Abstract: Aiming at the issue that the characteristics of hydropower unit vibration faults are difficult to extract under strong background noises, this paper presents a fault diagnosis method using the techniques of stochastic resonance (SR) denoising and multidimensional permutation entropy (MPE) for extraction of the characteristic vectors from vibration signals. This method first denoises a vibration signal using stochastic resonance to enhance its stochastic resonance, then uses MPE to extract its feature vectors. Taking the feature vectors as input, an improved particle swarm algorithm and support vector machine model is able to achieve identification and diagnosis of the signal faults. Our simulations show that the method enables the fault diagnosis of hydropower units with high accuracy.