In human magnetoencephalography （MEG） research, the spatial location and orientation information of neural sources inside the human are difficult to obtain directly, making it challenging to intuitively and repeatedly evaluate the performance of spin-exchange relaxation-free (SERF) magnetometer array systems for magnetoencephalography (MEG) imaging. To address this issue, neuronal activity in the human brain can be modeled using equivalent current dipoles (ECDs) characterized by explicit position and orientation information. Based on this premise, we designed a brain-like physical dry phantom supporting multiple orientations with 25 ECDs at different positions, thereby providing controllable and known magnetic source information. Furthermore, a joint optimal orientation estimation method was proposed to simultaneously estimate the single dipole orientation under different signal-to-noise ratios. A potential source space of 3 mm resolution was created within the phantom, and dipole localization experiment was constructed using a 7-channel-SERF magnetometer array. Experimental results showed that, under known-source conditions, the SERF array achieves a mean localization error of 16.86 mm and an average orientation error of 15.35°. These findings indicate that constructing a physical phantom with known magnetic sources provides an effective approach for evaluating the feasibility of multi-channel SERF magnetometer arrays for MEG imaging, and offers a reliable basis for optimizing design under different channel configurations. In addition, the proposed physical phantom exhibits good repeatability and can be employed for routine calibration, performance consistency verification and operational maintenance of SERF-MEG systems. Overall, the physical phantom and orientation estimation method presented in the study provide reference value for performance assessment and engineering implementation of SERF magnetometer array in MEG imaging.