Machine learning for 3D simulated visualization of laser machining
Machine learning for 3D simulated visualization of laser machining
复制标题
DOI：
10.1364/oe.26.021574
复制
发表时间：
2018-08-20
期刊：
影响因子：
3.8
通讯作者：
Mills, Ben
中科院分区：
文献类型：
--
作者：
Heath, Daniel J.;Grant-Jacob, James A.;Mills, Ben
激光加工可以依赖于许多复杂和非线性物理过程的组合。因此，基于第一性原理（如光子-原子相互作用）的激光加工模拟很难扩大到实验上有用的尺寸。在这里，我们展示了一种使用神经网络的模拟方法，该方法不需要对底层物理过程有任何了解，而是直接使用实验数据来创建实验模型。研究表明，神经网络建模方法可以准确预测暴露于各种空间强度分布后激光加工表面的3D表面轮廓，并用于发现实验数据中固有的趋势，否则这些趋势很难发现。由The Optical Society根据Creative Commons Attribution 4.0 License条款发布。
Laser machining can depend on the combination of many complex and nonlinear physical processes. Simulations of laser machining that are built from first-principles, such as the photon-atom interaction, are therefore challenging to scale-up to experimentally useful dimensions. Here, we demonstrate a simulation approach using a neural network, which requires zero knowledge of the underlying physical processes and instead uses experimental data directly to create the model of the experiment. The neural network modelling approach was shown to accurately predict the 3D surface profile of the laser machined surface after exposure to various spatial intensity profiles, and was used to discover trends inherent within the experimental data that would have otherwise been difficult to discover. Published by The Optical Society under the terms of the Creative Commons Attribution 4.0 License.