A Deep Learning Approach for Rolling Bearing Intelligent Fault Diagnosis

In order to improve the accuracy and diagnosis speed of automatic identification of instrument fault types by deep learning technology, a new and improved fast intelligent diagnosis method is proposed, which uses a small data subset (Mini batch) to optimize the K-means algorithm, which can be used for the diagnosis of instrument fault types. By comparing the experiment of Mini batch K-means+ algorithm with K-Means algorithm, the fault diagnosis of Mini batch K-means+ algorithm effectively reduces the number of training parameters and convergence time of traditional K-means algorithm. The results show that the identification time is only 16.60ms, which increases by about 34%. Compared with Birch algorithm and K-mean algorithm, it can be seen that the diagnosis accuracy and recall rate are basically the same, which verifies the feasibility of the fault diagnosis of Mini batch K-means algorithm.