To identify broken rotor bar faults in induction motors accurately and rapidly, this paper illustrates a novel method to diagnose broken rotor bar fault on the basis of Hilbert Modulus and BP Neural Network evolved by Chaos Particle Swarm Optimization(CPSO). Firstly, Hilbert Modulus of stator current can transform the power frequency component into DC component to weaken the influence of the fundamental frequency signal in stator current, which can help extract the feature vector accurately. Compared with BP neural network, CPSO-BP neural network have superior initial weights and can strengthen the classification correctness. As a result, the experiment reminds the effectiveness and superiority of the proposed method.