FedMqADV: A Unified Framework for End-to-End Evaluation of MQTT-Based Federated Learning in Adversarial Setting

Industrial IoT and smart factories increasingly use AI to address operational challenges. Over distributed networked control systems, federated learning enables decentralized AI training with enhanced privacy protections and, recently, more and more security features such as robust aggregation, homomorphic encryption, and differential privacy techniques. However, deploying such emerging robust techniques across factory systems challenges real-time and efficient process monitoring. This research introduces FedMqADV, a scalable MQTT broker performance benchmark framework for handling federated learning across multiple distributed clients in an adversarial setting. The framework supports several open-source MQTT brokers such as Eclipse Mosquitto, NanoMQ, and VerneMQ and integrates adaptive adversarial attacks and defense strategies tailored for industrial IoT applications. We evaluated communication efficiency and reliability under different network conditions in a distributed water treatment plant based on specific application requirements. The results demonstrate that NanoMQ offers the optimal balance between privacy, reliability and performance; making it the most suitable choice for resource-constrained environments.