The water-lubricated thrust bearing works as an important component in the submersible motor. The water-lubricated thrust bearing is often damaged by a poor working environment, overload, and alternating heat and cold resulting in failure of the sealing structure. The disassembling and assembling the acceleration sensor in the submersible motor is difficult and the acquisition of vibration signals is influenced by water damping pollution. To solve these problems, a mathematical model of the submersible motor under fault conditions is proposed, and the current signal is obtained by simulation. Based on the AlexNet model, the motor current signal (MCS) in different states is trained. The simulation results verify that the model can quickly identify fault signals with high accuracy, has strong robustness, and meets the requirements of predictability and real-time performance of fault diagnosis.