基于ANN模型和HPSO算法的数控机床可靠性分布模型研究
Numerical control machine reliability distribution model research based on the ANN model and HPSO algorithm
作者：付涛(集美大学机械与能源工程学院)；弓清忠(集美大学机械与能源工程学院)；王大镇(集美大学机械与能源工程学院)；祁丽(集美大学机械与能源工程学院)
Author：FU Tao(School of Mechanical Engineering, Jimei University)；GONG Qin-Zhong(School of Mechanical Engineering, Jimei University)；WANG Da-Zhen(School of Mechanical Engineering, Jimei University)；QI Li(School of Mechanical Engineering, Jimei University)
收稿日期：2014-05-04 年卷（期）页码：2015,52(2):262-268
期刊名称：四川大学学报: 自然科学版
Journal Name：Journal of Sichuan University (Natural Science Edition)
关键字：神经网络; 混合粒子群算法; 极大似然法; 参数估计
Key words：Neural network; HPSO; Maximum likelihood method; Parameter estimation
基金项目：福建省产学研重大项目资助（2012H6016）； 福建省自然科学基金计划资助项目（2011J01321）
中文摘要
针对数控机床可靠性研究过程中, 由于可靠性数据较难收集导致可靠性分布模型不唯一的问题, 采用ANN模型对收集到的少量可靠性数据进行扩充, 扩充后的数据采用K S检验法进行分析以确定可靠性数据模型, 同时在求解确定可靠性分布模型参数过程中, 将混合粒子群优化(hybrid particle swarm optimization, HPSO)算法引入极大似然估计中, 解决其在小样本数据下求解某些复杂分布模型时易陷于局部最优解和求解效率低的问题. 实例分析结果表明: 采用混合粒子群算法求解可以在求解效率和收敛性性能上达到较好的平衡, 比较所有的求解模型结果, 经过ANN模型扩充后的2重3参数威布尔分布的相对均方差最小, 其值为0. 0425, 说明利用该方法求解数控机床的可靠性分布模型是可行的, 而且能够获得较精确的结果.
英文摘要
According to the problem that reliability distribution model is not unique in reliability research of numerical control machine tool due to reliability datas are difficult to collect, ANN model is adopted to expand the collected reliability data which is then analyzed by K S inspection to determine the reliability data model. At the same time, hybrid particle swarm optimization is used to maximum likelihood estimation in the process of solving the reliability distribution model to solve the problem that maximum likelihood estimation is easily to trappe in local optimal solution and low efficient with small samples. The results show that hybrid particle swarm algorithm does good performance in solution efficiency and convergence. Comparing all the results, the NRSE of 2 fold 3 parameter Weibull distribution of ANN extended models is 0. 0425, proving that this method is feasible to estimate the mixed Weibull distribution parameters and more accurate results can be obtained
【关闭】
