Design and implementation of drawing number apply system based on machine learning
-
摘要: 针对企业中图号申请效率低下的现状,开发了基于机器学习的图号申请系统。首先,对企业PDM库中的历史图号申请记录进行去流水号和去重处理得到数据集。其次,采用K-means++算法将数据集和需要申请图号的新零部件共同聚类划分为若干簇,遍历每簇中的新零部件并利用KNN算法得到其属性图号。针对“同名异号”件采用基于多视图卷积神经网络的三维模型检索技术得到其属性图号。最后,对属性图号分配最新流水号得到完整图号。以企业某批次冷藏车厢体为例,系统图号申请正确率达到95%以上,效率提高5~6倍。Abstract: For the current situation of low efficiency of drawing number apply in enterprises, a drawing number apply system based on machine learning was developed. First, the dataset is obtained by remove serial number and remove duplicates processing of the existing historical drawing number apply records in the enterprise PDM database. Secondly, the K-means++ algorithm is used to divide the dataset and the parts that need to apply for drawing numbers into several clusters, traverse the new parts in each cluster and use the KNN algorithm to get their attribute drawing numbers. For the “same name and different number” parts, the MVCNN-based 3D model retrieval technology is used to obtain the attribute drawing number. Finally, assign the latest serial number to the attribute drawing number to obtain the complete drawing number. Taking a batch of refrigerated trucks in an enterprise as an example, the correct rate of system drawing number apply is over 95%, and the efficiency is increased by 5~6 times.
-
Key words:
- drawing number apply /
- K-means++ /
- KNN /
- CNN
-
表 1 部分零部件名称One-hot编码示例
零部件名称 侧门 总成 底壁 顶壁 … 骨架 折弯件 门杠 属性图号 折弯件 0 0 0 0 … 0 1 0 KF8.623 底壁骨架 0 0 1 0 … 1 0 0 KF6.109 侧门门杠
总成1 1 0 0 … 0 0 1 KF6.604 表 2 K-means++算法聚类结果
簇号 最终聚类中心 样本数 1 <封胶板,KF8.610>
(0, 0, ···, 1, ···, 0, 0)45 2 <右壁木骨架,KF6.113>
(0, 0, ···, 1, ···, 1, ···, 0, 0)68 ··· ··· ··· 22 <底架总成,KF6.103>
(0, 0, ···, 1, ···, 1, ···, 0, 0)33 表 3 KNN算法中欧式距离d(xinc ,xi)的计算结果
序号 样本二元组 d(xinc, xi) 1 <右壁木骨架,KF6.113>
(0, 0, ···, 1, ···, 1···, 0, 0)$ \sqrt 2 $ 2 <顶壁木骨架,KF6.712>
(0, 0, ···, 1, ···, 1···, 0, 0)0 ··· ··· ··· 68 <反转隔板木骨架,KF6.103>
(0, ···, 1, ···, 1, ···, 1, ···, 0)$ \sqrt 3 $ -
[1] 孙宜然, 赵嵩正. 面向工装管理的工装图号编码方法的设计与实现[J]. 制造业自动化, 2007(8): 25-28. doi: 10.3969/j.issn.1009-0134.2007.08.007 [2] 黄坤, 孙向东, 谢志华, 等. B/S模式产品图号信息系统的开发与应用[J]. 火控雷达技术, 2017, 46(1): 79-82. doi: 10.3969/j.issn.1008-8652.2017.01.019 [3] 李毅民. 机械产品图号编码设计方法与原则[J]. 机械工业标准化与质量, 2002(6): 31-34. doi: 10.3969/j.issn.1007-6905.2002.06.014 [4] 韩廷超, 仇健, 葛任鹏, 等. K-means阈值算法下数控车床功率监控系统的研究[J]. 制造技术与机床, 2019(2): 69-72. doi: 10.19287/j.cnki.1005-2402.2019.02.014 [5] Lu Y H, Zhen M M, Fang T. Multiview based neural network for semantic segmentation on 3D scenes[J]. Science China (Information Sciences), 2019, 62(12): 248-250. [6] 梁杰, 陈嘉豪, 张雪芹, 等. 基于独热编码和卷积神经网络的异常检测[J]. 清华大学学报:自然科学版, 2019, 59(7): 523-529. [7] 朱连江, 马炳先, 赵学泉. 基于轮廓系数的聚类有效性分析[J]. 计算机应用, 2010, 30(S2): 139-141. [8] Lin C H, Kumar A. Contactless and partial 3D fingerprint recognition using multi-view deep representation[J]. Pattern Recognition, 2018, 83: 314-327. doi: 10.1016/j.patcog.2018.05.004 [9] Dhomne A, Kumar R, Bhan V. Gender recognition through face using deep learning[J]. Procedia Computer Science, 2018, 132: 2-10. doi: 10.1016/j.procs.2018.05.053 [10] Fu Y S, Song J, Xie F X, et al. Circular fruit and vegetable classification based on optimized GoogLeNet[J]. IEEE Access, 2021, 9: 113599-113611. doi: 10.1109/ACCESS.2021.3105112 [11] Phawinee S, Cai J F, Guo Z Y, et al. Face recognition in an intelligent door lock with ResNet model based on deep learning[J]. Journal of Intelligent and Fuzzy Systems, 2021, 40(4): 8021-8031. [12] Xin Z, Du M, Guo H X, et al. Light-weight FaceNet based on MobileNet[J]. International Journal of Intelligence Science, 2020, 11(1): 1-16. [13] 高淑萍, 赵清源, 齐小刚, 等. 改进MobileNet的图像分类方法研究[J]. 智能系统学报, 2021, 16(1): 11-20. doi: 10.11992/tis.202012034 [14] Zhou Z Y, Huang H, Fang B H. Application of weighted cross-entropy loss function in intrusion detection[J]. Journal of Computer and Communications, 2021, 9(11): 1-21. doi: 10.4236/jcc.2021.911001 [15] Gao T W, Chai Y T. Improving stock closing price prediction using recurrent neural network and technical indicators[J]. Neural Computation, 2018, 30(10): 1-22.