月球风化层钻取采样过程密实度分类研究

钻取采样是月球风化层土壤样品获取的重要方式，密实度是重要的风化层月壤原位特性，对钻进过程中的策略制定有重要影响。本文结合钻取采样过程特点，提出了通过采样机构的力、速度、电流、温度等传感器获取的瞬时信息感知月壤密实度的方法，利用深度学习方法构建一类适应于可变长度序列数据的门控型循环神经网络，实现钻进过程月壤密实度在线分类。研究表明，该分类方法在风化层钻进过程中月壤密实度感知滞后时间约为33 s，对未知序列数据识别正确率大于89.36%，具有较高的分类精度和泛化能力。

Research on classification of relative density in lunar regolith drilling

Drilling is an important approach to acquire samples from lunar regolith. The relative density is an in situ characteristic of lunar regolith and will influence the strategy of drilling process. According to the working principle of drilling, the perception scheme of lunar soil relative density is proposed with the instantaneous information derived from sensors of force, velocity, current, and temperature. Based on the deep learning method, a class of gated recurrent unit neural network is constructed for variable length data sequence. The network can achieve the online classification of relative density of lunar soil. The delay time for relative density perception is about 33 s and the correct ratio is higher than 89.36% to unknown data sequence, indicating that the method has high classification precision and generality.