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 evolution 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 combination of thermal processing conditions, long experimental cycles, and high costs make it difficult to obtain complete microstructural data under all operating conditions. A method for generating microstructure data of titanium alloys based on improved Condition Generative Adversarial Network (CGAN) under cross-condition was proposed to address the above issues. This method is divided into a microstructure data generation module based on global local generator fusion and a data quality transfer verification module based on adaptive fine-tuning. The microstructure data generation module proposes that the global local generator efficiently generates microstructure data under new working conditions, and the data quality transfer verification module proposes an adaptive fine-tuning strategy combined with transfer learning technology to verify the quality of generated data. Experiments were conducted on the microstructure sample data of TC18 titanium alloy, and the results showed that the cross-condition microstructure data generation method can effectively complete the real data while maintaining the generated data, reducing the average absolute error and root mean square error by more than 27.6% compared to existing methods, which is significantly better than similar methods.

摘要： 钛合金广泛应用于航空航天等领域，其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实验手段，通过物理试样制备与表征获取数据，但热加工工况组合复杂，实验周期长、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。该方法分为基于全局?局部生成器融合的显微组织数据生成模块和基于自适应微调的数据质量迁移验证模块，其中显微组织数据生成模块提出全局?局部生成器在新工况下高效生成显微组织数据，数据质量迁移验证模块提出自适应微调策略结合迁移学习技术验证生成数据质量。在TC18钛合金显微组织试样数据上进行实验，结果表明，跨工况显微组织数据生成方法在保持生成数据可以有效补全真实数据的同时，使平均绝对误差和均方根误差较现有方法降低27.6%以上，显著优于同类方法。