Fault Diagnosis Method Based on Improved Deep Boltzmann Machines

With the increasing complexity of mechanical equipment, the traditional signal-based fault diagnosis methods cannot meet the current needs of fast, accurate and intelligent fault diagnosis due to its low efficiency and over-reliance on experience and subjective judgment of diagnosticians. Deep learning has powerful feature extraction and pattern recognition ability, and once the model is established, it can perform rapid pattern recognition. Based on this, a fault diagnosis method based on deep Boltzmann machines is proposed in this paper. Firstly, to solve the problem that DBMs can only deal with binary data, the Gaussian units are used to replace the binary visible units of the deep Boltzmann machines to construct the improved deep Boltzmann machines model, enabled the deep Boltzmann machines to process real-valued data. After the model is constructed, it is applied to process vibration signals for fault diagnosis. We present result on the CWRU bearing datasets, which shows that the improved DBMs learn generative models well and are good at fault recognition tasks.