机械工程学报
機械工程學報
궤계공정학보
CHINESE JOURNAL OF MECHANICAL ENGINEERING
2010年
2期
145-149
,共5页
径向基函数神经网络%可靠性%电火花线切割机床%聚类学习算法%柯尔莫哥洛夫-斯米尔诺夫检验法
徑嚮基函數神經網絡%可靠性%電火花線切割機床%聚類學習算法%柯爾莫哥洛伕-斯米爾諾伕檢驗法
경향기함수신경망락%가고성%전화화선절할궤상%취류학습산법%가이막가락부-사미이낙부검험법
Radial basis function neural network%Reliability%Wire electrical discharge machining%Cluster learning algorithm%Kolmogorov-smirnov test method
针对电火花线切割机床(Wire electrical discharge machining,WEDM)可靠性数据分布模型无法确定的问题,提出应用径向基函数(Radial basis function,RBF)神经网络对可靠性数据进行模拟仿真,扩大可靠性数据样本量,从而确定其分布模型的方法.选取聚类学习算法作为神经网络学习方法,通过无监督学习确定RBF神经网络中各隐节点的数据中心,并根据各数据中心之间的距离确定隐节点的扩展常数,然后用有监督学习训练各隐节点的输出权值.经过对原始可靠性数据进行拟合训练后生成一套RBF神经网络,随机产生100个可靠度数据输入该神经网络产生与原始可靠性数据具有相同失效统计规律的数据.对扩充后的可靠性数据通过图估计法和柯尔莫哥洛夫-斯米尔诺夫(Kolmogorov-smirnov, K-S)检验法确定电火花线切割机床可靠性数据分布模型为对数正态分布模型,同时对可靠性模型的参量估计更加准确.
針對電火花線切割機床(Wire electrical discharge machining,WEDM)可靠性數據分佈模型無法確定的問題,提齣應用徑嚮基函數(Radial basis function,RBF)神經網絡對可靠性數據進行模擬倣真,擴大可靠性數據樣本量,從而確定其分佈模型的方法.選取聚類學習算法作為神經網絡學習方法,通過無鑑督學習確定RBF神經網絡中各隱節點的數據中心,併根據各數據中心之間的距離確定隱節點的擴展常數,然後用有鑑督學習訓練各隱節點的輸齣權值.經過對原始可靠性數據進行擬閤訓練後生成一套RBF神經網絡,隨機產生100箇可靠度數據輸入該神經網絡產生與原始可靠性數據具有相同失效統計規律的數據.對擴充後的可靠性數據通過圖估計法和柯爾莫哥洛伕-斯米爾諾伕(Kolmogorov-smirnov, K-S)檢驗法確定電火花線切割機床可靠性數據分佈模型為對數正態分佈模型,同時對可靠性模型的參量估計更加準確.
침대전화화선절할궤상(Wire electrical discharge machining,WEDM)가고성수거분포모형무법학정적문제,제출응용경향기함수(Radial basis function,RBF)신경망락대가고성수거진행모의방진,확대가고성수거양본량,종이학정기분포모형적방법.선취취류학습산법작위신경망락학습방법,통과무감독학습학정RBF신경망락중각은절점적수거중심,병근거각수거중심지간적거리학정은절점적확전상수,연후용유감독학습훈련각은절점적수출권치.경과대원시가고성수거진행의합훈련후생성일투RBF신경망락,수궤산생100개가고도수거수입해신경망락산생여원시가고성수거구유상동실효통계규률적수거.대확충후적가고성수거통과도고계법화가이막가락부-사미이낙부(Kolmogorov-smirnov, K-S)검험법학정전화화선절할궤상가고성수거분포모형위대수정태분포모형,동시대가고성모형적삼량고계경가준학.
For determining the distribution model of wire electrical discharge machining (WEDM) reliability data, the radial basis function (RBF) neural network is applied to simulating the original reliability data, and more reliability data are achieved that have the same distribution rules with the original reliability data. The cluster learning algorithm is chosen as the learning method of the neural networks. The data centers of hidden nodes are determined by unsupervised learning, and the extended constants of the hidden nodes are determined by the distances of each data center, then the output weights of the hidden nodes are achieved by the supervised learning method. After simulating and calculating, the extended reliability data is achieved by the trained RBF neural networks, and the reliability distribution model of WEDM reliability data is confirmed as log-normal distribution model by the graphical estimation method and Kolmogorov-smirnov (K-S) test method. And it is more accurate for estimating characteristic parameters of the reliability distribution model.
