基于神经网络的汽油车三效催化器老化诊断

Aging Diagnosis on Three-Way Catalytic Converter of Gasoline Vehicle Based on Neural Network

随着汽油车后处理系统复杂程度的提高,三效催化器的故障诊断需要解决建模难度大、标定成本高等问题。结合神经网络在处理非线性问题上的优势,提出一种基于神经网络的汽油车三效催化器老化诊断算法。算法以三效催化器的老化机理为基础,采集催化器前后氧传感器信号作为特征输入,结合不同故障进行编码,构成网络训练和测试所需的数据集;分别应用反向传播神经网络(Back Propagation Neural Network,BPNN)和深度置信网络(Deep Belief Network,DBN)的模型架构,对训练过程各参数进行调优,并对诊断结果进行测试。试验结果表明:基于神经网络的诊断算法建模简便,具有较高的诊断精度和泛化能力;诊断架构上,DBN相对于BPNN简化了特征提取过程,拥有更高的诊断精度。

With the complexity increase of gasoline vehicle after-treatment system,the problems such as large modeling difficulty and high calibration cost should be solved during the fault diagnosis process of three-way catalytic converter.Based on the advantages of neural network in dealing with non-linear problems,a neural network based aging diagnosis algorithm for three-way catalytic converter of gasoline vehicle was proposed.According to the aging mechanism of three-way catalytic converter,the oxygen sensor signals before and after the catalyst were collected as feature inputs and the data set required for network training and testing was determined combined with different fault codes.Back propagation neural network(BPNN)and deep belief network(DBN)were used separately to optimize the parameters of training process and test the diagnostic results.The experimental results show that the neural network-based diagnosis algorithm is simple in modeling and has high diagnostic accuracy with good generalization capability.From the perspective of diagnostic framework,DBN simplifies the feature extraction process and has higher diagnostic accuracy than BPNN.