基于神经网络的发动机早期故障诊断研究

Early Fault Diagnosis of Engine Based on Neural Network

采用BP(Back Propagation)神经网络,借助德国莱尔浩福二合公司(Reilhofer KG)的德尔塔分析仪(delta-ANALYSER),将对发动机进行等角度采样的振动信号进行阶次分析所得到的阶次谱作为神经网络的输入,故障代码作为输出,对发动机进行早期故障诊断。结合企业自身优势,通过数十台发动机的台架耐久试验对神经网络进行了训练和验证。结果表明,经过大量案例训练过的神经网络对该企业发动机耐久试验过程中常出现的排气门断裂故障以及拉缸的诊断率达到95%,神经网络配合德尔塔分析仪进行发动机早期故障诊断达到了预期的效果,为发动机台架耐久试验早期故障诊断工作提供了一个有价值的方法。

Using BP （Back Propagation） neural network, with the help of （Reilhofer KG） delta-ANALYSER, the vibration signal of order spectrum was taken as the input of neural network, and the fault code as output, for diagnosis of the early fault of engines in the engine endurance test. Combining the advantages of the enterprise itself, the neural network was trained and validated by dozens of engine tests on the test bench. The results show that, after a large number of cases, the fault diagnosis rate of the trained neural network to failure of the valve and cylinder scoring reached more than 90%, the neural network combined with the delta analyzer achieved the desired results for early fault diagnosis of engine. It provides a valuable method for early fault diagnosis of engine endurance test on the test bench.