep learning method are introduced.[Result&Conclusion]Experimental results demonstrate that under conditions of data class imbalance or limited sample size,the CNN-LSTM model can effectively improve the accuracy of fault diagnosis.Compared with other fault diagnosis models,the CNN-LSTM model performs better.The proposed method is effective and can meet the accuracy requirements of application scenarios.