Abstract

Surface measurement using scattered light has attracted much attention as it has many advantages, such as high speed, robustness and long working distances. However, the inverse problem to determine the surface geometry from the scattering signal is complex and it is difficult to find solution with complex surfaces. To address this issue, we propose a novel method powered by artificial intelligence (AI), combining a rigorous light scattering model, which can address the reverse problem effectively. We demonstrate the method with a prototype system. In the system, a machine learning platform based on a densely connected neural network was designed and trained by simulating thousands of surfaces and their associated far-field scattering signals with different combinations of parameters. Thus, the machine learning model can recognise any unseen scattering signal that falls within the range of the simulation parameters. Laser light was projected onto the surface and was scattered and recorded by a sensor. The scattering signal was fed into the machine learning model and the characteristics of the surface being measured can be output directly. Experimental results show that the proposed method can measure surfaces with high accuracy, robustness and efficiency. We expect the method will be widely adapted in various in-process industrial applications.