汽轮机叶片长链条产线的上下游工艺参数智能优化算法研究

Steam turbine blade is one of the most important parts of steam turbine. The geometric precision of blade directly affects the overall performance of the equipment. Therefore, the precision of multistage processes and long chain processing is increasingly demanded. At present, the process optimization of long chain production line mainly depends on experienced engineers, and the relationship between various process data is complex, which restricts the intelligent optimization of the production line. Therefore, this project focuses on the long chain processing production line of steam turbine blades and investigates the intelligent optimization algorithm of its process parameters. Firstly, the coupling mechanism between upstream and downstream process parameters is analyzed. Based on the data of industrial IoT, key process parameters are extracted by feature fusion. The optimal model of the production line is established by using statistical analysis and deep learning algorithm. Then, in the dynamic environment, the combination of deep reinforcement learning and transfer learning is wisely integrated to generate processing parameter optimization strategy under various operational conditions. The proposed processing parameters are fed to process researchers. Therefore, a man-machine interaction mechanism is constructed to achieve intelligent optimization decision-making. Finally, the validation is carried out on the turbine blade production line to reduce the defective rate of the production line and improve the processing accuracy. This project will improve the theory of intelligent process optimization for steam turbine blades, and also provide a theoretical reference for multi-process optimization of other complex surface parts.

汽轮机叶片是汽轮机最重要的零件之一，叶片几何精度直接影响设备整体性能，因此对其多工序、长链条加工的精细化要求越来越高。目前其长链条产线工艺优化仍主要依存于经验丰富的工程师，且各种工艺数据之间关系错综复杂，制约了产线智能升级。为此，本项目以汽轮机叶片的长链条加工产线为研究对象，对其工艺参数智能优化算法进行研究。首先，分析上下游工艺参数关联耦合机制，基于工业物联数据，通过融合式特征筛选提取关键工艺参数，运用统计学分析和深度学习方法，建立产线最优模型表达；然后，在动态加工环境下，有机结合深度强化学习和迁移学习，开发兼具策略反馈和适应性的叶片加工参数智能优化算法，为工艺研究人员反馈最佳工艺参数建议，构建人机合作的智能优化决策系统。最后，在汽轮机叶片产线进行验证，以期降低产线次品率并提升加工精度。本项目将完善汽轮机叶片智能工艺优化理论，也为其他复杂曲面类零件多工序加工工艺优化提供理论参考。

复杂曲面类零件长链条加工的工艺感知和精度控制能力直接决定了零件质量，特别是叶片类零件的加工，如航空发动机整体叶轮、汽轮机叶片、燃汽轮机叶片等，因其具有薄壁结构，上下游加工过程极易导致器件变形和振动，其几何精度直接影响设备整体性能、作业效率和使役寿命，因此需要运用大数据智能技术实现零件加工精度的预测、工艺参数的优化、加工质量的提升。本项目研究基于大数据学习的复杂曲面零件长链条加工工艺建模与参数智能优化问题，基于复杂曲面长链条加工产线物联数据实时采集装置，开展了叶片加工中上下游工艺关联与耦合规律研究，提出了基于大数据学习的复杂曲面零件加工工艺精度预测与溯源方法，建立了加工工艺与叶片质量之间最优建模表达，开发了航发叶片长链条加工的智能优化算法。数据采集装置应用于无锡透平叶片一条汽轮机叶片长链条加工产线，实现了叶片加工场景下精度预测可视化；所提算法应用于无锡透平叶片两条X5CrNiCuNb16-4（17-4 PH）复杂曲面叶片加工产线，溯源了关键加工工序中（综合铣、铣叶根等）的核心工艺参数（减薄区间、进气边曲率等），各工序参数优化后加工工序能力指数超过1.57，平均工艺调整时间减少10.83%。