Sérgio Luciano Avila

A wind project is financed against a nameplate power curve, but delivers only the energy a congested grid lets it inject. We build realized power curves — what a plant achieves at its connection point — for the 1,051 curtailment-exposed wind plants of the Brazilian system, resolved across all nine curtailed-wind states, from a public pipeline of system-operatorconstrained-off records and the national generation registry. A capacityfactor decomposition separates reference, available and delivered yield, isolating the curtailment tax: the physical share of available energy the grid withholds. Fleet delivered capacity factor is 38.5% against an available 49.7% — an 11.3-point tax, 22.6% of available energy, about twice the residual gap to a reference turbine. An audit shows the operator’s constrained-off field is a model counterfactual, bounding the true loss between 18.2% and 22.6%. Exposure is locational, and its wind-speed signature rather than the annual rate separates two siting regimes: it rises steeply along one corridor carrying 80.5% of national forgone energy, and is flat to falling elsewhere, including a South whose 16.7% rate sits inside the Northeast’s own range. A plant-sited battery recovers at most about a third of it, with network headroom rather than battery size binding.

Supplying power from shore to electrify offshore oil and gas production can substantially reduce carbon dioxide emissions. This strategy, however, presents a complex planning problem involving the optimal placement of offshore collection hubs and the selection of suitable transmission technologies for extended subsea distances. This study presents a planning framework for designing offshore transmission layouts for multiple floating production, storage, and offloading units connected to an onshore grid using High-voltage alternating current, Low-frequency alternating current, or High-voltage direct current links. Distance-capacity envelopes for inter-array and export cables are developed through optimal power-flow analyses, which guide the screening of permissible connection lengths and export ratings. The feasible region is discretised into candidate hub locations. A mixed-integer linear programming model, implemented in Gurobi, determines hub placement, assigns units to hubs, and selects an admissible export option to minimise offshore transmission investment. The methodology is demonstrated using the Mero field in the Brazilian pre-salt region, where five units import 75 percent of their electrical demand from shore over distances of approximately 190 to 200 kilometres. For this case, the alternating-current configuration selects two hubs, while the low-frequency and direct-current configurations converge to a single hub. In the single-hub layout, High-voltage direct current achieves the lowest offshore-specific investment cost, and Low-frequency alternating current results in intermediate costs. Sensitivity and benchmark analyses are presented to assess the influence of demand uncertainty and to quantify the value of coordinated hub placement over independent per-asset planning, yielding a 57.8% reduction in total offshore transmission investment. In the Mero case study, the offshore search area is discretised on a 19 km grid; finer spatial resolutions are achievable at increased computational expense.

Mechanical vibration sensing often generates datasets with high dimensionality and redundancy, posing substantial challenges for the development of robust and interpretable predictive models. Drawing on a real-world dataset, this study presents an integrated approach to advanced wind turbine drivetrain monitoring by coupling factor analysis with Random Forest regression. Vibration data collected from distributed sensors are examined to identify the most relevant explanatory variables for predicting the behavior of target sensors. Factor analysis is applied to reduce dimensionality and uncover latent structural patterns, while Random Forest regression captures nonlinear relationships and ranks variable importance. The findings show that variables such as defect factor, crest factor, peak value, and RMS are the most reliable predictors of vibrational behavior. Sensors located on intermediate shafts, planetary gearboxes, and generators consistently emerge as critical sources of explanatory variables, reinforcing the importance of these components within the coupled dynamic system that constitutes the drivetrain. The proposed method enhances model interpretability and generalizability, providing a firm foundation for predictive maintenance strategies. Limitations and avenues for future research, including the use of raw signals and cross-validation across diverse operational contexts, are also discussed.

Wind turbines must operate reliably to support the current transition to renewable energy. Failures can cause significant operational and financial losses. Even low-severity short-circuits accelerate generator degradation and compromise long-term turbine performance. Severe faults, such as inter-phase short-circuits, trigger standard protection devices, but inter-turn and inter-winding short-circuits often remain undetected. These low-severity faults do not cause immediate damage but gradually reduce generator lifespan and increase maintenance requirements, highlighting the need for monitoring strategies capable of identifying subtle deviations in generator behavior. This work introduces a feature designed to identify short-circuits in wind turbine generators, tested on both PMSG and SCIG using controlled experimental setups with sensor measurements. The feature detects faults in 92% of scenarios with more than 5% of turns short-circuited for PMSG, whereas SCIG detection remains limited at 30%. These results demonstrate that the feature provides a reliable and real-time-capable indicator for PMSG, which can serve as the basis for developing more precise detection algorithms. The findings highlight the potential of the feature to improve short-circuit monitoring and enhance operational reliability in wind turbine applications.

Modern wind turbines are a complex electromechanical energy conversion system that is prone to mechanical and electrical faults. This article addresses a common type of electrical fault that affects turbines with Doubly-Fed Induction Generator, which is winding failure inside the generator, such as short-circuits between turns, also called interturn faults. A simplified electrical model of the generator with interturn fault is presented and validated with experimental data in a laboratory, and then used as support to develop a method for fault identification. Firstly, we analyzed the experimental and simulation data of stator and rotor currents, under both healthy and faulty conditions, and the negative-sequence current is observed to be an indicator of interturn faults. Furthermore, it is shown that the position of the vector of the negative-sequence current in the “dq” domain changes according to the affected phase, and this feature can be used to identify the faulty phase. Another contribution of this work, which was developed through the analysis of model-based and experimental data, is a machine learning method capable of identifying and locating the faulty winding affected by a short circuit, designed to ensure traceability in the decision process and to improve the ability to interpret the results. In addition, we introduce a generalized approach aimed at reducing the need for frequent retraining under different operational conditions. The model achieved 98% accuracy on experimental test data collected in various wind turbine operating scenarios, having been trained solely on simulated data generated simplified DFIG model. These contributions aim to promote less effort with adequate accuracy in the processes of training models, identifying faults and understanding the results for different operational conditions and different DFIG wind turbines.

Purpose: We explore the enhancement of deep learning (DL) algorithms from time series data analysis for rotating machinery monitoring. Design/methodology/approach: Our proposed method combines the strengths of three distinct neural network architectures: a Multilayer Perceptron (MLP), a Long Short-Term Memory (LSTM) and a Convolutional Neural Network (CNN). The skills of each architecture allow us to tune them to process data from time, frequency and time-frequency domains, respectively. The multi-domain stacking approach has proven effective, confirming its versatility and potential for applicability in several diagnostic scenarios on industry applications. Findings: It is challenging to work with high-quality and extensive datasets, remarkably like electrical currents. Electrical current can be interpreted in time, frequency and time-frequency domains. Multi-domain stacking DL models can be tuned to be more suitable for electrical current analysis. Multi-domain stacking approach has applicability in several diagnostic scenarios in industry applications. Originality/value: We introduce a novel approach to fault detection by exploiting machine learning algorithms and signal processing techniques across time, frequency and time-frequency domains. Our method integrates data from these domains using three specialized neural network architectures: MLP for structured data from the Fast Fourier Transform (FFT), CNN for matrix-like data from spectrograms and LSTM networks for sequential data with dependencies over time. We stack these specialized architectures to form a neural network framework that enhances data utility, improving classification accuracy rate and noise robustness. It is applicable not only to electrical signal analysis but also to vibration data, offering broad potential for preventive maintenance and operational safety across various industry applications.