航空发动机磨损趋势变权重组合预测技术研究

摘要

由于航空发动机滑油中金属元素含量受许多复杂因素的影响,所以磨损趋势预测精度相对较低.针对这个问题提出了RBF网络变权重组合预测(RBFNN-VWCF)模型对航空发动机零部件的磨损趋势进行研究.由于输入维数对模型的预测精度影响较大,引入混沌理论中的C-C方法重构相空间确定模型最佳输入输出样本的维数,选取BP网络和SVM模型作为子预测模型对铁元素含量的变化趋势进行预测,将得到的预测值作为RBFNN-VWCF模型的输入变量进行变权重组合预测,利用正交最小二乘法训练网络模型,确定子模型不同时刻的权重,并对影响模型预测精度的参数进行讨论.仿真结果表明,RBFNN-VWCF模型充分利用了两种子预测模型的有效信息,更客观地反映了发动机零部件的磨损趋势,与单一模型相比具有较高的预测精度和很强的实用性,为发动机下一步的维修决策提供了有力支持.%The contents of metal in the spectral oil of the aeroengine are affected by many complex factors, so wear trend prediction accuracy is relatively low.To solve this problem, an RBF neural network variable weight combination forecasting model (RBFNN-VWCF) is proposed for wear trend prediction of aircraft engine.As the dimensions of input layer affect the prediction accuracy severely,C-C method that originates from chaos theory is introduced to determine the dimensions of input samples and output samples.BP network and the SVM model are carried on as sub-prediction model to predict the trend of iron,and then put the prediction value of sub-model as input variable of RBFNN-VWCF model,orthogonal least square is used to train the RBFNN-VWCF model,determine the weight of sub-model at different times,at the same time parameters which affecting the prediction accuracy are discussed in this paper.The simulation results show that,RBFNN-VWCF model takes full advantage of effective information of sub-model, and it can be able to reflect the wear trend of engine more objectively. The forecasting result is robust,compared with the single model RBFNN-VWCF model has higher prediction accuracy and strong practical value,provides powerful support for decision-making of the engine next step.