Research on Instantiation Method of Object-Oriented Bayesian Network for Fault Diagnosis Based on Dynamic Weight Parameter Transfer

Bayesian Networks (BNs) are extensively utilized for equipment fault diagnosis, and Object-Oriented Bayesian Networks (OOBNs) offer an effective framework for efficient model construction. In OOBNs, parameter learning, also referred to as instantiation, generally requires an adequate number of training samples. In practical fault diagnosis scenarios, however, such samples are often limited. To address this limitation, an instantiation approach based on variable-weight parameter migration is proposed. When training samples are insufficient, fully instantiated objects from related domains are employed as sources, and their parameters are transferred to enhance instantiation in the target domain. The transfer weight of each source object is dynamically adjusted according to its similarity to the target domain, thereby mitigating the risk of negative transfer. Experimental validation confirms that the proposed method significantly improves OOBN instantiation performance in data-scarce conditions.