Blockchain-Driven Federated Learning for Secure Industrial Predictive Maintenance

Factories with similar machines could build better failure-prediction models by sharing data, but most do not want to share their raw records with competitors or a central server. Federated learning (FL) solves this problem by allowing each factory to keep its data locally while training a shared model. However, FL still depends on a central server and does not clearly track contributions. This paper shows a framework that combines FL with a lightweight permissioned ledger. Each factory trains a local model, signs its update, and stores only a hash of the update on the ledger. This provides data integrity, an audit trail, and controlled participation. Using the AI4I 2020 predictive-maintenance dataset across five simulated factories, the framework achieved a ROC-AUC of 0.949, close to the centralized model’s 0.974 and much better than a single factory’s 0.790. The ledger added only about 6% extra training time and required very little storage. This helps identify and remove malicious participants, making industrial data sharing more secure and trustworthy.