航空发动机振动趋势预测的过程神经网络法  

Prediction of Aeroengine Vibration Trend Using Process Neural Network

作　　者：金向阳[1,2] 林琳[1] 钟诗胜[1] 丁刚[1] 刘义翔[2] 

机构地区：[1]哈尔滨工业大学机电工程学院,哈尔滨150001 [2]哈尔滨商业大学轻工学院,哈尔滨150028

出　　处：《振动．测试与诊断》2011年第3期331-334,396-397,共4页Journal of Vibration,Measurement & Diagnosis

基　　金：国家高技术研究发展计划("八六三"计划)重点资助项目(编号:2009AA043404);国家自然科学基金重点资助项目(编号:60939003;50805032)

摘　　要：提出一种基于过程神经网络思想的航空发动机振动趋势预测方法。利用过程神经网络具有输出函数对输入函数在时间上的聚合效应和非线性映射能力,预测方法的网络结构选择为9个输入节点,第2层和第3层各有9个隐层节点,1个输出节点,参数外推预测,将选取的振动历史数据分为学习样本和检测样本两组,学习样本用于网络训练,检测样本用于检验预测模型的精度。在相同条件下,与传统人工神经网络进行趋势预测比较,提高了网络训练速度,降低了预测误差。将所提出的预测方法应用到某型航空发动机的振动趋势预测中,预测结果与实际值的误差符合要求。A prediction for about aeroengine vibration trend is proposed based on the process neural network.The analysis of vibration trend is influenced by many complicated factors during the practical operation period of aeroengines.It is difficult for the traditional methods to predict vibration change tendency effectively.Utilizing the polymerization effect and continuous input-output mapping of the system realized by nonlinear mapping capability to the time variable of process neural networks,the prediction method for aeroengine vibration trend had one input and output node,nine hidden nodes.Historical data are separated as learning samples and detection samples.The corresponding network model and learning algorithm are given.Under the same condition,compared with the traditional artificial neural network,the new network training speed improves and the prediction error to be decreased.Finally,the prediction method with the corresponding learning algorithm is used to predict the vibration trend of some aeroengines.Results are satisfactory.

关 键 词：航空发动机 振动预测 过程神经网络 学习算法 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] V263.6[自动化与计算机技术—控制科学与工程]