Bearing Fault Diagnosis Under Small Data Set Condition: A Bayesian Network Method With Transfer Learning for Parameter Estimation

Bearings are broadly applied in various types of industrial systems. Fault diagnosis, as a promising way for reliability modern internet thing applications, has attracted increasing attention from both academia and industry fields. Being ideal modeling inference tool uncertainty situations, Bayesian network (BN) is becoming increasingly popular many However, practical uncertain complicated engineering surroundings, it’s difficult or expensive to collect massive labeled fault data the sake diagnosis model learning. To address issue BN parameter learning under small set conditions, this paper proposes Varying Coefficient Transfer Learning (VCTL) algorithm based on aggregation transfer learning, that considers knowledge resource domain relevance contributions. The balancing weight function designed determine whether task activated. Relevance factors proposed measure target parameters quantitatively, by combing information domains with those obtained domain, using maximum posterior (MAP) likelihood estimation (MLE). Finally, aggregated initial domain. Based VCTL, bearing approach verified. experimental results show that, condition set, accuracy VCTL varying coefficient better than MLE algorithm, MAP state-of-the-art method, local linear pooling (LoLP) algorithm. Under sufficient approaches classical MAP, correctness Moreover, we illustrate successful application real-world case where had access expert-provided real data.