基于数据挖掘与深度语义模型的工艺序列推荐方法

doi:10.11996/JG.j.2095-302X.2025040864

图学学报 ›› 2025, Vol. 46 ›› Issue (4): 864-873.DOI: 10.11996/JG.j.2095-302X.2025040864

基于数据挖掘与深度语义模型的工艺序列推荐方法

郑佳辉(), 郭宇, 吴涛, 王胜博, 黄少华(), 郑凯文

南京航空航天大学机电学院，江苏南京 210016

收稿日期:2024-09-25 修回日期:2024-12-26 出版日期:2025-08-30 发布日期:2025-08-11
通讯作者:黄少华(1990-)，男，讲师，博士。主要研究方向为大数据，物联网，智能决策等。E-mail：shaohuah@nuaa.edu.cn
第一作者:郑佳辉(2000-)，男，硕士。主要研究方向为智能制造、数字化制造和人工智能。E-mail：zheng_jiahui@nuaa.edu.cn
基金资助:
中国航空科学基金资助项目(2023M043052001)

Data mining and deep structured semantic model-based process sequence recommendation method

ZHENG Jiahui(), GUO Yu, WU Tao, WANG Shengbo, HUANG Shaohua(), ZHENG Kaiwen

College of Mechanical & Electrical Engineering, Nanjing University of Aeronautics and Astronautics, Nanjing Jiangsu 210016, China

Received:2024-09-25 Revised:2024-12-26 Published:2025-08-30 Online:2025-08-11
First author：ZHENG Jiahui (2000-), master student. His main research interests cover intelligent manufacturing, digital manufacturing and artificial intelligence. E-mail：zheng_jiahui@nuaa.edu.cn
Supported by:
China Aeronautical Science Foundation(2023M043052001)

摘要/Abstract

摘要：

为了应对航空制造工艺设计中传统的“经验驱动”方法面临的“数据超载”问题，难以实现航空复杂零件的智能化工艺设计，提出一种基于数据挖掘与深度语义模型的工艺序列推荐方法。通过采用PrefixSpan算法与BERT大语言模型相结合从零件实例数据中挖掘典型制造工艺序列及其相关能力，构建了可重用、可更新的制造工艺知识库。在此基础上，针对航空制造数据的特点提出了一种改进的空间通道注意力机制，进行零件实例数据隐式特征提取，同时针对零件实例不均衡分布导致的“冷启动”问题，结合自监督学习挖掘数据的深层结构，保证模型泛化能力和小样本实例的学习能力。通过基于双通道注意力的深度语义模型与自监督学习相结合的方法，使得模型在数据不平衡的情况下更好地提取特征、学习知识以及准确地推荐更加符合航空工艺设计的工艺序列。以某航空零件为例，进行了制造工艺序列的推荐与验证。实验结果表明，该方法在制造工艺序列推荐的各项指标上均优于基准模型，验证了该方法的有效性，且能满足航空工艺设计人员的智能化工艺设计需求。

关键词: 数据挖掘, 自监督学习, 深度语义模型, 航空复杂零件, 制造序列推荐

Abstract:

To address the challenge of “data overload” encountered with traditional “experience-driven” methods in aerospace manufacturing process designs, we propose a process sequence recommendation method based on data mining and deep semantic models. This method integrates the PrefixSpan algorithm and BERT to extract typical manufacturing process sequences and their associated capabilities from component instance data, thereby constructing a reusable and updatable manufacturing process knowledge base. Building on this foundation, an enhanced spatial channel attention mechanism is introduced to accommodate the characteristics of aerospace manufacturing data, enabling implicit feature extraction form part instance data. Additionally, to mitigate the “cold start” issue caused by the uneven component instance distribution, self-supervised learning is employed to uncover the deep structure of the data, thereby ensuring the model’s generalization ability and improving its capability to learn from small sample instances. By combining a dual-channel attention-based deep semantic model with self-supervised learning, this approach effectively extract features, acquire knowledge, and accurately recommend process sequences suitable for aerospace manufacturing, even in the presence of data imbalance. Finally, a case study involving a specific aerospace component was conducted to validate the proposed method. Experimental results demonstrate that this method consistently outperforms benchmark models across various metrics in manufacturing process sequence recommendation, confirming its effectiveness and fulfilling the intelligent process design requirements of aerospace engineers.

Key words: data mining, self-supervised learning, deep semantic models, aviation complex parts, manufacturing sequence recommendation

中图分类号:

郑佳辉, 郭宇, 吴涛, 王胜博, 黄少华, 郑凯文. 基于数据挖掘与深度语义模型的工艺序列推荐方法[J]. 图学学报, 2025, 46(4): 864-873.

ZHENG Jiahui, GUO Yu, WU Tao, WANG Shengbo, HUANG Shaohua, ZHENG Kaiwen. Data mining and deep structured semantic model-based process sequence recommendation method[J]. Journal of Graphics, 2025, 46(4): 864-873.

图/表 12

图1 基于数据挖掘与深度语义模型的工艺序列推荐方法框架图

Fig. 1 Framework diagram of process sequence recommendation method based on data mining and deep semantic models

图2 制造工艺知识库(MPKD)构建流

Fig. 2 The flow of manufacturing process knowledge database (MPKD) construction

图3 双通道注意力深度结构化语义模型框架图

Fig. 3 The framework diagram of dual-channel attention deep structured semantic model

图4 双通道注意力模型结构图

Fig. 4 The structure of dual-channel attention mode

表1 抽取记录样例

Table 1 The extraction of record samples

元素	内容
几何属性特征	{“15.0 mm”,“1.5 mm”,“75.0 mm”,“6%”,“0°”}
材料属性特征	{“铝”,“LF2”,“75*1.5”,“M”,“管材”}
制造工艺特征	{“无氧化”,“无焊接”,“无酸洗”,“无扩口”,“除油”,“无喷漆”}
制造工艺序列	[“下料”，“领料”，“标记”，“除油”，“检查”，“标记”，“成品检验”]

表2 平衡数据的制造工艺序列推荐实验对比结果

Table 2 The experimental comparison results of process sequence recommendation for balanced data

模型	AUC	MRR@K			NDCG@K
模型	AUC	K=1	K=3	K=10	K=1	K=3	K=10
LSTM	0.938 4	0.571 8	0.721 9	0.743 5	0.571 8	0.770 6	0.808 2
Transformer	0.947 2	0.681 5	0.795 7	0.810 1	0.681 5	0.831 8	0.858 0
DSSM	0.974 7	0.746 7	0.864 8	0.866 5	0.746 7	0.897 9	0.901 1
DA-DSSM	0.981 7	0.834 5	0.908 0	0.910 5	0.834 5	0.928 9	0.933 5
DA-SLDSSM	0.988 8	0.887 2	0.940 4	0.940 4	0.887 2	0.954 1	0.955 8

图5 平衡数据的制造工艺序列推荐指标迭代结果((a) 损失迭代曲线；(b) AUC指标迭代曲线)

Fig. 5 The iterative results of process sequence recommendation metrics for balanced data ((a) Loss iteration curve; (b) AUC metric iteration curve)

表3 长尾数据的制造工艺序列推荐实验对比结果

Table 3 The experimental comparison results of process sequence recommendation for long-tail data

模型	AUC	MRR@K			NDCG@K
模型	AUC	K=1	K=3	K=10	K=1	K=3	K=10
LSTM	0.881 8	0.522 0	0.625 6	0.669 4	0.522 0	0.662 5	0.749 1
Transformer	0.935 3	0.634 1	0.754 8	0.773 7	0.634 1	0.793 7	0.829 4
DSSM	0.959 0	0.672 1	0.807 2	0.815 2	0.847 6	0.847 6	0.862 4
DA-DSSM	0.974 5	0.794 5	0.882 2	0.886 3	0.907 3	0.907 3	0.915 3
DA-SLDSSM	0.984 7	0.869 5	0.928 4	0.928 9	0.869 5	0.942 6	0.947 1

图6 长尾数据的制造工艺序列推荐实验对比结果((a) 损失迭代曲线；(b) AUC指标迭代曲线)

Fig. 6 The experimental comparison results of process sequence recommendation for long-tail data ((a) Loss iteration curve; (b) AUC metric iteration curve)

表4 零件实例推荐验证结果

Table 4 The validation results of part instance recommendation

零件ID	零件实例部分特征	推荐的典型工艺序列ID (K=3)
001	[15，1.5，75，6，0，铝，LF2，75 $∗$ 1.5，M，无氧化、无焊接，无扩口······]	[9，7，8]
042	[20，0，0，0，6，0，辅料，ET200，0，无，有氧化，有焊接，无扩口······]	[3，1，4]
091	[15，0，0，0，6，0，辅料，6602，0，无，无氧化，无焊接、有扩口······]	[4，3，7]
165	[20，1.5，75，6，90，铝，LF2，75 $∗$ 1.5，M，无氧化、无焊接，无扩口······]	[7，5，8]
320	[25，1，16，6，90，不锈钢，1Cr18Ni9Ti，16 $∗$ 1，固溶，无氧化、无焊接，无扩口······]	[11，12，8]

表4 零件实例推荐验证结果

Table 4 The validation results of part instance recommendation

零件ID	零件实例部分特征	推荐的典型工艺序列ID (K=3)
001	[15，1.5，75，6，0，铝，LF2，75 $∗$ 1.5，M，无氧化、无焊接，无扩口······]	[9，7，8]
042	[20，0，0，0，6，0，辅料，ET200，0，无，有氧化，有焊接，无扩口······]	[3，1，4]
091	[15，0，0，0，6，0，辅料，6602，0，无，无氧化，无焊接、有扩口······]	[4，3，7]
165	[20，1.5，75，6，90，铝，LF2，75 $∗$ 1.5，M，无氧化、无焊接，无扩口······]	[7，5，8]
320	[25，1，16，6，90，不锈钢，1Cr18Ni9Ti，16 $∗$ 1，固溶，无氧化、无焊接，无扩口······]	[11，12，8]

表5 扩展数据集样例

Table 5 The data of expanded dataset

元素	复杂辅材数据	简单钣金数据
几何属性特征	{纤维厚度：15 mm，树脂层厚度：1.5 mm，面板宽度：75 mm，纤维角度偏差：6%，纤维方向：0°，铺层间距：3 mm，边缘容差：0.1 mm，孔径：5 mm}	{板材厚度：2 mm，孔径：10 mm，折弯半径：5 mm，长度：100 mm，宽度：50 mm，边缘容差：0.2 mm}
材料属性特征	{纤维密度：1.6 g/cm³，孔隙率限制：1.5%，复合材料弹性模量：70 GPa，断裂伸长率：2%，剪切模量：5 GPa}	{材料类型：不锈钢，材料密度：7.85 g/cm³，屈服强度：250 MPa，抗拉强度：450 MPa，硬度：45 HRC}
制造工艺特征	{固化时间：120 min，层数：5层，模具温度：180 ℃，铺层顺序：0°/90°/45°，固化速率：1 ℃/min，固化结束温度：180 ℃，真空度：95%，压实时间：30 min，铺层类型：UD单向，铺层厚度偏差：±0.02 mm}	{切割方式：激光切割，折弯角度：90°，表面处理方式：喷砂，涂层厚度：10 μm，加工精度：±0.1 mm}
制造工艺序列	{材料准备，模具制作与处理，初始铺层，层间检查，真空袋密封，热压罐固化，压力控制，缓冷处理，脱模，后固化，切割修边，钻孔与组装，表面打磨，无损检测，尺寸检测，机械性能测试，表面涂层}	{材料准备，切割，折弯成型，边缘处理，焊接，表面处理，质量检测，包装}

图7 模型泛化性实验结果图

Fig. 7 The results of the model generalization experiments

参考文献 22

[1]	汤文俊, 王裴岩, 蔡东风, 等. 基于编码器—解码器的工艺过程生成方法[J]. 计算机集成制造系统, 2023, 29(11): 3656-3668.
	TANG W J, WANG P Y, CAI D F, et al. Encoder-decoder based process generation method[J]. Computer Integrated Manufacturing System, 2023, 29(11): 3656-3668 (in Chinese).
[2]	张一鸣, 刘金锋, 陈亚杰, 等. 零件加工隐性工艺知识获取方法研究[J]. 图学学报, 2024, 45(2): 399-408. DOI
	ZHANG Y M, LIU J F, CHEN Y J, et al. Tacit process knowledge acquisition methods for the parts machining[J]. Journal of Graphics, 2024, 45(2): 399-408 (in Chinese). DOI
[3]	何柳, 安然, 刘姝妍, 等. 基于知识图谱的航空多模态数据组织与知识发现技术研究[J]. 图学学报, 2024, 45(2): 300-307. DOI
	HE L, AN R, LIU S Y, et al. Research on knowledge graph-based aviation multi-modal data organization and discovery method[J]. Journal of Graphics, 2024, 45(2): 300-307 (in Chinese). DOI
[4]	冯锦丹, 何楠, 刘金山, 等. 基于知识驱动的航天复杂产品装配工艺重构方法[J]. 计算机集成制造系统, 2024, 30(12): 4314-4327.
	FEN J D, HE N, LIU J H, et al. Knowledge-driven assembly process reconfiguration method for aerospace complex products[J]. Computer Integrated Manufacturing Systems, 2024, 30(12): 4314-4327 (in Chinese).
[5]	曹勇. 基于数据挖掘的工艺知识发现与重用研究[D]. 济南: 山东大学, 2019.
	CAO Y. Research on process knowledge discovering and reusing based on data mining[D]. Jinan: Shandong University, 2019 (in Chinese).
[6]	SAXENA A, PRASAD M, GUPTA A, et al. A review of clustering techniques and developments[J]. Neurocomputing, 2017, 267: 664-681.
[7]	ZHAO C X, DINAR M, MELKOTE S N. A data-driven framework for learning the capability of manufacturing process sequences[J]. Journal of Manufacturing Systems, 2022, 64: 68-80.
[8]	CHEN X C, YAO L N, MCAULEY J, et al. Deep reinforcement learning in recommender systems: a survey and new perspectives[J]. Knowledge-Based Systems, 2023, 264: 110335.
[9]	陈治宇, 鲍劲松, 郑小虎, 等. 基于长短期记忆网络的装配工艺语义识别方法[J]. 计算机集成制造系统, 2021, 27(6): 1582-1593. DOI
	CHEN Z Y, BAO J S, ZHENG X H, et al. Semantic recognition method of assembly process based on LSTM[J]. Computer Integrated Manufacturing Systems, 2021, 27(6): 1582-1593 (in Chinese).
[10]	KOREN Y, RENDLE S, BELL R. Advances in collaborative filtering[M]//RICCI F, ROKACH L, SHAPIRA B. Recommender Systems Handbook. 3rd ed. New York: Springer, 2022: 91-142.
[11]	HUANG P S, HE X D, GAO J F, et al. Learning deep structured semantic models for web search using clickthrough data[C]// The 22nd ACM International Conference on Information & Knowledge Management. New York: ACM, 2013: 2333-2338.
[12]	KRICHENE W, MAYORAZ N, RENDLE S, et al. Efficient training on very large corpora via Gramian estimation[EB/OL]. [2024-03-24]. https://dblp.org/db/conf/iclr/iclr2019.html#KricheneMRZYHCA19.
[13]	GUO C, MOUSAVI A, WU X, et al. Breaking the glass ceiling for embedding-based classifiers for large output spaces[C]// The 33rd International Conference on Neural Information Processing Systems. New York: ACM, 2019: 445.
[14]	PANDA D K, RAY S. Approaches and algorithms to mitigate cold start problems in recommender systems: a systematic literature review[J]. Journal of Intelligent Information Systems, 2022, 59(2): 341-366.
[15]	YAO T S, YI X Y, CHENG D Z, et al. Self-supervised learning for large-scale item recommendations[C]// The 30th ACM International Conference on Information & Knowledge Management. New York: ACM, 2021: 4321-4330.
[16]	DEVLIN J, CHANG M W, LEE K, et al. BERT: Pre-training of deep bidirectional transformers for language understanding[C]// 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers). Minnesota: Association for Computational Linguistics, 2019: 4171-4186.
[17]	林佳瑞, 程志刚, 韩宇, 等. 基于BERT预训练模型的灾害推文分类方法[J]. 图学学报, 2022, 43(3): 530-536.
	LIN J R, CHENG Z G, HAN Y, et al. Disaster tweets classification method based on pretrained BERT model[J]. Journal of Graphics, 2022, 43(3): 530-536 (in Chinese). DOI
[18]	张晋, 刘检华, 赵文浩, 等. 基于预训练模型与相似度算法的复杂产品装配技术问题风险预警方法[J]. 机械工程学报, 2024, 60(24): 317-329.
	ZHANG J, LIU J H, ZHAO W H, et al. Risk warning method for complex product assembly technology problems based on pre-training model and similarity algorithm[J]. Journal of Mechanical Engineering, 2024, 60(24): 317-329 (in Chinese).
[19]	SUBAKTI A, MURFI H, HARIADI N. The performance of BERT as data representation of text clustering[J]. Journal of Big Data, 2022, 9(1): 15. DOI PMID
[20]	PEI J, HAN J W, MORTAZAVI-ASL B, et al. PrefixSpan: mining sequential patterns efficiently by prefix-projected pattern growth[C]// The 17th International Conference on Data Engineering. New York: IEEE Press, 2001: 215-224.
[21]	WOO S, PARK J, LEE J Y, et al. CBAM: convolutional block attention module[C]// The 15th European Conference on Computer Vision. Cham: Springer, 2018: 3-19.
[22]	AKIBA T, SANO S, YANASE T, et al. Optuna: a next-generation hyperparameter optimization framework[C]// The 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. New York: ACM, 2019: 2623-2631.

基于数据挖掘与深度语义模型的工艺序列推荐方法

Data mining and deep structured semantic model-based process sequence recommendation method

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 12

参考文献 22

相关文章 7

编辑推荐

Metrics

本文评价

[1]	刘鸿硕, 白静, 晏浩, 林淦. 面向三维点云的平衡泛化和特化的细粒度分类网络[J]. 图学学报, 2025, 46(3): 602-613.
[2]	王雪婷, 郭新, 汪松, 陈恩庆. 基于变分自编码器掩蔽重建的骨骼点动作识别方法[J]. 图学学报, 2025, 46(2): 270-278.
[3]	林晓, 张秋阳, 郑晓妹, 杨启哲. 基于自监督的主动标签清洗[J]. 图学学报, 2024, 45(3): 495-504.
[4]	安峰 , 戴军 , 韩振 , 严仲兴 . 引入注意力机制的自监督光流计算[J]. 图学学报, 2022, 43(5): 841-848.
[5]	孔震 1,张华鲁 1,2，岳圣凯 3,袁明磊 3,路通 3. 基于时域卷积网络的多尺度双线性天气预测模型[J]. 图学学报, 2020, 41(5): 764-770.
[6]	李农勤 1，杨维信 2,3 . 基于生成式对抗神经网络的手写文字图像补全[J]. 图学学报, 2019, 40(5): 878-884.
[7]	王汉友. 基于用户导向粗糙集算法的旅行箱设计规则研究[J]. 图学学报, 2017, 38(5): 675-682.