A Bayesian Probabilistic Score Matrix Based Collaborative Filtering Recommendation System for Rolling Bearing Fault Identification

Abstract

As the amount of data generated by monitoring the condition of rolling bearings is increasing, matrix factorization-based collaborative filtering can effectively dig out valuable fault information from it. However, in practice, the amount of data generated by the normal state of the bearing is much larger than the amount of data of the bearing fault. As the total amount of data increases, this imbalance will become more and more and more severe, bearing fault information is often overwhelmed in it. In response to this problem, this paper starts from the perspective of mathematical statistics, a method of mean conjugate prior is proposed for the bearing normal condition data of bearing score matrix, from which the prior distribution of the probability distribution parameters of the bearing fault data is obtained. Then combined with the Bayesian method, we get the posterior distribution. According to the distribution, the random number is used to construct the Bayesian probabilistic scoring matrix (BPSM). Relying on BPSM, the collaborative filtering recommendation algorithm is used to identify different types of faults in rolling bearings. Under unbalanced data, comparing with the identification under a conventional joint score matrix (CJSM), the model built based on BPSM has a better identification effect on bearing fault state.