基于主成分分析和集成神经网络的发动机故障诊断模型研究

针对发动机废气排放参数和故障之间复杂的非线性关系，提出了一种基于主成分分析和集成神经网络技术的发动机故障诊断分析模型。该模型首先运用主成分分析方法降低故障诊断样本的输入维数，然后按发动机不同运转状态将样本分组，并用于子网络训练；故障诊断时，各子网络分别诊断出相应的结果，最后采用投票法融合各输出结果。试验结果表明，这种模型能有效简化训练样本和样本属性参数，优化网络结构，其诊断精度及学习能力优于单一神经网络诊断模型，能较好地解决网络规模大、训练速度慢、诊断精度低等缺点。

Fault diagnosis model for engines based on principal component analysis and integrated neural network

Aimed at the complicated non-linear relationship between the emission parameters of the engine and their faults, a fault diagnosis model for engines based on principal component analysis (PCA) and integrated neural network was developed. First, PCA was used to reduce the dimension of the input numbers. Second, all samples were classified into different groups by the operation condition of engine, and then were trained on the sub-networks. Lastly, the different diagnosis conclusions resulted from different sub-networks were combined and outputted by voting method. Experimental results demonstrate that the model can simplify the training samples and their characteristic parameters, optimize the structure of the network, and improve the diagnosis precision, and study ability is superior to unitary neural network model, which can solve the problems of the former neural network with the shortcomings of too large scale networks, slow training speed, low diagnosis precision, etc..