Reliable degradation monitoring of hydraulic turbomachinery is challenging because direct measurement of runner erosion is rarely possible during operation, while measured features are strongly affected by operating conditions. This study develops a degradation-aligned health indicator (HI) for erosion monitoring using a centrifugal pump-as-turbine (PAT) as an experimental platform. Progressive impeller degradation was introduced through controlled accelerated erosion, and cumulative mass loss was measured after each damage stage as the physical damage index for supervised HI construction. Vibration signals were acquired across multiple operating conditions and transformed into 54 candidate features representing impulsiveness, sparsity, high-frequency energy redistribution and low-frequency structural dynamics.A supervised partial least squares (PLS) framework was used to construct a sensor-derived HI aligned with measured mass loss. Variable importance in projection (VIP) scores were computed across retained PLS components, after which candidate top-N feature subsets were refitted and evaluated using PHM-oriented metrics. The final HI was selected as the smallest feature subset achieving at least 99% of the best Fitness score while maintaining monotonicity above 0.95. This selected a 31-feature HI, achieving correlation of 0.9875 with cumulative mass loss, monotonicity of 0.9809, trendability of 0.9458 and prognosability of 0.9578.Leakage-safe leave-one-operating-condition-out validation was performed using training-fold-only standardisation. The selected HI maintained strong held-out degradation alignment, with median and minimum correlations of 0.9888 and 0.9556, respectively. Compared with PCA-based and category-based indicators, the proposed HI provided better degradation alignment and physical relevance. The framework offers a transparent route for HI development in hydraulic turbomachinery, supporting degradation tracking and prognostics.

Hydraulic turbines increasingly operate under variable hydraulic regimes, creating a need for condition monitoring (CM) systems that support diagnostics and predictive maintenance, rather than only protection or threshold-based alarms. However, sensor selection in hydropower monitoring studies is often reported as an experimental detail, with limited guidance on how monitoring objectives, degradation mechanisms, measurable signatures, sensor modalities, data acquisition (DAQ) requirements and field constraints should be linked. This paper presents a sensor specification and deployment framework for hydraulic turbine CM in practice. Its main contribution is an operational decision process that links monitoring objectives to mechanisms, signatures, sensor roles, DAQ requirements, deployment constraints and site-specific sensor-suite selection. The framework was verified on a laboratory pump-as-turbine (PAT) rig using flow, pressure, vibration and acoustic emission (AE) measurements under three quasi-steady regimes defined by Q/Q_max = 0.90, 0.63 and 0.33, where Q_max is the maximum measured PAT flow. Flow measurement separated the regimes with coefficients of variation of 2.47–7.61%; pressure measurements provided hydraulic-loading and fluctuation context; accelerometers showed direction-dependent vibration responses; and Hsu–Nielsen testing confirmed AE reproducibility within the ASTM E976 criterion. Field adaptation to five Pelton, Francis and Kaplan plants showed how the same specification logic produced site-specific sensor layouts.