Abstract: To improve the fault diagnosis accuracy of hydropower units, this paper proposes a fault diagnosis model based on a combination of Subtraction-based Optimization Algorithm (SABO)-optimized Variational Mode Decomposition (VMD) and Improved Dung Beetle Optimization Algorithm (IDBO)-Kernel Extreme Learning Machine (KELM). Firstly, the SABO algorithm is used to optimize the key parameters of VMD (penalty factor and number of modes). The intrinsic mode functions (IMF) with the minimum compound function of permutation entropy and mutual information entropy, obtained from the SABO-VMD decomposition, are selected as the optimal components. Their corresponding time-domain feature parameters are calculated to construct the fault signal feature vector. Next, multiple strategies such as Tent chaotic mapping and adaptive t-distribution perturbation are introduced to improve the Dung Beetle Optimization Algorithm. The IDBO algorithm is then used to optimize the parameters of the KELM model, creating the IDBO-KELM fault diagnosis model for hydropower units. Finally, the model is validated using the rotor experimental platform at Wuhan University to simulate shaft system faults. The results show that the proposed method achieves a fault diagnosis accuracy of 99.375% in diagnosing shaft system faults of hydropower units, providing a new solution for high-precision fault diagnosis in hydropower units.