Aiming at the problems of difficulties in fault detection and the lack of maintenance experience for unmanned mine-sweeping vehicles, a novel method is proposed, featuring rapid detection and high diagnostic accuracy. Building upon the extreme learning machine (ELM) algorithm, the bald eagle search (BES) algorithm is optimized by incorporating the Lévy flight strategy and simulated annealing mechanism. The improved bald eagle search algorithm (IBES) is then utilized to optimize the parameters of the extreme learning network. A fault diagnosis model for the power system of unmanned mine-sweeping vehicles is established, based on the extreme learning machine optimized by the improved bald eagle search algorithm. Experimental results indicate that the fault diagnosis accuracy can reach 98.18%, significantly outperforming the pre-improvement model and other methods. This approach holds both theoretical value and practical significance in engineering applications.