1.Southwest China Research Institute of Electronic Equipment,Chengdu 610036,China
2.School of Computer Science,Sichuan University,Chengdu 610065,China
3.Dongfang Electric (Chengdu) Innovation Research Co. ,Ltd. ,Chengdu 610213,China
Show less
文章历史+
Received
Accepted
Published
2025-12-09
2026-02-04
2026-09-28
Issue Date
2026-10-22
PDF (2126K)
摘要
面向智能制造背景下工艺设计自动化的迫切需求,本文指出当前研究方法存在明显局限性。现有基于视觉的工艺生成模型则过度聚焦于图像特征提取,普遍忽视图纸中的关键文本标注(如尺寸公差、材料规格),导致多模态信息协同不足。这些不足的深层次问题在于,现有方法缺乏对工程图纸多模态特性(视觉-文本关联性)的有效建模,且传统融合策略难以实现跨模态的渐进式对齐,从而制约了工艺描述生成的准确性与逻辑一致性。针对上述不足,本研究提出一种基于多模态信息融合的工艺生成式设计算法。该算法通过引入(Optical Character Recognition, OCR)文本特征与视觉特征的深度融合,有效解决了工艺图纸中技术参数提取不完整、语义表达不准确等关键问题。本算法在双语评估替补指标(Bilingual Evaluation Understudy,BLEU)、面向召回率的摘要评估指标(Recall-Oriented Understudy for Gisting Evaluation,ROUGE)等指标上均优于传统视觉模型,且未明显增加计算开销,其为工艺设计智能化转型提供了可靠的技术支撑,对推动制造业数字化转型具有重要实践意义。
Abstract
Facing the urgent demand for process design automation in the context of intelligent manufacturing, this study identifies significant limitations in current research approaches.Existing vision-based process generation models tend to overemphasize image feature extraction while generally neglecting critical textual annotations in drawings (such as dimensional tolerances and material specifications), leading to insufficient multimodal information coordination.The underlying issues stem from the inadequate modeling of the multimodal characteristics (visual-textual correlations) inherent in engineering drawings by existing methods, and the inability of traditional fusion strategies to achieve progressive cross-modal alignment, thereby constraining the accuracy and logical consistency of generated process descriptions.To address these shortcomings, this study proposes a generative process design algorithm based on multimodal information fusion.By deeply integrating OCR-extracted textual features with visual features, the algorithm effectively addresses key challenges such as incomplete extraction of technical parameters and inaccurate semantic representation from process drawings.Our algorithm demonstrates superior performance over traditional vision-only models on metrics including BLEU and ROUGE, without significantly increasing computational overhead.It provides reliable technical support for the intelligent transformation of process design and holds significant practical value for advancing the digital transformation of the manufacturing industry.
近年来,多模态学习在计算机视觉和自然语言处理领域的融合应用取得了显著进展。从计算机视觉的角度来看,图像中的内容极具价值[1]。以图像描述生成为代表的多模态任务,通过深度神经网络实现了视觉信息与语言描述的有效关联,成为连接图像理解与文本生成的重要技术桥梁。特别是在Encoder-Decoder架构的推动下,基于注意力机制的图像描述模型在通用场景中展现出令人瞩目的性能。然而,当这些通用模型迁移到工艺设计等专业领域时,其适用性面临严峻挑战。在工艺设计这一特殊应用场景中,工程图纸不仅仅是简单的二维形状集合,更是承载着丰富设计意图和技术规范的专业文档[2],包含着大量丰富的文字标注与精确参数[3]。然而,当前主流的计算机辅助工艺设计(Computer Aided Process Planning,CAPP)系统或基于图像的工艺生成方法在应对工程图纸的多模态特性时表现乏力,严重制约了智能化转型的进程。具体而言,现有CAPP系统多基于规则或模板,虽能处理结构化数据,但难以自动解析非结构化的工程图纸,尤其缺乏对图纸中视觉与文本信息的协同利用。工程图纸包含大量文本标注(如尺寸公差、材料规格和工艺要求等关键参数)。当前基于图像的工艺生成方法(如通用图像描述模型)主要聚焦于视觉特征的提取,往往忽略文本标注信息,导致生成的工艺描述丢失关键参数。例如,在复杂零件图纸中,尺寸公差(如“⌀25±0.1”)或材料要求(如“5A06-H112”)若未被有效提取,会引发工艺规划错误(如加工余量不足或检测遗漏)。研究表明,单纯依赖视觉特征仅能捕捉图纸的几何轮廓,无法理解技术规范,亟需开发多模态融合方法以弥补这一语义鸿沟。
当前,基于Encoder-Decoder架构的通用图像描述模型在工艺设计任务中主要存在三方面局限:首先,多模态信息融合机制不完善,现有模型大多忽视Optical Character Recognition文本的集成[4],无法有效协调图像特征与图纸标注信息;其次,长文本生成过程中的语义一致性难以保证[5],关键工艺参数在生成长序列时易出现遗漏或失真;此外,模型计算复杂度高[6],难以满足实际工业应用对效率的严格要求。
KushwahaR, BiswasA.Hybrid feature and sequence extractor based deep learning model for image caption generation[C]//2021 12th International Conference on Computing Communication and Networking Technologies (ICCCNT), 2021: 1-6.
[2]
GuiH, YangZ, HarishA R, et al.GenPattern: Dual-graph enhanced sewing pattern generation via multimodal large language model[J].J Manuf Syst, 2025, 83: 822-838.
[3]
SunC W, HouJ L, LiuX G, et al.Large language model prompt generation method for engineering drawing understanding[J].J Comput Appl, 2025, 45(3): 801-807.
XuK, BaJ, KirosR, et al.Show, attend and tell: Neural image caption generation with visual attention [C]//Proceedings of the 32nd International Conference on Machine Learning (ICML), 2015: 2048-2057.
[6]
SeeA, LiuP J, ManningC D.Get to the point: Summarization with pointer-generator networks [C]//Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 2017: 1073-1083.
[7]
TANM, LEQ V.EfficientNet: Rethinking model scaling for convolutional neural networks[C]//Proceedings of the 36th International Conference on Machine Learning (ICML).Long Beach, USA: PMLR, 2019: 6105-6114.
[8]
McLaughlinN, Martinez del RinconJ, MillerP.Recurrent convolutional network for video-based person re-identification[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016: 1325-1334.
[9]
ElmanJ L.Finding structure in time[J].Cogn Sci, 1990, 14(2): 179-211.
[10]
SimonyanK, ZissermanA.Very deep convolutional networks for large-scale image recognition[PP/OL].V1.(2014-09-04) [2025-05-06].
[11]
HeK, ZhangX, RenS, et al.Deep residual learning for image recognition[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016: 770-778.
LiuX G, GuoY, LiY, et al.A method for constructing Domain-Specific large language models oriented to manufacturing process design[J].J Sichuan Univ Nat Sci Ed, 2025, 62(3): 513-521.
VaswaniA, ShazeerN, ParmarN, et al.Attention is all you need[C]//Proceedings of the 31st Annual Conference on Neural Information Processing Systems (NeurIPS), 2017: 6000-6010.
[16]
LiJ, LiD, XiongC, et al.BLIP: Bootstrapping language-image pre-training for unified vision-language understanding and generation [C]//Proceedings of the 39th International Conference on Machine Learning (ICML), 2022: 12888-12900.
[17]
AlayracJ-B, BarrI, BarreiraR, et al.Flamingo: A visual language model for few-shot learning [C]//Advances in Neural Information Processing Systems 35, 2022: 23716-23736.
[18]
LuJ, ClarkC, LeeS, et al.Unified-IO 2: Scaling autoregressive multimodal models with vision, language, audio, and action[C]//2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024: 26429-26445.
[19]
LiJ, LiD, SavareseS, et al.BLIP-2: Bootstrapping language-image pre-training with frozen image encoders and large language models[C]//Proceedings of the 40th International Conference on Machine Learning, 2023: 19730-19742.
[20]
DaiW, FungP N, HoiS, et al.InstructBLIP: Towards general-purpose vision-language models with instruction tuning[C]//Advances in Neural Information Processing Systems 36.New Orleans, Louisiana, USA: Neural Information Processing Systems Foundation, Inc.(NeurIPS), 2023: 49250-49267.
[21]
BengioY, SimardP, FrasconiP.Learning long-term dependencies with gradient descent is difficult [J].IEEE Trans Neural Netw, 1994, 5(2): 157-166.
[22]
KeW, ZhengY, LiY, et al.Large language models in document intelligence:A comprehensive survey, recent advances, challenges, and future trends[J].ACM Trans Inf Syst, 2026, 44(1): 1-64.
[23]
KiranaK C, KumalasariI, OktagaluG Q, et al.Comparison of tesseract OCR, easy OCR, and transformer OCR on handwritten image [C]//2025 9th International Conference on Electrical, Electronics and Information Engineering (ICEEIE), 2025: 1-6.
[24]
ShiB, BaiX, YaoC.An end-to-end trainable neural network for image-based sequence recognition and its application to scene text recognition[J].IEEE Trans Pattern Anal Mach Intell, 2017, 39(11): 2298-2304.
[25]
GravesA, SchmidhuberJ.Framewise phoneme classification with bidirectional LSTM and other neural network architectures[J].Neural Netw, 2005, 18(5/6): 602-610.
[26]
BautistaD, AtienzaR.Scene text recognition with permuted autoregressive sequence models[M]//Computer Vision-ECCV 2022.Tel Aviv, Israel: Springer Nature Switzerland, 2022: 178-196.
[27]
AtienzaR.Data augmentation for scene text recognition[C]//2021 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW),2021: 1561-1570.
[28]
LuY, ChenY, ZhangS.Research on the method of recognizing book titles based on paddle OCR[C]//2024 4th International Signal Processing, Communications and Engineering Management Conference (ISPCEM), 2024: 1044-1048.
[29]
SarkarO, SinhaS, JenaA K, et al.Automatic number plate character recognition using paddle-OCR [C]//2024 International Conference on Innovations and Challenges in Emerging Technologies (ICICET).Nagpur, India: IEEE, 2024: 1-7.
[30]
BaltrušaitisT, AhujaC, MorencyL-P.Multimodal machine learning: A survey and taxonomy[J].IEEE Trans Pattern Anal Mach Intell, 2019, 41(2): 423-443.
KarpathyA, LiF F.Deep visual-semantic alignments for generating image descriptions[J].IEEE Trans Pattern Anal Mach Intell, 2017, 39(4): 664-676.
[33]
NagraniA, YangS, ArnabA, et al.Attention bottlenecks for multimodal fusion[C]//Proceedings of the 35th International Conference on Neural Information Processing Systems, 2021: 14200-14213.
[34]
AndersonP, HeX, BuehlerC, et al.Bottom-up and top-down attention for image captioning and visual question answering[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018: 6077-6086.
[35]
TayY, DehghaniM, BahriD, et al.Efficient transformers: A survey[J].ACM Comput Surv, 2023, 55(6): 1-28.
[36]
LiuC, LiuC, LuoZ, et al.ARIF: An adaptive attention-based cross-modal representation integration framework[C]//Artificial Neural Networks and Machine Learning-ICANN 2024, 2024: 3-18.
[37]
MehtaN K, Arvind, PrasadS S, et al.Optimizing multitask industrial processes with predictive action guidance[J].IEEE Trans Automat Sci Eng, 2025, 22: 22833-22844.
[38]
RenL, WangH, WangY, et al.Foundation models for the process industry: Challenges and opportunities [J].Engineering, 2025, 52: 53-59.
[39]
KingmaD P, BaJ.Adam: A method for stochastic optimization[C]//Proceedings of the 3rd International Conference on Learning Representations (ICLR), 2015.
[40]
HintonG E, SrivastavaN, KrizhevskyA, et al.Improving neural networks by preventing co-adaptation of feature detectors[J].Computer Science, 2012, 3(4): 212-223.
[41]
PapineniK, RoukosS, WardT, et al.BLEU: A method for automatic evaluation of machine translation[C]//Proceedings of the 40th Annual Meeting on Association for Computational Linguistics - ACL '02, 2001: 311.
[42]
LinC-Y.ROUGE: A package for automatic evaluation of summaries[C]//Text Summarization Branches Out, 2004: 74-81.