Renewable energy projects rarely get into trouble because someone got the physics wrong. They get into trouble because a number could not be defended when it mattered. We re-derive those numbers from the physics, test them across the uncertainty, and document them end to end. That is research work, and it is most of what we do: loads, reliability and lifetime, extreme-weather survivability, and the atmospheric and ocean modelling underneath.
One modelling chain underneath all of it. What changes is the question and what gets handed over.
The two assumptions that move energy yield most: what the wind does over the site, and how much of it the turbines take from each other. Complex terrain, large arrays, neighbouring farms and offshore structures, resolved with simulation rather than a superposition rule.
See the work →Coupled aeroelastic simulation for onshore, offshore and floating machines: design load cases, component fatigue, availability and remaining life. For OEMs, certification bodies, engineering consultancies and owners' engineers.
See the work →Tropical cyclones, derechos and a changing climate. What a specific storm at a specific site does to a specific machine, carried from turbulence-resolving atmospheric simulation through to structural response and downtime.
See the work →Which dataset is fit for which decision, scored in the units of that decision rather than as a global error statistic. Whether a higher-resolution product actually changes the answer, and what combining sources measurably adds.
See the work →Probabilistic generation and load forecasts with calibrated uncertainty bands, compressed to run on the device next to the meter. Benchmarked in the open; the measurements are below.
See the work →A lender needs one to release capital. Underneath each of those numbers sit assumptions about the governing loads or the site conditions, made early in the project and rarely stress-tested again. The fleet is meanwhile changing faster than the data behind it: machines are the first of their size, sites are moving into harsher climates, and the grid underneath is run closer to its limits every year. When the record runs out, fleet averages and rules of thumb fill the gap.
A 15 MW floating turbine in a typhoon basin does not behave like the 2 MW onshore fleet behind the historical curves. Extrapolating across that gap is not analysis; it is guessing with confidence intervals.
Certification says an asset passed a standard. It does not say what a Category 4 at this site does to this machine, how long the outage lasts, or what the downtime distribution looks like. Those are separate calculations, and most projects never see them.
Distributed solar, heat pumps and EV charging are reshaping feeder loads, yet most operational forecasts are deterministic means. The decisions that matter live in the upper percentiles, and reaching them takes calibrated, probabilistic models running close to the data.
Four things we run. Where a number is quoted it is measured on a stated protocol and published, not asserted in a brochure.
Cloud platform for aeroelastic simulation, live at turbinex.app. OpenFAST 4.2.0. All 26 IEC 61400-1 Ed.4 design load cases templated, offshore included. BeamDyn for geometrically exact composite blades and OLAF free-vortex wake. Over 2,000 orchestrated simulations run in beta. Certification-style reporting: ultimate loads with partial safety factors, Weibull-weighted fatigue, Campbell diagrams.
Most wind modelling stops at the mesoscale and parameterises the turbulence away. We carry the chain down. A mesoscale outer domain with a moving 1.5 km vortex-following nest supplies boundary conditions for turbulence-resolving large-eddy simulation at 33 m and 11 m, which resolves the eddies that actually load a rotor instead of approximating them. Hurricane Laura was run this way, eyewall turbulence reproduced from the physics and published in the Quarterly Journal of the Royal Meteorological Society. It is the core underneath the extreme-weather work. What it resolves, and what it does not →
Probabilistic forecasting compressed to run on edge-class ARM hardware. On a public household-power benchmark: 0.84 MB model bundle, 0.52 ms per forecast, 80.0% empirical coverage against 80% nominal on a fully held-out year, and pinball loss reduced by 59.5% and 38.8% for active and reactive power at the 15-minute horizon. Method and figures →
The IEA-15-240 and IEA-22-280 reference turbines reproduced as open verification cases, and benchmarks published with their figures and protocols so a reviewer can disagree with the method rather than take the result on trust. Forty peer-reviewed papers sit behind the methods.
The engine was not assembled from scratch last year. It came out of a decade of funded research, and the standards work continues.
A number survives challenge when it can be traced back to the physics that produced it. That is research work, and it is most of what we do. Parametrica is a European SME, registered on the EU Funding and Tenders Portal as PIC 862445819, and open for business as a consortium partner.
Work packages in atmospheric and ocean modelling, structural and aeroelastic response, machine-learning surrogates, and uncertainty quantification and calibration. Usually the part of a proposal where the physics is the hard problem.
An SME partner at the higher funding rate, with the person doing the work named in the proposal rather than hidden behind a subcontract chain. We write our own sections, carry our work packages to completion, and expect the results to be publishable.
Send the call identifier and the deadline. We will tell you within a day whether the fit is real, what we would own, and what we would need from the coordinator.
Send the question, the site or the asset, and the date you need an answer by. If it is not something we should be doing, we will say so on the first call.
Talk to us