基于状态和质量特征的半导体设备智能维护方法研究

项目名称： 基于状态和质量特征的半导体设备智能维护方法研究

项目编号： No.51275093

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 机械、仪表工业

项目作者： 敖银辉

作者单位： 广东工业大学

项目金额： 80万元

中文摘要： 半导体生产线是一个多品种、多批量，带"重入"的复杂制造系统，其设备维护是否得当直接影响生产。本项目以半导体复杂制造过程为对象，考虑生产计划，研究集设备状态、产品质量、历史维护数据为一体的智能维护策略。首先基于Markov决策过程模型进行单设备的维护调度方法研究，实现适应动态失效和变动作业的长期维护计划；其次，考虑生产线在制品WIP、库存、产品质量等变化，研究基于离散事件Monte Carlo仿真的多目标优化的维护策略，实现短期维护的时间点设置，适应设备实时状态的重调度。项目提出采用"虚拟测量"进行全产品测量和质量跟踪，利用质量特征分布间接确定设备状态转移概率。研究全局贝叶斯网络分解、分类回归的模型来表达多工况下不确定性变量间关系。最后，项目对典型半导体生产模型和实际OLED生产线分别进行仿真和实验验证。本项目所研究的动态维护调度模型对完善智能维护理论，促进其在半导体生产中的应用有重要意义。

中文关键词： 不均衡集成采样；马尔科夫决策；多设备机会维护；离散系统仿真；

英文摘要： Semi-conductor fabrication is a complex manufacturing process with multiple products and batches. The jobs have re-entry characteristics. The maintenance policy may affect production directly. This research will combine the status, quality and maintenance records togother and provide a integrated policy for production plan and intelligent maintenance. A hierarchy of two levels is presented. The first level is modelled with semi-markov decision process aimed at production and maintenance planning for single machine considering varied jobs and dynamic failures. The second level is discrete event simulation for multi-machines to optimize the maintenance time with balanced Work In Process, inventry and quality etc.. Virtual quality metrology is employed to obtain the unmeasured quality through status and controller data. Bayes network and Support Vector Machine are used to model the relationships with varieties between quality and status data.This project will give some simulation results with typical FAB models. At last, a real OLED production line in a coopration will be studied and used as an experimental object with the studyed model and strategy. This research is aimed at solving the problem of joint planning for production and mainteance in semi-conductor fabrications. It will provide a new strategy for semi-c

英文关键词： Non balanceed sampling；Markov Decesion；Multiple equipment maintenance；Discrete System Simulation；