Contribution au diagnostic et au pronostic par mesure indirecte des actifs d'énergies renouvelables à partir des données SCADA avec quantification d'incertitude

Abstract

The global energy transition relies on the increasing integration of renewable energy sources, notably wind and hydropower, whose sustainability depends on the reliability and availability of production assets. Subject to harsh operating conditions, these systems are exposed to degradation mechanisms such as cavitation or fatigue of critical components. In this context, Prognostics and Health Management (PHM) constitutes a key lever for anticipating degradation and reducing unplanned outages. PHM approaches rely on the availability of reliable data. Condition Monitoring Systems (CMS) offer high sensitivity to degradation but remain sparsely deployed due to economic and operational constraints. Conversely, SCADA data are available across all installations but suffer from limited temporal resolution and partial observability of degradation mechanisms. In collaboration with Hydro-Québec and Power Factors, this thesis proposes a PHM framework exclusively based on SCADA data, aiming, when possible, to replace CMS with indirect measurement approaches in order to ensure continuous and fleet-scale deployable monitoring. The contributions focus on the development of data-driven indirect measurements, their exploitation for diagnosis and prognosis despite the limitations of SCADA data, and the integration of valid and well-calibrated uncertainty quantification. Two industrial case studies are investigated. The first addresses the monitoring and diagnosis of erosive cavitation in Francis hydraulic turbines, where an artificial intelligence-based indirect measurement approach enables the estimation of mass loss due to erosive cavitation from SCADA data, replacing CMS while ensuring transferability between turbines. The second focuses on the prognosis of approximately 200 commercial wind turbines using SCADA data only, through prognostic classification of thermal anomalies followed by remaining useful life prediction using health indicators adapted to SCADA constraints. All approaches integrate uncertainty quantification methods, in particular conformal methods. The results show that, for certain observable degradation mechanisms, SCADA data, combined with indirect measurement approaches, constitute a credible alternative to CMS for monitoring, diagnosis, and prognosis, paving the way for large-scale industrial deployment.

Date	28 Apr 2026
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Antoine Tahan (Supervisor) & Martin Gagnon (Co-supervisor)