基于人工智能的深潜耐压球壳应力场映射

目的 针对深潜耐压球壳在真实下潜过程中全局应力场难以直接获取的问题，提出一种基于人工智能的深潜耐压球壳应力场映射算法。方法 构建深潜耐压球壳有限元模型，并开展仿真分析。提出深潜耐压球壳监测布点方案，进而利用长短时记忆神经网络(Long-short Term Memory Network，LSTM)，将测点应力信息作为输入，将全局应力场信息作为输出，构建深潜耐压球壳应力场映射模型。最后，对不同测点下的映射结果进行分析。结果 与模型试验结果相比，仿真误差小于2%。与DNN模型及BP模型相比，映射误差分别下降94.92%与97.76%。结论 所提映射算法可在部分测点失效的情况下仍可以保持较高精度。

Stress Field Mapping Algorithm of Deep-sea Pressurized Spherical Shell Based on Artificial Intelligence

Aiming at the problem of difficulty in directly obtaining the global stress field of the deep-sea pressurized spherical shell during the actual diving process, the work aims to propose a stress field mapping algorithm for deep-sea pressurized spherical shells based on artificial intelligence. Firstly, a finite element model of the deep-sea pressurized spherical shell was constructed and simulated. The simulation error was less than 2% compared with the model test results. Secondly, a monitoring point layout plan was proposed. Furthermore, the Long-short Term Memory Network (LSTM) was used to construct the stress field mapping model for deep-sea pressurized spherical shells with motoring point stress information as input and global stress field information as output. Compared with the DNN model and BP model, the mapping error decreases by 94.92% and 97.76%, respectively. Finally, the mapping results under different monitoring points are analyzed, and the results show that the mapping algorithm proposed can still maintain high accuracy in the case of partial monitoring point failure.