a large dataset of 609,934 real Modbus packets

a large dataset of 609,934 real Modbus packetsIn industrial control systems (ICS), the Modbus protocol remains a cornerstone for communication between field devices and supervisory systems. However, limited availability of diverse and realistic datasets for Modbus TCP traffic hinders the development of advanced cybersecurity solutions. In this study, we present a novel approach to synthesize Modbus TCP packet data using a Wasserstein Generative Adversarial Network (WGAN) augmented with Gumbel-Softmax sampling. Our system learns from a large dataset of 609,934 real Modbus packets, replicating their structure, byte distribution, and variability to produce synthetic yet protocol-valid packet data. We design and implement a generator-critic architecture, optimizing training with gradient penalties and leveraging PyTorch for scalability. Evaluation of the generated packets demonstrates a 93.92% Test System Reception Rate (TSRR) when interacting with a Modbus server, confirming their validity and realism. Furthermore, statistical comparisons between real and generated packet byte distributions reveal significant alignment, while diversity metrics and structural conformity ensure utility in anomaly detection and robustness testing. This synthetic data solution has promising implications for training intrusion detection systems (IDS) in resource-constrained or privacy-sensitive environments, providing a secure, scalable, and high-fidelity alternative to raw packet captures.