基于特征选择算法的复杂产品关键质量特性识别研究

发布时间：2018-03-05 01:19

本文选题：复杂产品　切入点：关键质量特性识别　出处：《天津大学》2016年博士论文　论文类型：学位论文

【摘要】：复杂产品包含大量质量特性,识别影响产品质量的关键质量特性(CTQ)是产品质量改进中必不可少的一步。实际生产中,大量质量特性数据从生产线收集,为后续分析提供了基础。本文针对复杂产品数据的高维性、数据非平衡性、质量特性时序性等特点,建立基于特征选择算法的CTQ识别算法。首先,针对数据高维性,建立了两个平衡数据CTQ识别算法。分别是基于ReliefF改进的识别算法和基于遗传模拟退火算法的识别算法。在基于ReliefF改进的方法中,提出一个新的CTQ数确定方法。结果表明,改进方法能够比传统ReliefF更加有效过滤无关、冗余质量特性。在基于遗传模拟退火算法的方法中,建立了结合质量特性数和质量特性重要性两个度量的综合适应度函数用于特征优化。实验结果表明,综合适应度函数能够增强算法降维能力。其次,针对非平衡制造数据,建立了一个两阶段CTQ识别框架。第一阶段,使用多目标优化算法识别出一组备选关键质量特性集。第二阶段,使用理想点法从备选解中选择最佳调和解,即CTQ集。基于该框架,分别使用多目标优化算法——改进NSGA-II和改进DMS,建立两个识别算法。在这两个算法中,分别定义了两种基于非平衡数据的质量特性重要性度量。实验结果表明,两个算法能够在数据非平衡条件下有效识别CTQ。最后,质量特性对应于产品加工各个阶段,具有时序性的特点。考虑各质量特性的时序性,并提出两阶段非平衡数据CTQ识别算法。在第一阶段,利用基于遗传算法和DMS的混合多目标优化算法选择备选关键质量特性集。第二阶段,从备选解中选择最佳调和解。在该阶段,考虑质量特性时序性,优先识别较早加工阶段的质量特性作为产品CTQ。实验结果表明了该方法的有效性。
[Abstract]:Complex products contain a large number of quality characteristics, the identification of key quality characteristics that affect product quality (CTQ) is an essential step in product quality improvement. In actual production, a large number of quality characteristics data are collected from the production line. In this paper, the CTQ recognition algorithm based on feature selection algorithm is established for the characteristics of complex product data, such as high dimension, unbalanced data, and quality characteristics. Firstly, aiming at the high dimension of data, the paper proposes a new algorithm based on feature selection algorithm. Two CTQ recognition algorithms for balanced data are established, one is based on the improved ReliefF algorithm and the other is based on the genetic simulated annealing algorithm. In the improved method based on ReliefF, a new method for determining the CTQ number is proposed. The results show that, The improved method can filter irrelevant and redundant quality characteristics more effectively than traditional ReliefF. In the method based on genetic simulated annealing algorithm, In this paper, a comprehensive fitness function combining two measures of mass characteristic number and quality characteristic importance is established for feature optimization. The experimental results show that the synthetic fitness function can enhance the dimensionality reduction ability of the algorithm. Secondly, for non-equilibrium manufacturing data, In this paper, a two-stage CTQ recognition framework is established. In the first stage, a set of alternative critical mass characteristic sets is identified by using multi-objective optimization algorithm. In the second stage, the optimal harmonic solution is selected from the alternative solution by using the ideal point method, that is, the CTQ set. Two recognition algorithms are established by using improved NSGA-II and improved NSGA-II respectively. In these two algorithms, two kinds of importance measures of quality characteristics based on unbalanced data are defined respectively. The experimental results show that, The two algorithms can effectively identify CTQs under the condition of non-equilibrium data. Finally, the quality characteristics correspond to each stage of product processing and have the characteristics of timing. In the first stage, the hybrid multi-objective optimization algorithm based on genetic algorithm and DMS is used to select the set of key quality characteristics. The optimal harmonic solution is chosen from the alternative solution. In this stage, the quality characteristics of the earlier processing stage are first identified as the CTQs. The experimental results show the effectiveness of the method.
【学位授予单位】：天津大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：F273.2

【相似文献】