Federated Learning-Driven Decentralized Intelligence for Explainable Anomaly Detection in Industrial Operations
Prabhakar Marry, Y. Mounika, S. Nanditha, R Shiva, R. Saikishore · 2024
In today's rapidly evolving landscape of smart industry, timely detection of anomalies within industrial machinery stands as a critical factor in preventing breakdowns and upholding safety standards. While Machine Learning (ML) offers an automated solution, leveraging the Internet of Things (IoT) for data collection, it encounters challenges like data security risks and uneven data distribution. To overcome these challenges, this study proposes an innovative Autoencoder-based Federated Learning Framework. This framework conducts multiple rounds of training on sensor data gathered from randomly selected IoT devices. This study aims to experiment with model aggregation techniques such as FedA vg and FedAvgM, followed by a thorough comparative analysis. Furthermore, by integrating Explainable AI techniques, we enhance the transparency and interpretability of our anomaly detection models, providing stakeholders with insightful explanations of their functioning.