摘要
质量检测作为质量管理的基础工作,是防止不良品流入市场的关键。针对工业大数据中不合格品(正类)密度极低,但现有算法多以全局最优为目标,难以检测出不合格品的问题,提出CS-AdaBoost-DT智能质检模型。使用AdaBoost自适应集成算法级联多棵决策树,以解决单个分类器易于陷入局部最优的问题;同时,引入代价敏感因子对分类结果、产品种类不同的样本权重进行差异化赋值,进而降低不平衡数据对模型分类结果偏移的影响,以期减少企业质检的损失。以Bosch公司家电生产线的大数据为例,使用决策树、SVM、AdaBoost-DT、CS-AdaBoost-DT模型进行实证分析,并进行十折交叉验证以分析模型性能。结果显示:CS-AdaBoost-DT漏检率均值为9.88%,AUC均值为95.21%,G-mean均值为83.9%,都优于其他三种模型,且各指标的标准差更小,表明CS-AdaBoost-DT模型不仅提高了产品质量检测的准确性,且具有更高的稳定性。
As the fundamental work in quality management,quality inspection was the key process to prevent the defective products from entering the market. Actually,there were only a small number of defective products in production and the existing algorithms were mostly aimed at the overall optimization,which means the defective products were difficult to be detected accurately.Considering this problem,the CS-AdaBoost-DT intelligent quality inspection model was proposed. In this model,many decision trees would be cascaded by the adaptive integration algorithm of AdaBoost,so as to solve the problem that a single classifier was easy to fall into local optimum. Besides,the cost sensitive factor was introduced to differentiate the sample weights of different product types. The CS-AdaBoostDT model aimed to decrease the impact from imbalanced data on the classification results,so it reduced the company loss in quality inspection. Based on the big data from Bosch household appliance production line,four models of Decision Tree,SVM,AdaBoost-DT,CS-AdaBoost-DT were used for products quality inspection,and ten-fold cross-validation was adopted to evaluate their performance.The results show that the CS-AdaBoost-DT model,with the average missed detection rate 9.88%,the average AUC 95.21% and the average G-mean 83.9%,is better than the other three models.Moreover,it shows even smaller standard deviation. This all suggest that the CS-AdaBoost-DT model can not only improve the accuracy of the product quality inspectionbut alsoachieve higher stability.
作者
吴增源
周彩虹
刘畅
郑素丽
WU Zengyuan;ZHOU Caihong;LIU Chang;ZHNEG Suli(College of Economics and Management,China Jiliang University,Hangzhou 310018,China;College of Electrical and Electronic Engineering,Shanghai Institute of Technology,Shanghai 201418,China)
出处
《工业工程与管理》
CSSCI
北大核心
2020年第5期42-49,共8页
Industrial Engineering and Management
基金
国家自然科学基金项目(71572187)
浙江省自然科学基金项目(LY20G010008)。
