In materializing the previous objectives, the WINDMIL project:
1. Developed data-driven performance assessment methods, which are able to account for the lack of precise input (loading) information as well as environmental and modeling uncertainties. These schemes are holistic in that they operate across two temporal scales, namely the short- and long-term one. The former refers to the handling of sudden anomalies typically linked to extreme events (strong winds, waves, earthquake), while the latter refers to deterioration processes that evolve across a lengthier time span and which form an adverse factor for extending the life-cycle of these structural components, as fatigue damage.
2. Established a minimum intervention principle in order to deploy optimal sensor configurations, able to balance costs versus quality of information.
3. Proposed a new monitoring paradigm for WT facilities which relies on sensory feedback for the extraction of quantifiable metrics facilitating the LCA of WTs both in the short- long-term, in an effort to manage these systems from cradle-to-grave.
We show that by exploiting structural monitoring information and combining this to the typically available SCADA information stream, it is possible to reliably estimate and forecast processes that relate to damage accumulation (e.g. fatigue, wake induced effects); a process which underpins the life expectancy of these systems.
The methodological tools that form the core of WINDMIL have been organized along two main interacting tracks:
- In the forward track, we focus on the development of computational tools that are able to offer a virtualization - a digital representation - of wind turbine structures. Critical in this procedure is the tackling of uncertainties relating to diverse loadings (wind, wave, wake effects, etc), site-specific conditions (e.g. soil-structure-foundation interaction), as well as modeling simplifications.
- The inverse track deals (i) with the extraction and handling of information obtained via monitoring and (ii) with the derivation of data-driven system representations. The latter includes parametric and nonparametric representations, as well as time-invariant and time-variant models for tracking the evolution of the WT dynamics.
The novel methods developed as part of WINDMIL are fueled by the synergy of the forward and inverse engineering, building on the fusion of data with models. We term such schemes, coupling physics with data, hybrid.
In the forward track, WINDMIL delivered a methodological framework for developing multi-type models (physics-based, data-driven or hybrid) that can be used for reality enhancement. In other words, these digital and data-powered models can simulate the complex turbine dynamics with affordable computational cost at the component (e.g. blade, see Mylonas & Chatzi, 2021/Sensors) and system level (e.g. wake effects, see Avendaño-Valencia et al. 2018/ICVRAM/ISUMA). Since these representations require the use of models of diversified granularity, WINDMIL proposed an ensemble learning approach to aggregate output from multiple simulators and measurements (Abdallah, Tatsis, Chatzi, 2020/RESS).
In the inverse track, WINDMIL devised a powerful hybrid simulation tool relying on measurements of structural vibration from WT components (e.g. blades, tower) at sparse yet optimally chosen locations, combined with the information stemming from a reduced order model (ROM) of the structure. The method provides reliable forecasting of the turbine's response (e.g. in the form of strains) at critical "hotspot" locations (Noppe et al., 2018/ISMA, Tatsis et al., 2017/Eurodyn, Tatsis et al. 2018/IALCCE).
Finally, one of the main challenges for deploying Structural Health Monitoring (SHM) methodologies on in-service WTs lies in the variability of Environmental and Operational Parameters (EOPs) as reported by the standardly assembled SCADA. WINDMIL put forth a stochastic modeling approach for tackling this challenge (Avendaño et al., 2020/MSSP; 2017/Frontiers, Bogoevska et al., 2017/Sensors).
Under availability of the described data-driven indicators, or (hybrid) indicators relying on the fusion of data and models, it is possible to quantitatively describe the performance of these complex dynamic systems. This allows the subsequent development of decision support tools. In (Abdallah et al., 2018/ESREL) we develop a framework relying on use of decision tree learning algorithms to detect faults, errors, damage, patterns, anomalies and abnormal operation by running wind turbine telemetry data through the decision tree.
