基于改进混合采样均衡化的质量规则获取

Quality Rule Acquisition Based on Improved Mixed Sampling

零件加工质量规则获取对于改善零件质量有重要意义。由于零件加工数据存在不均衡的问题,直接挖掘难以得到不合格品相关规则,而现有的采样均衡化方法往往存在误删关键样本、破坏样本总体分布、产生噪声等缺陷,会导致挖掘结果不准确,为此提出了面向不均衡数据集的基于改进混合采样的质量规则获取模型。先利用提出的基于关键规则欠采样与改进SMOTE过采样的混合采样均衡化方法,对质量数据均衡化处理,再使用FP-tree关联规则算法挖掘质量规则。利用企业实际生产数据,与基于FP-tree的方法、基于随机混合采样的方法进行对比分析,得出该模型在不减少获得的合格类规则及保证规则正确性的前提下,可以获取到更多的不合格类规则。

The acquisition of quality rules for part processing is of great significance for improving the quality of parts.Due to the imbalance of the part processing data,it is difficult to directly mine the relevant rules for unqualified products,and the existing sampling equalization methods often have defects such as accidentally deleting key samples,destroying the overall distribution of the samples,and generating noise,which will lead to inadequate mining results.Accurate,for this reason,an imbalanced data set-oriented quality rule acquisition model based on improved mixed sampling is proposed.First,the proposed hybrid sampling equalization method based on key rule under sampling and improved SMOTE oversampling is used to equalize the quality data,and then the FP-tree association rule algorithm is used to mine quality rules.Using the actual production data of the enterprise,it is compared with the method based on FP-tree and the method based on random mixed sampling,and it is concluded that the model can obtain more without reducing the qualification rules obtained and ensuring the correctness of the rules.Of unqualified rules.