Automated robot inspection for spherical tank is beneficial to improving work efficiency and reducing costs. This paper presents an inspection robotic system that integrates software and hardware. The designed inspection robot can adsorb and climb on the vertical surface of tanks, carrying some equipment and tools. The mechanical structure of the robot includes four magnetic wheels, two drive components, a vision component and a multi-layer robot frame. Modular magnetic wheels can be installed and replaced quickly and easily. Additional grinding mechanism, cleaning mechanism and detection mechanism are also designed to achieve automated inspection and maintenance. The robotic system software is developed to achieve the remote control and monitoring of the robot. The deep networks of Mask R-CNN are applied to intelligently identify weld seams. Force balance and obstacle-negotiation performance of the robot are analyzed and simulated. Climbing experiments results indicate that the robot can carry 10.1 kg payload and climb stably on the vertical tank wall. In obstacle-surmounting experiments, the robot could smoothly negotiate the weld seam with a height of 4 mm. The robotic system can identify weld seams in real time after training and learning, the identification time of each image is about 0.2-0.25s. The field tests on a 3000 m3 spherical tank were carried out to test the practicality of the robotic system. The application of this inspection robotic system can effectively improve the automated inspection and maintenance of spherical tanks.