A reproducible preprocessing and feature-engineering workflow for hydraulic transient data under controlled experimental conditions

High-frequency hydraulic measurements commonly exhibit missing segments, impulsive artefacts and timing inconsistencies that distort transient signatures and hinder the reliable interpretation of pressure and flow responses. This study presents a reproducible preprocessing and feature-engineering workflow for transient hydraulic signals under controlled experimental conditions. The proposed methodology is organised into three sequential blocks: (i) temporal reconstruction and standardisation, including missing-value handling; (ii) robust anomaly detection and spike correction using moving-median-based statistics; and (iii) feature extraction in the time, statistical and frequency domains. The workflow was evaluated using 268 controlled transient scenarios from the SANDBOX GIRS platform in Colombia, comprising valve-closure, pressure-regulating-valve and hydropneumatic tank events. Validation included a benchmark comparison against a conventional moving-average baseline and a sensitivity analysis of the moving-median window length. Across representative scenarios, the proposed workflow achieved lower RMSE values, higher NSE values and substantially improved preservation of transient gradients than the baseline, while maintaining stable performance over a practical range of window lengths. The resulting cleaned and structured dataset provides a reproducible preprocessing layer for subsequent transient interpretation, anomaly screening and data-driven hydraulic monitoring applications.