HOW WE SUPPORT OUR CUSTOMERS
SERVICES
Eight ways we work with companies and investors — from the first sketch of a product to the mentoring that gets it built.
01
MECHANICAL DESIGN
We take a product through every phase of mechanical design: from the specification and the feasibility study that decides whether an idea survives contact with physics, to the detailed design a workshop can actually build from.
The work spans aerospace, motorsport, wind energy, pumps and turbomachinery, maritime, oil & gas and power generation — sectors whose constraints differ wildly while the method does not.
We also sit alongside the team rather than deliver to it: start-ups and established companies developing a new product get project management, design and redesign, engineering, problem solving and prototype construction — whichever of those the company is short of at the time.
Mechanical designFeasibilityDetailed designPrototypingTechnical consulting
02
NUMERICAL SIMULATIONS
CFD and FEM, and the two coupled. We compute aerodynamics, turbulent flow, pressure loss, noise and cavitation; stresses and deformations in isotropic and composite materials; and the fluid-structure interaction where each drives the other — for automotive, marine, HVAC, energy, turbomachinery, aeronautical and piping applications. The same coupled models carry into biomedical work — arteries, aneurysms and prostheses — where flow and structure cannot be separated either.
Every analysis can be coupled with our optimisation algorithms, which modify the geometry to improve performance rather than only reporting how the current one behaves.
A model is only worth what it can be checked against. Mesh sensitivity, boundary conditions and solver settings are recorded alongside the results, and validated against measurement or published data wherever either exists. What comes back is a number with its uncertainty stated rather than a picture — together with the geometry and the mesh that produced it, so the analysis can be repeated or extended later.
CFDFEMCAECADPython
03
EXPERIMENTAL TESTING
A test campaign is planned for the fewest runs that still give the accuracy you need — choosing the right kind of test for the application, then the minimum number of them, saves both time and cost.
We design the benches that carry the tests, help select instruments, sensors and acquisition hardware, and are there at your facility while the prototype is on the rig.
We also run the campaign ourselves: our own laboratory, with accurate instrumentation and a wind tunnel, so a test can be set up and repeated without waiting for someone else's slot.
Instruments are calibrated and the acquisition chain checked end to end before a campaign starts, so a result that disagrees with the model is a finding rather than a doubt. You get the raw signals as well as the processed curves, so anyone on your side can re-derive what we concluded instead of taking the conclusion on trust.
TestingWind tunnelSensorsTest benchAcquisition
04
CUSTOM TOOL DEVELOPMENT
We write the software that accelerates and automates a design process, so an engineer spends the afternoon on the part of the problem that is genuinely hard rather than on the part that merely repeats.
That covers analytical models and algorithms which estimate how a complex product, process or system will perform, and digital twins of a machine or an environment that report its state and its efficiency while it runs. Coupled with our optimisation algorithms, those same models return the operating parameters and the geometries worth building rather than merely describing the ones you have.
Design and artificial intelligence are still an uncommon pairing. We build neural-network models that stand in for the expensive part of a simulation, turning a design loop that ran overnight into one that answers while the question is still interesting.
AutomationDigital twinAlgorithmsNeural networksPython
05
PREDICTIVE MAINTENANCE
A machine learning model that predicts the remaining useful life of a machine from its own sensors, so maintenance is scheduled against measured condition rather than a calendar. We built one for an aircraft engine: the network, the sensor analysis and the scheduling output wrapped in a semi-automatic tool an operator can run. Fewer spare parts held, less maintenance bought, more uptime.
The work is mostly in the data. Sensor streams arrive as CSV, are filtered for invalid readings and for the stretches recorded while a sensor was broken, then reorganised into a temporal distribution the network can learn from — input steps, future steps, output steps.
A correlation matrix then decides what is worth keeping: of 18 channels — temperature, pressure, vibration — 8 carried real influence on remaining life, direct or inverse, and the rest were noise for this purpose. An LSTM network trained on 16 failure events reproduces the trend of cycles to failure closely, and closest where it matters, in the approach to breakage.
Machine learningLSTMRemaining useful lifeSensorsCondition monitoring
06
TECHNICAL DUE DILIGENCE FOR INVESTORS
An independent read on whether the engineering behind a company does what its deck says it does. We come from the disciplines being claimed — aerospace, turbomachinery, energy, marine — so the assessment is of the physics and the maturity of the work, not of the pitch.
The team doing the reading is a mix of engineers, computer scientists and physicists, and it does not stop at the documents: where a claim can be settled by measurement, we can test the solution in our own laboratory rather than take it on trust.
The output is a written opinion you can act on: what is proven, what is modelled, what is still an assumption, and what it would take to close the gap.
The reading starts with what already exists — test reports, simulation decks, patents, code — and then goes to the people who produced it, because the distance between a document and its author is usually where the risk sits. The report is written to be acted on by an investment committee with no engineer in the room, and every claim in it says where the evidence for it came from.
Performance checkFeasibilityTeam maturityRisks
07
ACADEMY
Training delivered on request, built around a real test case rather than a slide deck. The theory of the modelled phenomena sits next to the software module that implements it and the case you then run yourself.
The introductory OpenFOAM course covers CFD theory, the software and its folder structure, meshing, rotating regions, multiphase and Paraview — some modules general, others activated on request, including a test case of your own choosing.
Groups are kept small enough that exercises can be followed one by one, and the pace follows the room rather than a fixed syllabus. Everything used stays with the team afterwards — the cases, the scripts, the solver setups — so the course keeps returning value long after the last session.
CFDMATLABOpenFOAMArtificial intelligence
08
MENTORING STARTUPS
Taking a prototype to a product is a different problem from inventing it, and it is where most of the engineering effort actually goes. We join the team rather than report to it: an R&D group drawn from inside and outside the company, working in small steps that each prove one thing before moving on.
Because the team spans several disciplines, the support covers the whole of a product rather than one slice of it: software, hardware, AI, firmware, electronics, supply chain and business development, whichever of them is the thing currently holding the product back.
That means concepts, virtual simulation environments, targeted experimental tests and beta feedback — in whatever order the product needs, with the aim of satisfying the functionality while minimising what it costs to build.
Work usually opens with a short, honest assessment of where the product actually is, which is rarely where the roadmap says it is. From there we stay involved through the first production series — the least glamorous stretch, and the one where an unnoticed design decision turns into a recurring cost on every unit built.
PrototypeIdeaExecutionDevelopmentTeamSkills
NOT SURE WHICH ONE YOU NEED?
Tell us the problem rather than the service. Most of our work starts as a question nobody has been able to answer yet.