This paper proposes a predictive maintenance framework based on Federated Learning to address the challenges of data privacy and efficiency in Industry 4.0 manufacturing. The framework employs a combined network called a one-dimensional convolutional neural network (1DCNN) and a bidirectional long short-term memory (BiLSTM) network (1DCNN-BiLSTM) model for predictions, enabling cross-client collaborative training without sharing raw data. Using the Microsoft Azure Predictive Maintenance dataset, our experimental results demonstrate that Federated Learning can achieve 81% prediction accuracy, representing a 27% improvement over individual models. The federated model maintains at least 71% prediction performance even with 50% reduced training data or 25% client participation while showing excellent generalization ability on unseen data. These results validate the framework's effectiveness in balancing data privacy, efficiency, and model performance in real-world applications.
Robust Predictive Maintenance Framework Based on Federated Learning
2 Views
1 Downloads