固着磨料加工高精度碳化硅反射镜粗糙度的人工神经网络研究

Study on the Artificial Neural Network of the RB-SiC Mirror Surface Roughness Fabricated with the Fixed Abrasive Technology

固着磨料工艺主要针对某空间相机的高精度平面折反镜而开发，分别从微观结构的仿真计算和人工神经网络两个角度对此工艺加工碳化硅反射镜表面粗糙度进行分析。一方面引入了二维粗糙度的在微观结构仿真概念，在人工神网络方面使用双隐层神经网络对固着磨料工艺的加工结果进行了分析，使得网络的性能大幅提高，收敛结果达到了8．4075×10^-5，并对网络性能进行了验证，标准化后的预测集与实验验证集距离偏差为0．2113。完全满足固着磨料工艺对表面粗糙度的预测需求。

Fixed abrasive technique mainly aims at the high precision plane of a space camera to fold the mirror. The microcosmic structure simulation calculation and artificial neural networks analysis of the reaction-bonded sintering （RB） SiC mirror surface roughness fabricated with fixed abrasive technique is analyzed. In the microcosmic structure simulation calculation part, the concept of the two-dimensional （2D） surface roughness is introduced. In the artificial neural networks analysis part, the concept of double hidden layer neural network is introduced to analyze the experimental results. The network performance is improved remarkably through training. The last performance value is 8. 4075 ×10^-5. The network performance is validated after training, the error between simulation data and experimental data is 0. 2113, which meets the prediction requirement of the fixed abrasive technique surface roughness.