Data-driven ultrashort pulse laser processing using deep neural network for shape prediction and in-process monitoring

We report on recent research on the development of data-driven ultrashort pulse laser processing to achieve higher productivity and quality. We are developing in-process monitoring, artificial intelligence (AI) optimization, and fast active control of the laser based on them. These key technologies are introduced for micro-drilling of metals and transparent materials and laser-induced periodic surface structure (LIPSS) formation on a silica glass. We demonstrate a fast pulse-to-pulse modulation of the fluences to control the ablation efficiency. A deep neural network was utilized to predict the 3-dimensional shapes of the ablation craters depending on the laser parameters (fluence and pulse duration). The scheme was extended to 10 sequential modulations of fluences. An in-process monitoring of the crack formation on glasses was implemented by optical transmission imaging with deep neural network. The optical reflection/transmission technique was also employed to probe the quality of the LIPSS formation on a silica glass.