Replication Data for: Machine Learning-Driven Optimization of Thermoelectric Materials Laser-Printed on Flexible Substrate

Replication data for Machine Learning-Driven Optimization of Thermoelectric Materials Laser-Printed on Flexible Substrate . Abstract: The development of flexible, lightweight, and scalable thermoelectric (TE) generators is critical for powering next-generation wearable and Internet-of-Things (IoT) devices. However, conventional fabrication routes for the benchmark Bi2Te3-based material are limited to the production of rigid and small devices. Emerging printing processes like laser bed powder fusion (LBPF) enable flexibility and free shaping, but at the expense of decreased performance. In this study, we utilize machine learning (ML) to optimize the TE performance of LBPF-fabricated Bi0.5Sb1.5Te3-based materials printed on a flexible substrate. By comparing multiple ML frameworks, we developed a predictive tool able to prescribe the Sb concentration, laser processing parameters, and printing atmosphere, leading to the highest performance. A ~4% improvement in average power factor, up to 1282 µW m-1 K-2, is achieved compared to the initial best-performing materials identified by an exhaustive traditional search of the experimental space. A flexible printed TE energy harvesting module produced with the optimized materials and occupying 4 cm2 displays a power output of 44 µW at ΔT = 30 K. This approach advances data-driven TE materials manufacturing and enhances the understanding of LPBF–property relationships, paving the way to high-performance, flexible TEGs for next-generation IoT and wearable systems.