ISC Analytics — Variance-AR1 Coupling Analysis Code and Processed Results: TEP, DAMADICS, and SWaT Validation

This dataset contains all analysis code and processed results supporting the manuscript "Variance-AR1 Coupling as a Baseline Susceptibility Predictor for Industrial Process Fault Detection: Validation on the Tennessee Eastman Process, DAMADICS Actuator Benchmark, and Secure Water Treatment Testbed" (submitted to Reliability Engineering and System Safety, 2026). The core method introduces variance-AR1 coupling C = corr(Var_W, AR1_W) computed from rolling windows of normal-operation time series as a baseline susceptibility predictor of relative fault response magnitude. Baseline coupling predicts which process variables will respond most strongly to faults before any fault occurs, using only normal operation data with no fault examples, labeled training conditions, or prior knowledge of variable roles. Contents include: Python analysis scripts (isc_core.py, tep_analysis.py, pca_comparison.py, robustness_tests.py), processed CSV result files from TEP benchmark analysis (52 variables, 21 fault types, 4 window sizes), DAMADICS real actuator plant validation (Cukrownia Lublin sugar factory, Poland), Secure Water Treatment (SWaT) infrastructure validation under cyberattack conditions, PCA vs coupling comparison results, and six robustness analyses (permutation test, leave-one-fault-out, window stability, noise robustness, alternative predictor comparison, multivariate regression). Key results: TEP Pearson r = -0.781 (p = 8.9e-12, n = 52 variables); DAMADICS r = -0.770 (p = 2.0e-6, n = 28 variables); SWaT r = -0.649 (p = 1.4e-5, n = 37 variables). Coupling outperforms PCA PC1 loading by 7.16x in fault response magnitude (p = 0.002). Permutation test confirms result robust to variable correlation (empirical p < 0.0001, 10,000 shuffles). Raw input datasets (TEP .dat files, SWaT sensor logs, DAMADICS actuator records) are not included as they are publicly available from their original sources. See README.md for data source URLs and file descriptions. Author: Ryan W. Malone, Independent Researcher, Spring TX USA. ORCID: 0009-0002-9583-232X.