Abstract: To solve the problems of multiple correlations, high data dimensions, small samples existing in the key quality characteristics identification of multistage manufacturing process, the Ada-Lasso method is improved by Principle Component Regression, integrated with the state space idea and Bootstrap method to identify the key quality characteristics in multistage process. Firstly, the State Space idea is introduced to construct the identification model of key quality characteristics in multistage process; the samples are reconstructed by Bootstrap method to expand the sample size. Then, the improved Adaptive Lasso method is adopted to identify the key quality characteristics, and the effectiveness of Adaptive Lasso, Lasso, Ridge Regression and improved Adaptive Lasso is through simulation verification under different correlation degrees between quality characteristics. Finally, the application process of improved Adaptive Lasso method is illustrated by an example. The simulation and example results show that the improved Adaptive Lasso method has a good ability to identify the key quality characteristics of multistage processes, especially when there is a strong correlation between quality characteristics, it is significantly superior to the other two methods.

Key words: multistage manufacturing process, key quality characteristics, state space model, bootstrap, improved adaptive lasso

摘要： 为解决多工序制造过程关键质量特性识别中存在的质量特性间具有多重相关性以及数据高维度, 小样本等问题,本文采用主成分回归改进Adaptive Lasso方法并融合状态空间思想和Bootstrap方法实现多工序过程关键质量特性识别。首先引入状态空间思想构建多工序过程关键质量特性识别模型,然后利用Bootstrap方法重构样本,扩大样本量;进而采用改进Adaptive Lasso方法识别关键质量特性,并通过仿真验证改进Adaptive Lasso方法与Lasso,Adaptive Lasso和岭回归方法在质量特性间不同相关度下识别的有效性;最后通过实例说明改进Adaptive Lasso的具体应用过程,仿真及实例结果显示,改进Adaptive Lasso方法对多工序过程有良好的关键质量特性识别能力,特别当质量特性间有较强相关性时显著优于其它两种方法。

关键词: 多工序制造过程, 关键质量特性, 状态空间模型, Bootstrap, 改进Adaptive Lasso