重建自由曲面的神经网络算法

利用神经网络方法解决退向工程设计中的自由贡面重建问题，在ＢａｃｋＰｒｏｐａｇａｔｉｏｎ算法的基础上，吸取了模拟退火的算法优点，神经网络的学习采用按概率随机随机接受一个不成功训练值的方法，使学习跳出局部最优解，最终收敛于全局最优解，试验证明，用这种方法解决自由曲面的重建问题，可以达到良好的效果，优于以往文献提出的方法。

NEURAL NETWORK METHOD TO RECONSTRUCT THE FREEFORM SURFACE

This article presents an artificial neural network approach to solve the problem of reconstruction and manufacture of freeform surfaces in reverse engineering. Taking advantage of the global minimum property of Simulated Annealing Procedure, a technique is proposed to accept a temporally failed training result in accordance to probability. Using this technique, the training can jump out of the local minimum and converge to the global minimum. The method is better than the algorithm given in the article , when it is used to solve the problem of reconstruction and manufacture of freeform surfaces.