Journal of Networking Technology Tran Trung Duy 17 4 2026 https://doi.org/10.6025/jnt/2026/17/4/204-221 https://www.dline.info/jnt/fulltext/v17n4/jntv17n4_2.pdf Programmable Logic Controllers (PLCs) are foundational to modern industrial automation, yet the datadriven characterisation of their input output dynamics at typical process historian sampling rates remains underexplored. This study investigates the dominant input output relationships within a closed loop PLC industrial control system using a high resolution dataset of 100,000 time stamped observations. We employed a sequential analytical pipeline comprising bivariate correlation, multivariable ordinary least squares (OLS) regression, random forest modelling, and dynamic ARX extensions, followed by rigorous residual diagnostics. Our findings reveal that the control loop behaves primarily as a near deterministic static mapping rather than a dynamic system with memory. The PLC output is overwhelmingly governed by a binary emergency stop interlock and a linear gain on the continuous input signal, achieving a predictive R2 of 0.9962 and an RMSE of 0.0974 V. Augmenting the model with lagged variables yielded negligible improvements, and residual analysis confirmed the absence of serial correlation. Discrete events, particularly emergency-stop activations, dictate absolute operating levels and safety transitions, while continuous variables like valve status play secondary roles. These results demonstrate that a compact, linear static surrogate is highly adequate for digital twin construction and anomaly detection at this temporal resolution, bypassing the need for complex dynamic modelling.