Cross-condition microstructure data generation method for titanium alloys based on improved CGAN

doi:10.11772/j.issn.1001-9081.2025070865

Journal of Computer Applications ›› 2026, Vol. 46 ›› Issue (7): 2373-2382.DOI: 10.11772/j.issn.1001-9081.2025070865

• Frontier and comprehensive applications • Previous Articles

Cross-condition microstructure data generation method for titanium alloys based on improved CGAN

Jing LIU¹^,², Shaoze ZHAO², Xingang LIU³, Haozhe NIU³, Haipeng JI¹^,⁴()

^1.State Key Laboratory of High Performance Roll Materials and Composite Forming （Tianjin University），Tianjin 300072，China
^2.School of Artificial Intelligence，Hebei University of Technology，Tianjin 300401，China
^3.School of Mechanical Engineering，Yanshan University，Qinhuangdao Hebei 066004，China
^4.School of Materials Science and Engineering，Hebei University of Technology，Tianjin 300401，China

Received:2025-08-01 Revised:2025-09-11 Accepted:2025-09-11 Online:2025-11-05 Published:2026-07-10
Contact: Haipeng JI
About author:LIU Jing， born in 1979， Ph. D.， research fellow. Her research interests include industrial artificial intelligence.
ZHAO Shaoze， born in 2001， M. S. candidate. His research interests include deep learning， microstructure prediction of titanium alloys.
LIU Xingang， born in 1978， Ph. D.， professor. His research interests include large forging process and optimization， aviation materials and processes， dynamic in-situ analysis， full-field high-throughput characterization.
NIU Haozhe， born in 2000， M. S. candidate. His research interests include high-throughput analysis for titanium alloys.
Supported by:
Tianjin Manufacturing Industry High-Quality Development Special Fund(20241047);Hebei Province Higher Education Science and Technology Research Project(CXY2024023);Hebei Provincial Major Science and Technology Support Program(24291001Z)

跨工况下基于改进CGAN的钛合金显微组织数据生成方法

刘晶¹^,², 赵邵泽², 刘鑫刚³, 牛浩哲³, 季海鹏¹^,⁴()

^1.高性能轧辊材料与复合成形全国重点实验室（天津大学），天津 300072
^2.河北工业大学人工智能与数据科学学院，天津 300401
^3.燕山大学机械工程学院，河北秦皇岛 066004
^4.河北工业大学材料科学与工程学院，天津 300401

通讯作者: 季海鹏
作者简介:刘晶（1979—），女，内蒙古包头人，研究员，博士，CCF杰出会员，主要研究方向：工业人工智能
赵邵泽（2001—），男，河北保定人，硕士研究生，主要研究方向：深度学习、钛合金显微组织预测
刘鑫刚（1978—），男，河南濮阳人，教授，博士，主要研究方向：大型锻件工艺及优化、航空材料与工艺、动态原位分析、全域高通量表征
牛浩哲（2000—），男，河北唐山人，硕士研究生，主要研究方向：钛合金的高通量分析
基金资助:
天津市制造业高质量发展专项资金资助项目(20241047);河北省高等学校科学技术研究项目(CXY2024023);河北省重大科技支撑计划项目(24291001Z)

Abstract

Abstract:

Titanium alloys are widely used in aerospace and other fields， and optimizing their microstructure is the key to improving material properties. Therefore， exploring the correlation between thermal processing parameters and microstructure evolutionary patterns of titanium alloys has become a research hotspot. Traditional research methods heavily rely on experimental methods， obtaining data through physical sample preparation and characterization. However， the complex combinations， long experimental cycles， and high costs of thermal processing conditions make it difficult to obtain complete microstructural data across all operating conditions. To address the above issues， a cross-condition microstructure data generation method for titanium alloys based on improved Conditional Generative Adversarial Network （CGAN）， namely GL-CGAN-TL， was proposed. The proposed method consists of two modules： first， a microstructure data generation module based on global-local generator fusion， which efficiently generates microstructure data under new operating conditions through the global-local generator； second， a data quality transfer validation module based on adaptive fine-tuning， which verifies the quality of generated data by combining transfer learning technology. Experimental results on the microstructure sample data of TC18 titanium alloy show that the proposed method can maintain the generated data completing the real data effectively， while reducing the Mean Absolute Error （MAE） and Root Mean Square Error （RMSE） by more than 27.6% compared to the existing methods， which is significantly better than similar methods.

Key words: cross-condition, microstructure, data generation, transfer learning, deep learning

摘要：

钛合金广泛应用于航空航天等领域，它的显微组织的优化是提升材料性能的关键。因此，探索钛合金热加工工艺参数与显微组织演变规律的关联性成为研究热点。传统研究方法高度依赖于实验手段，通过物理试样制备与表征获取数据，然而，热加工工况组合复杂、实验周期长且成本高，导致全工况下的显微组织数据难以完备。针对上述问题，提出一种跨工况下基于改进条件生成对抗网络（CGAN）的钛合金显微组织数据生成方法（GL-CGAN-TL）。该方法包括2个模块：1）基于全局-局部生成器融合的显微组织数据生成模块，通过全局-局部生成器在新工况下高效生成显微组织数据；2）基于自适应微调的数据质量迁移验证模块，结合迁移学习技术对生成数据质量进行验证。在TC18钛合金显微组织试样数据上进行实验的结果表明，本文方法在保持生成数据有效补全真实数据的同时，使平均绝对误差（MAE）和均方根误差（RMSE）较现有方法降低了27.6%以上，显著优于同类方法。

关键词: 跨工况, 显微组织, 数据生成, 迁移学习, 深度学习

CLC Number:

TP301.6

Jing LIU, Shaoze ZHAO, Xingang LIU, Haozhe NIU, Haipeng JI. Cross-condition microstructure data generation method for titanium alloys based on improved CGAN[J]. Journal of Computer Applications, 2026, 46(7): 2373-2382.

刘晶, 赵邵泽, 刘鑫刚, 牛浩哲, 季海鹏. 跨工况下基于改进CGAN的钛合金显微组织数据生成方法[J]. 《计算机应用》唯一官方网站, 2026, 46(7): 2373-2382.

Figures/Tables 14

Fig. 1 Schematic distribution of operating condition data in microstructure research

Fig. 2 Process of obtaining microstructure data of titanium alloys

Fig. 3 Framework of cross-condition microstructure data generation method for titanium alloys based on improved CGAN

Fig. 4 Framework of microstructure data generation module based on fusion of global-local generators

Fig. 5 Framework of data quality transfer validation module based on adaptive fine-tuning

Fig. 6 Process of titanium alloy sample preparation and characterization

Tab. 1 Partial microstructure sample data of TC18 titanium alloy

工况	序号	热加工工艺参数			组织特征参数
工况	序号	变形温度/℃	固溶温度/℃	应变量	α相含量	α相球化百分数/%
工况1	1	740	720	1.339 351	18.896	69.301 821 559 42
	2	740	720	1.339 681	17.892	65.372 701 509 26
						
	5 720	740	720	1.199 721	14.394	67.022 838 499 18
工况2	1	740	760	1.339 351	17.822	74.035 868 392 66
	2	740	760	1.339 681	20.686	65.269 634 875 49
						
	5 720	740	760	0.314 628	32.702	33.887 524 656 82
						
工况10	1	860	840	1.366 787	23.230	30.240 915 746 04
	2	860	840	1.367 015	17.116	18.870 452 437 39
						
	5 720	860	840	0.311 993	14.190	39.629 453 681 71

Tab. 2 Partially generated microstructure data of TC18 titanium alloy

工况	序号	热加工工艺参数			组织特征参数
工况	序号	变形温度/℃	固溶温度/℃	应变量	α相含量	α相球化百分数/%
新工况1	1	740	740	1.235 390	11.396	86.071 229 862 371
	2	740	740	1.238 930	17.892	88.781 906 790 139
						
	202	740	740	0.554 115	28.121	60.520 853 963 581
新工况2	1	740	750	1.216 206	12.429	86.524 323 471 475
	2	740	750	1.213 575	12.266	84.136 386 453 952
						
	212	740	750	0.571 749	30.123	61.345 440 159 697
						
新工况18	1	800	820	1.305 826	11.317	83.149 986 330 210
	2	800	820	1.263 338	12.602	84.862 242 157 222
						
	129	800	820	0.662 023	31.119	54.204 760 636 530

Fig. 7 Comparison of prediction accuracy for generated samples

Tab. 3 Validation results of confidence scores for generated samples

方法	工况1				工况2
	α相含量		α相球化百分数		α相含量		α相球化百分数
	RMSE	$M A E$	RMSE	$M A E$	RMSE	$M A E$	RMSE	$M A E$
Pretrained	18.368	17.869	0.246	0.223	5.726	5.394	0.214	0.189
Real-only	16.085	14.772	0.165	0.135	5.005	4.631	0.183	0.153
G25+R75	5.670	4.647	0.101	0.078	3.268	2.619	0.117	0.092
G50+R50	3.147	2.353	0.092	0.071	1.526	1.265	0.105	0.081
G75+R25	6.340	5.781	0.128	0.106	3.425	3.109	0.146	0.119
G100	10.809	10.399	0.130	0.107	4.034	3.576	0.155	0.128

Tab. 3 Validation results of confidence scores for generated samples

方法	工况1				工况2
	α相含量		α相球化百分数		α相含量		α相球化百分数
	RMSE	$M A E$	RMSE	$M A E$	RMSE	$M A E$	RMSE	$M A E$
Pretrained	18.368	17.869	0.246	0.223	5.726	5.394	0.214	0.189
Real-only	16.085	14.772	0.165	0.135	5.005	4.631	0.183	0.153
G25+R75	5.670	4.647	0.101	0.078	3.268	2.619	0.117	0.092
G50+R50	3.147	2.353	0.092	0.071	1.526	1.265	0.105	0.081
G75+R25	6.340	5.781	0.128	0.106	3.425	3.109	0.146	0.119
G100	10.809	10.399	0.130	0.107	4.034	3.576	0.155	0.128

Fig. 8 Comparison of alpha phase content and alpha phase spheroidization percentage prediction results of various experimental methods under two operating conditions

Tab. 4 Comparison of prediction results among C-DCGAN， RCWGAN， and GL-CGAN-TL

方法	工况1				工况2
	α相含量		α相球化百分数		α相含量		α相球化百分数
	RMSE	$M A E$	RMSE	$M A E$	RMSE	$M A E$	RMSE	$M A E$
C-DCGAN	10.728	10.232	0.167	0.142	4.227	3.799	0.177	0.145
RCWGAN	6.174	5.529	0.127	0.106	3.442	2.980	0.158	0.132
GL-CGAN-TL	3.147	2.353	0.092	0.071	1.526	1.265	0.105	0.081

Tab. 4 Comparison of prediction results among C-DCGAN， RCWGAN， and GL-CGAN-TL

方法	工况1				工况2
	α相含量		α相球化百分数		α相含量		α相球化百分数
	RMSE	$M A E$	RMSE	$M A E$	RMSE	$M A E$	RMSE	$M A E$
C-DCGAN	10.728	10.232	0.167	0.142	4.227	3.799	0.177	0.145
RCWGAN	6.174	5.529	0.127	0.106	3.442	2.980	0.158	0.132
GL-CGAN-TL	3.147	2.353	0.092	0.071	1.526	1.265	0.105	0.081

Fig. 9 Comparison of alpha phase content and alpha phase spheroidization percentage ablation results of various experimental methods under 2 operating conditions

Tab. 5 Comparison of prediction results among No-CUTREG， CGAN-TL and GL-CGAN-TL

方法	工况1				工况2
	α相含量		α相球化百分数		α相含量		α相球化百分数
	RMSE	$M A E$	RMSE	$M A E$	RMSE	$M A E$	RMSE	$M A E$
No-CUTREG	6.620	5.784	0.189	0.164	1.984	1.540	0.165	0.146
CGAN-TL	6.679	5.058	0.173	0.143	3.449	2.863	0.190	0.158
GL-CGAN-TL	3.147	2.353	0.092	0.071	1.526	1.265	0.105	0.081

Tab. 5 Comparison of prediction results among No-CUTREG， CGAN-TL and GL-CGAN-TL

方法	工况1				工况2
	α相含量		α相球化百分数		α相含量		α相球化百分数
	RMSE	$M A E$	RMSE	$M A E$	RMSE	$M A E$	RMSE	$M A E$
No-CUTREG	6.620	5.784	0.189	0.164	1.984	1.540	0.165	0.146
CGAN-TL	6.679	5.058	0.173	0.143	3.449	2.863	0.190	0.158
GL-CGAN-TL	3.147	2.353	0.092	0.071	1.526	1.265	0.105	0.081

References 20

[1]	刘晶，董志红，张喆语，等.基于联邦增量学习的工业物联网数据共享方法［J］.计算机应用， 2022， 42（4）： 1235-1243.
	Liu Jing， Dong Zhihong， Zhang Zheyu， et al. Data sharing method of industrial internet of things based on federal incremental learning ［J］. Journal of Computer Applications， 2022， 42（4）： 1235-1243.
[2]	Shi P， Qi Q， Qin Y， et al. Highly interacting machining feature recognition via small sample learning ［J］. Robotics and Computer-Integrated Manufacturing， 2022， 73： No.102260.
[3]	Cook A， Karakuş O. LLM-Commentator： novel fine-tuning strategies of large language models for automatic commentary generation using football event data ［J］. Knowledge-Based Systems， 2024， 300： No.112219.
[4]	Wu F， Courty N， Jin S， et al. Improving molecular representation learning with metric learning-enhanced optimal transport ［J］. Patterns， 2023， 4（4）： No.100714.
[5]	Ran H， Li W， Li L， et al. Learning optimal inter-class margin adaptively for few-shot class-incremental learning via neural collapse-based meta-learning ［J］. Information Processing and Management， 2024， 61（3）： No.103664.
[6]	Tian C， Gao H， Wang P， et al. An enhanced GAN for image generation ［J］. Computers， Materials and Continua， 2024， 80（1）： 105-118.
[7]	Liu Y， Lee J， Park M， et al. Learning to propagate labels： transductive propagation network for few-shot learning ［PP/OL］. V5. arXiv （2019-02-08）［2025-04-10］. .
[8]	Gayathri R， Sangeetha S K B， Mathivanan S K， et al. Enhancing heart disease prediction with reinforcement learning and data augmentation ［J］. Systems and Soft Computing， 2024， 6： No.200129.
[9]	Zhang R， Che T， Ghahramani Z， et al. MetaGAN： an adversarial approach to few-shot learning ［C］// NeurIPS 2018. Red Hook： Curran Associates Inc.， 2018： 2371-2380.
[10]	Kabir H M D， Mondal S K， Alam S B， et al. Transfer learning with spinally shared layers ［J］. Applied Soft Computing， 2024， 163： No.111908.
[11]	Molina-Cabanillas M A， Jiménez-Navarro M J， Arjona R， et al. DIAFAN-TL： an instance weighting-based transfer learning algorithm with application to phenology forecasting ［J］. Knowledge-Based Systems， 2022， 254： No.109644.
[12]	Yoo J， Zwartjes P. Attenuation of seismic migration smile artifacts with deep learning ［J］. Artificial Intelligence in Geosciences， 2022， 3： 123-131.
[13]	Himeur Y， Al-Maadeed S， Kheddar H， et al. Video surveillance using deep transfer learning and deep domain adaptation： towards better generalization ［J］. Engineering Applications of Artificial Intelligence， 2023， 119： No.105698.
[14]	Castillo-García G， Morán-Fernández L， Bolón-Canedo V. Feature selection for domain adaptation using complexity measures and swarm intelligence ［J］. Neurocomputing， 2023， 548： No.126422.
[15]	Zhang L， Gao X. Transfer adaptation learning： a decade survey ［J］. IEEE Transactions on Neural Networks and Learning Systems， 2024， 35（1）： 23-44.
[16]	Ma X， Xu P， Wang Z， et al. Domain adaptation with BERT-based domain classification and data selection ［C］// DeepLoRes NLP 2019. Stroudsburg： ACL， 2019： 76-83.
[17]	Liang J， Hu D， Feng J. Domain adaptation with auxiliary target domain-oriented classifier ［C］// CVPR 2021. Piscataway： IEEE， 2021： 16627-16637.
[18]	Qiao Z， Luo X， Xiao M， et al. Semi-supervised domain adaptation in graph transfer learning ［C］// IJCAI 2023. California： ijcai.org， 2023： 2279-2287.
[19]	Tan Y， Zhang Q， Shi L， et al. A novel short-term wind power scenario generation method combining multiple algorithms for data-missing wind farm considering spatial-temporal correlativity ［J］. International Journal of Electrical Power and Energy Systems， 2024， 162： No.110227.
[20]	Yilmaz B， Korn R. Synthetic demand data generation for individual electricity consumers： Generative Adversarial Networks （GANs）［J］. Energy and AI， 2022， 9： No.100161.

Cross-condition microstructure data generation method for titanium alloys based on improved CGAN

跨工况下基于改进CGAN的钛合金显微组织数据生成方法

RichHTML

PDF

Knowledge

Abstract

Cite this article

share this article

Figures/Tables 14

References 20

Related Articles 15

Recommended Articles

Metrics

[1]	Xiangyi WU, Hailiang YE, Feilong CAO. Point cloud completion method based on smooth-sharpen graph convolution [J]. Journal of Computer Applications, 2026, 46(7): 2267-2276.
[2]	Xiuli DU, Xing GAO, Xiaoyu ZHANG, Chengsheng PAN, Qijie ZOU. Video snapshot compressive imaging reconstruction method based on dense spatio-temporal deformable attention [J]. Journal of Computer Applications, 2026, 46(7): 2288-2296.
[3]	Miaogen LING, Rui JING, Wei FANG. Survey of research on applications of explainable deep learning in tropical cyclone forecasting [J]. Journal of Computer Applications, 2026, 46(7): 2318-2326.
[4]	Yi DU, Mingjin XU, Jiayi KONG, Liyao WANG, Chen ZHAO. Low-rank adaptive parameter-efficient fine-tuning algorithm based on YOLOv11 [J]. Journal of Computer Applications, 2026, 46(6): 1738-1745.
[5]	Zhenkai XIONG, Mengjun XU, Yinyin SUN, Xin WANG. Maritime ship detection algorithm under complex weather environments based on enhanced YOLOv8 [J]. Journal of Computer Applications, 2026, 46(6): 1998-2006.
[6]	Lili HE, Meng CAO, Lei ZHANG, Hongjun PAN, Yi LIU, Chengxin SUN. Sign language generation model based on Kolmogorov-Arnold network and diffusion Transformer [J]. Journal of Computer Applications, 2026, 46(6): 1801-1810.
[7]	Xing SHENG, Sunxian WENG, Kuosong CHEN, Zhongping WANG, Ruifeng REN, Yong LIU. Deep learning-based patent value evaluation for power grid enterprises [J]. Journal of Computer Applications, 2026, 46(5): 1468-1474.
[8]	Xinyao LIU, Jun LIANG, Jiahao LONG, Renliang YAN. Fine-grained Chinese herbal medicine image classification based on feature fusion and channel information compensation [J]. Journal of Computer Applications, 2026, 46(5): 1677-1683.
[9]	Huijie GUO, Tianfeng DOU, Zhenlin ZHANG, Kaiyuan QI, Dong WU, Zhijian QU, Zhao LI, Chongguang REN. Time-interdependency-aware dynamic Bayesian network for traffic prediction [J]. Journal of Computer Applications, 2026, 46(5): 1507-1517.
[10]	Haoxuan CHEN, Peichang YE, Lei LIU, Chengming LIU, Wenhua HU. Survey of automated code edit suggestion [J]. Journal of Computer Applications, 2026, 46(4): 1227-1237.
[11]	Songsen YU, Huang HE, Guopeng XUE, Hengtuo CUI. Quantitation and grading method for ceramic tile chromatic aberration based on improved fractal encoding network [J]. Journal of Computer Applications, 2026, 46(3): 959-968.
[12]	Jian ZHANG, Jianbo YU, Jian TANG. Municipal solid waste incineration state recognition method based on multilayer preprocessing [J]. Journal of Computer Applications, 2026, 46(3): 940-949.
[13]	Yongwei JIANG, Xiaoqing CHEN, Linjie FU. Elastic medical image registration model with high-frequency preservation based on spectrum decomposition [J]. Journal of Computer Applications, 2026, 46(3): 924-932.
[14]	Min CHEN, Xiaolin QIN, Shaohan LI, Hao YANG, Taohong LI. Review of deep learning applications in severe convective weather prediction [J]. Journal of Computer Applications, 2026, 46(3): 980-992.
[15]	Kejian CUI, Zhiming WANG, Zhaowen QIU. Method for retinal vessel segmentation and coronary artery disease prediction using optical coherence tomography angiography [J]. Journal of Computer Applications, 2026, 46(2): 640-651.