【作者】 刘利平；

【导师】 马义中；

【作者基本信息】 南京理工大学 ， 系统工程， 2011， 博士

【摘要】 进入21世纪以来,质量受到了前所未有的重视。统计过程调整作为一种在线质量控制技术,主要通过对过程的控制变量进行调整,来实现在线减小过程的波动,从而提高产品质量,是实施过程改进和保证产品质量所必须的技术之一。对于复杂产品而言,随着产品复杂程度的不断增加,其制造过程呈现少样机甚至无样机化生产、在线调整受影响的因素多、多质量特性等特点,这些都对统计过程调整技术的研究和应用提出了更高的要求。本论文在已有的统计过程调整技术研究的基础上,以复杂产品离散制造过程的在线质量控制为研究对象,通过对不同类型的过程进行分析和建模,以及对过程输出质量特性进行估计和控制,为复杂产品制造过程的在线质量控制提供技术和方法；结合过程的费用结构,设定过程的调整策略,在保证过程产品质量的同时降低过程的总体质量损失。具体的研究内容包括：首先,研究了参数未知自相关过程的在线调整问题。在复杂产品的实际生产中,由于少样机甚至无样机化生产使得过程参数往往很难精确知道。基于此,通过对目标过程建立状态空间方程模型,并利用贝叶斯方法估计过程的未知参数,考虑调整不需要成本和每次调整的成本恒定的情形,基于序贯蒙特卡洛法分别给出了两种情形下使过程总体损失最小的在线调整策略。其次,针对过程的调整存在随机误差,而且很多情形复杂产品生产的批次比较少、设置偏差对过程输出质量影响尤为突出的实际,对考虑调整随机误差的设置调整问题进行了分析。考虑到每次调整需要成本,通过对过程输出质量特性偏离目标值引起的损失和调整成本之间进行权衡,基于过程输出质量特性的预报分布,设计了过程相应的边界形式的调整策略。在该调整策略中,调整边界是随生产阶段而变化的,并且由于调整的随机误差会影响过程波动,每个阶段的调整边界还与过程波动的估计相关。仿真分析结果验证了该调整策略的可行性与有效性。最后,考虑到许多复杂产品最终产品的质量往往需要多个质量特性值来进行刻画,本论文对多变量的统计过程调整问题进行了研究。对过程参数已知和未知两种情形进行分析,由于调整需要成本,需要在过程输出质量特性偏离目标引起的损失和调整成本之间进行权衡,给出了两种情形下的调整策略。在过程参数已知的情形中,推导了该情形下边界调整策略后,还分析了过程参数、调整费用等对调整边界和调整效果的影响。在过程参数未知的情形中,通过分析发现过程输出质量特性向量的分布与过程参数已知情形下的分布完全不同,给出了根据下一阶段输出质量特性预报分布的边界形式的调整策略,并分析了过程生产阶段,先验信息等对调整效果的影响。通过与其它调整策略进行比较的仿真分析结果来看,两种情形下的调整策略都能更为有效地减少过程的质量损失。本论文通过对不同过程模型以及费用结构下的过程调整策略进行研究,分别实现了对复杂产品制造过程中参数未知、调整存在随机误差、多质量特性等情形下的在线质量控制。本论文的研究是对统计过程调整理论的丰富,并为复杂产品制造过程提高质量,降低成本提供了有效的指导。更多还原

【Abstract】 Quality has been given attention more than ever before since 21st century. Statistical process adjustment is one kind of online quality control technologies that can be used to reduce the process variation online to improve the quality of products, mainly by adjusting the level of control factors. It is therefore also one of the necessary technologies of process improvement and quality assurance. With the increasing degree of complexity of products, the manufacturing processes of complex products (system) have shown many features that raise higher requirements for the research and application of statistical process adjustment, such as no pilot or little pilot study, online adjustment easily influenced by so many factors, multiple quality characteristics, and so on.Based on existing research of statistical process adjustment, more embedded study of the online quality control of the discrete manufacturing process of complex products (system) is conducted and presented. Technologies and methods are provided through the modelling and analysis of different kind of processes, and the estimation, control of the characteristics of process output quality. Cost structure is taken into account in setting up the adjustment strategies in order to reduce the overall cost and ensure the quality of the manufacturing process of complex products (system) at the same time.This research can be summarized as follows:Firstly, the adjustment problem for autocorrelation process with unknown parameters is discussed. In practice, since there is very little data from pilot study, the process parameters are too difficult to obtain. By building the state space model for the process, and estimating the unknown parameters by using Bayesian method, given the situations of the adjustment that has a cost or not, the adjustment strategy to minimize the total process lost based on sequencial Monte Carlo method is provided.Secondly, the setup adjustment problem with adjustment error is discussed given that in the manufacturing of complex products (system), short-run processes are becoming more prevalent due to high-degree customization, the adjustment of the process with initial offset is very important, and the adjustment action may also have random error. Since the adjustment has a cost, we trade off the loss caused by the deviation from target of the quality characteristic and the adjustment cost. Based on the deduced predictive distribution of the quality characteristic, the deadband form adjustment strategy is proposed. In this adjustment strategy, the adjustment limits are time-varying, and depending on the stage of process and the process variance. Simulation results demonstrate the feasibility and effectiveness of the proposed adjustment strategy.Lastly, considering that the quality of end products needs to be described by several characteristics in many manufacturing processes of complex products (system), the multivariate statistical process adjustment problem is discussed. We analyze the problem under the known or unknown process parameters respectively. Given fixed adjustment cost and the loss caused by the deviation from target to the quality characteristics, the adjustment strategy is presented respectively. In the situation of known process parameters, the sensitivity of the adjustment strategy with respect to the process parameters and adjustment cost are investigated. In the situation of unknown process parameters, the sensitivity of the adjustment strategy with respect to the number of process stage and prior information are also investigated. With comparison studies, it could be seen that both presented adjustment strategies can achieve good adjusment performance as well as reduce lost effectively in the process.The adjustment strategies for different processes and with different cost structure proposed in the thesis are adaptable to multiple conditions, such as process with unknown parameters, adjustment with random error, and multiple quality characteristics. This research enriches the theory of the statistical process adjustment, and provides guidance to enhance the quality of products, minimize quality loss in the manufacturing of complex products (system).更多还原