Abstract: Aiming at the uncertainty and non-stationarity of bearing vibration signal as well as the slow convergence rate and poor stability of BP neural network learning algorithm, a rolling bearing fault pattern recognition method based on cloud theory and ensemble extreme learning machine was proposed. The pre-processed signal was cloudized to generate the signal cloud under different states of rolling bearing. Three parameters that can determine the signal cloud distribution were as follows: expectation, entropy and hyper entropy. These three parameters were taken as the characteristic quantities that characterize the bearing state and the original bearing state data set was constructed. Then, the fault feature data set that was normalized sent to the ensemble extreme learning machine for identification. The results showed that the cloud-ensemble extreme learning machine method could effectively realize bearing fault pattern recognition. Compared with the traditional neural network recognition method, the method had higher recognition rate and stability, and the ensemble extreme learning machine had a better performance in anti-noise.

Key words: rolling bearing, fault diagnosis, cloud theory, ensemble learning, neural network