An Integrated Health Index, Remaining Life Assessment, and Probability of Failure Framework for Critical Electrical Assets

This repository contains the simulated (synthetic) datasets and trained surrogate models supporting the paper "An Integrated Health Index, Remaining Life Assessment, and Probability of Failure Framework for Critical Electrical Assets." All data were generated using the configuration-driven simulation pipeline described in the manuscript; no proprietary, customer, or measured field data are included, and production scoring thresholds and per-parameter weightings are replaced with generalized representative values. Contents: - Remaining Life Assessment (RLA): Monthly_Data_Cable.xlsx and Monthly_Data_Motor.xlsx simulated monthly Health Index trajectories for an XLPE cable and a high-tension motor, used to fit the hybrid Weibull–polynomial degradation model and project Remaining Useful Life. fig_rla_cable.png, fig_rla_good.png == corresponding RLA output plots. - Health Index (HI) surrogate datasets: synthetic training corpora for the three medium-voltage air-insulated switchgear subsystems == busbar chamber (BUS_BAR*.csv), cable chamber (CT*.csv), and circuit-breaker chamber (CB_OP_HI_WRT.csv, MV_AIS_CB_input.csv). Records are generated from the JSON-defined Condition Assessment Scale (CAS) rules with a fixed random seed (42), 100,000 records per subsystem, and scored through the expert-weighted hierarchical aggregation to produce ground-truth HI labels. - Trained models: CatBoost and XGBoost regressors per subsystem (*.joblib) with their ONNX exports (*.onnx) and feature-mapping files, reproducing the HI surrogate models evaluated in the paper. - Diagnostic analysis: bus_bar_correlation_heatmap.png, cable_chamber_correlation_heatmap.png, circuit_breaker_correlation_heatmap.png == top correlated variables with the learned Health Index. readme.md & readme.txt file descriptions and column definitions. Data provenance and limitations: All data are fully synthetic, produced to validate and demonstrate the methodology across the entire operating envelope (including rare poor-condition cases). To comply with confidentiality requirements, the exact production thresholds and weightings used in the deployed system are not disclosed; the shared artifacts reproduce the structure and behaviour of the scoring relationships rather than any real asset's condition history. Related publication: Accompanies the manuscript (preprint DOI: 10.31224/7381) submitted to the Journal of Electrical Systems and Information Technology (ISSN 2314-7172).