SPTFuzz: A Fuzzing Approach Based on Spectral-Relational Fusion Transformer for Industrial Control Protocols

Industrial Control Systems (ICS) and consumer electronics rely on robust communication protocols to ensure reliable interactions. However, many industrial protocols lack essential security mechanisms, making them vulnerable to cyberattacks. Existing Transformer-based fuzzing approaches remain limited by their strict time-domain modeling bias, which struggles to capture the global periodic structures and long-range byte dependencies inherent in industrial protocols, while conventional mutation strategies often rely on localized heuristics that hinder efficient state-space exploration. To address these limitations, this paper proposes SPTFuzz, a Generative Artificial Intelligence (GenAI)-driven fuzzing framework for industrial control protocols. The framework employs an enhanced Transformer as a protocol-aware generative backbone and introduces a Spectral-Relational Fusion Mechanism to jointly model local syntactic dependencies and global spectral periodicities, significantly improving protocol conformity and sample diversity. In addition, a Particle Swarm Optimization (PSO)-based mutation strategy dynamically adjusts mutation probabilities according to execution feedback, enabling globally optimized vulnerability exploration. Experimental evaluation on representative protocols, including Modbus-TCP, MQTT, and S7Comm, demonstrates that SPTFuzz consistently outperforms state-of-the-art baselines in Test Case Recognition Rate (TCRR), Diversity of Generated Data (DGD), and anomaly-triggering efficiency, while maintaining strong robustness under limited-data conditions. These results highlight the potential of GenAI for enhancing the security of industrial and consumer technologies.