基于Elman神经网络的振镜扫描系统误差校正技术研究

Study on Error Correction in Dual Galvanometer Scanning System Based on Elman Recurrent Neural Network

为了消除激光快速成型机中振镜扫描系统的非线性扫描误差,以提高系统的扫描精度,提出了运用神经网络校正激光振镜扫描系统误差的方法.通过对BP神经网络、径向基神经网络和El-man神经网络3种不同网络的对比分析,运用Elman神经网络训练有限个标定坐标点的误差补偿值,构造全视场误差补偿曲面,然后对视场坐标进行校正补偿.通过实例分析,测试件在x、y方向的尺寸均方根误差由原来的0.229 6 mm、0.210 7 mm分别减小到0.023 2 mm和0.026 5 mm.实验结果表明,采用Elman神经网络构造振镜扫描系统误差补偿曲面,对振镜扫描系统进行动态误差校正,可以显著提高快速成型机的扫描精度.

Non-linear error of the dual galvanometer scanning system in the rapid prototyping machines directly influences the precision of the laser scanning system. A method based on neural network （NN） was proposed for correcting error in the system. Back propagation NN, radial basis function NN, and Elman recurrent NN were adopted and comparatively analyzed, and the Elman recurrent NN was verified as the best. The error data in the system were trained and the compensation surface was hence formed to correct the image field. The mean square roots of the specimen in the x and y direction were raised from 0. 229 6 mm, 0.210 7 mm to 0. 023 2 mm, 0. 026 5 mm respectively. The experimental results show the apparently improved scanning precision.