Application of Interpretability-Based Federated Learning to Fault Diagnosis of pumpjacks

Fault diagnosis of pumpjacks is crucial in oil extraction, but traditional centralised approaches are no longer applicable due to the challenges of data privacy and decentralised data. Federated Learning (FL), an emerging distributed learning approach, becomes a key technology for the solution by protecting data privacy while allowing multiple parties to share model updates. Although FL approaches protect data privacy, the transparency and interpretability of their model decision-making process need to be improved. To address this problem, we introduce the Logical Reasoning-based Interpretable Federated Learning (LRIFL) approach to enhance model transparency and interpretability. Under the LRIFL framework, each FL client generates logic rules from local data, which are uploaded to the server along with model updates without sharing the raw data. The server integrates these rules using intelligent logic connectors and dynamically adjusts the model weights according to the quality of the data reflected in the rules to ensure the accuracy and fairness of the model. Experimental results show that LRIFL has significant advantages in improving classification accuracy and rule fidelity, and is especially suitable for application scenarios with high data privacy and interpretability requirements.