The normal operation of pump equipment is crucial for the safety and stable operation of nuclear power plants. During long-term operations, they face various issues such as vibrations, excessive temperature, bearing damage, etc., leading to potential fault risks during operation. To facilitate quicker fault diagnosis and performance prediction by onsite maintenance personnel, reduce downtime, and lower repair costs, this paper proposes a method for implementing a mobile intelligent diagnostic terminal based on edge computing. By integrating traditional signal processing-based fault diagnosis methods with deep learning approaches, and then incorporating them into the intelligent diagnostic terminal, it is used for real-time health monitoring and fault diagnosis of key rotating components such as motors and gearboxes in nuclear power plant pump equipment. To ensure diagnostic accuracy, this paper employs lightweight neural networks, which not only improve diagnostic speed and real-time performance but also effectively reduce computational complexity and storage space requirements of the diagnostic devices. Finally, detailed functional verification, performance testing, and fault testing were conducted on the prototype of the intelligent diagnostic terminal on a circulating water pump test bench. The experimental results demonstrate that the prototype exhibits good diagnostic effectiveness and stability. The edge computing intelligent diagnostic terminal offers advantages such as low cost, strong real-time performance, and high accuracy, which are significant for ensuring the safe operation of pump equipment in nuclear power plants.