Tuning surface emissivity has been of great interest in thermal radiation applications, such as thermophotovoltaics and passive radiative cooling. As a low-cost and scalable technique for manufacturing surfaces with desired emissivities, femtosecond laser surface processing (FLSP) has recently drawn enormous attention. Despite the versatility offered by FLSP, there is a knowledge gap in accurately predicting the outcome emissivity prior to fabrication. In this work, we demonstrate the immense advantage of employing artificial intelligence (AI) techniques to predict the emissivity of complex surfaces. For this aim, we used FLSP to fabricate 116 different aluminum samples. A comprehensive dataset was established by collecting surface characteristics, laser operating parameters, and the measured emissivities for all samples. We demonstrate the successful application of AI in two distinct scenarios: (1) effective emissivity classification solely based on 3D surface morphology images, and (2) emissivity prediction based on surface characteristics and FLSP parameters. These findings open new pathways towards extended implementation of AI to predict various surface properties in functionalized samples or extract the required fabrication parameters via reverse engineering.

调整表面发射率在热辐射应用中引起了极大的兴趣，例如热光伏和被动辐射冷却。作为一种用于制造具有所需发射率的表面的低成本和可扩展技术，飞秒激光表面处理 (FLSP) 最近引起了极大的关注。尽管 FLSP 提供了多功能性，但在制造之前准确预测结果发射率方面存在知识差距。在这项工作中，我们展示了使用人工智能 (AI) 技术来预测复杂表面的发射率的巨大优势。为此，我们使用 FLSP 制造了 116 种不同的铝样品。通过收集所有样品的表面特征、激光操作参数和测量的发射率，建立了一个综合数据集。我们展示了人工智能在两个不同场景中的成功应用：（1）仅基于 3D 表面形态图像的有效发射率分类，以及（2）基于表面特征和 FLSP 参数的发射率预测。这些发现为扩展 AI 的实施开辟了新途径，以预测功能化样品中的各种表面特性或通过逆向工程提取所需的制造参数。