A digital twin (DT) is a cyberspace replica of a system, such as manufacturing equipment. A DT consists of statistical models and computer simulations of physical phenomena occurring in the system. The modeling is adjusted to the system based on signals from sensors attached to the system and their temporal changes. In general, a DT is utilized to (i) predict phenomena occurring in the system, (ii) optimize control parameters, and (iii) estimate part replacement schedules. We propose to use a DT to elucidate the unique solidification phenomena occurring in a type of metal 3D printing (i.e., additive manufacturing: AM) process. Thus, we propose that applications of DT that obtain scientific data be referred to as "digital twin science (DTS)." This paper first reviews the fundamental of the AM process, particularly powder bed fusion (PBF) and relevant computer simulations, and then studies on computer simulations conducted to elucidate the relationship between the extreme conditions characteristic of the PBF process and solidification microstructures. The findings achieved by the DTS approach indicate that the combination of experimental and simulation data aid the future development of techniques to obtain required microstructures exhibiting desired properties.