Hybrid MLP-GRU Federated Learning Framework for Industrial Predictive Maintenance
K. Praveena, M. Misba, Chamandeep Kaur, Mohammed Saleh Al Ansari, Veera Ankalu Vuyyuru, S Muthuperumal · 2024
Assuring the dependability and effectiveness of industrial gear, cutting downtime, and lowering maintenance costs all depend on predictive maintenance. We provide a hybrid MLP-GRU model-based Federated Learning-Enabled Advanced Predictive Maintenance Framework in this work for defect prediction and detection in industrial machinery. By using federated learning approaches, the framework is made to effectively utilize the combined intelligence of dispersed datasets while maintaining data security and privacy. Three distinct datasets, representing various types of equipment and failure scenarios, are integrated into the framework: the IMS Bearing Dataset, C-MAPSS Dataset, and Pump Sensor Dataset. The records are carefully combined into a training dataset by means of integration and preprocessing, which makes it easier to create a hybrid MLP-GRU model that can recognize intricate temporal correlations and fault patterns. The model may learn from a variety of sources without centralizing sensitive data thanks to the Federated Learning architecture, which facilitates collaborative model training across dispersed data subsets. Across dispersed datasets, the optimization layer effectively updates model parameters while decreasing loss functions by utilizing sophisticated optimization methods. The adapted framework's efficacy in defect detection and prognosis tasks across a range of industrial machinery types and fault circumstances has been demonstrated through training and assessment. All things considered, the suggested framework is a major step forward in industrial machinery predictive maintenance, providing a scalable, accurate, and privacy-preserving approach for proactive failure identification and prediction. Its use of hybrid MLP-GRU model architecture and federated learning approaches shows promising outcomes of 0.94% accuracy in practical industrial applications.