【摘要】 光谱油样分析是机械磨损状态监测与故障诊断的重要技术,基于光谱数据的机械状态预测有利于发现机械系统的早期磨损故障。由于神经网络对于非线性模型的辨识和非平稳信号的预测,与传统预测模型相比具有明显的优势,文章将神经网络预测方法运用于光谱分析,提出了基于神经网络预测的光谱分析监测技术。在预测模型中采用了三层BP网络模型,针对神经网络的结构对于信号预测或模型辨识的精度具有影响很大的问题,文章利用遗传算法,对神经网络输入节点数、隐层节点数和网络收敛的均方误差(MSE)目标值进行了优化,得到了最优的网络预测模型。最后,对某发动机实际的光谱分析数据进行了预测和分析,并与传统ARMA模型的预测结果进行了比较,结果充分表明了本方法的有效性和优越性。更多还原

【Abstract】 The spectrometric oil analysis (SOA) is an important technique for machine state monitoring and fault diagnosis, and (forecasting) machine state through SOA results has an advantage of finding out machine system wear fault early. Because Artificial (Neural) Network (ANN) possesses obvious advantages over traditional forecasting models for identifying non-linear model and (forecasting) non-even signal, the ANN forecasting approach was applied to monitoring technique by SOA, and the monitoring technique by SOA based on ANN forecasting was put forward. In the forecasting model, a 3-layer BP network structure was adopted. Aiming at the problem that ANN structure has a great effect on forecasting precision, the authors utilized the Genetic Algorithm (GA) to optimize the node number of input layer, the node number of hidden layer, and MSE (Mean of Squared Error) target value which was required for ANN training, and obtained the optimum forecasting model of ANN. Finally, the practical SOA data of some engine was analyzed and forecasted by ANN, and the forecasting result was compared with that of traditional ARMA model. The result fully shows the (superiority) and effectivity of the new method.更多还原