钻井用柴油发动机特征挖掘及故障诊断

Characteristic mining and failure diagnosis of the drilling diesel engine

柴油发动机状态监测及故障诊断的关键环节是特征参数的挖掘。提出了一种将距离测度和奇异值分解相结合的特征挖掘方法,将该方法应用到发动机的特征挖掘中,并将挖掘到的特征作为BP神经网络的输入向量用于发动机的故障诊断。结果表明,根据类间距离和类内距离之比定义的特征敏感度可以很好地区分不同的类,基于距离测度和奇异值相结合的方法可以优化特征,提高诊断的准确率。利用训练好的BP神经网络对现场12台发动机进行故障诊断,有11台诊断结论完全正确,诊断的准确率为91.67%。

The key step of diesel engine state monitoring and failure diagnosis is the mining of characteristic parameters.A kind of characteristic mining method which combines distance measurement and singular value decomposition was put forward.The method was applied to the mining of engine characteristics which were in turn used as the input vector of the BP neural network to diagnose the failure of engine.The result shows that different classes can well be distinguished according to the characteristic sensitivity of the definition of the ratio between inter-class distance and intra-class distance.The method based on the combination of distance measurement and singular value can optimize the characteristics and improve the accuracy of diagnosis.The well-trained BP neural network was adopted to conduct a field failure diagnosis of 12 engines and the diagnosis conclusion of the 11 engines was completely correct.The accuracy rate of diagnosis was 91.67%.