Measure. Iterate. Optimize.
calG builds a validated model of your engine from test data you have already paid for, so the search for a calibration happens in minutes at a desk instead of days in a booked test cell. The dyno still proves the answer — it stops being where you go looking for it.
Engine-out and SCR optimisation, NOx, soot and MAF estimation, across ESC, ETC, PTP, WHSC and WHTC. Eight modules, licensed individually, run by calibration engineers rather than data scientists.
How it works
calG does not replace your test cell, and it does not pretend you can calibrate without one. It changes what the test cell is for — generating good data and confirming the answer, rather than being the place you search for it. That is where the iterations go.
Feed calG the ESC, ETC and PTP runs your team has already run. It fits NOx, soot and MAF models and shows you the fit quality — actual against estimated, deviation, error bands, points inside each band — so the model is something you can interrogate rather than trust blindly.
Move a control variable or a map region and see the effect on emissions and fuel consumption immediately, for one operating point or across a whole cycle. Because the model runs locally, the iterations that used to queue for a test cell now cost minutes each.
Run optimisation trials against your emission and fuel targets with the NOx–soot–fuel trade-off visible at every step, interpolate the result into full 3D maps, and take it back to the ECU with the validation evidence that justifies it.
Measured result
The same deliverable, run two ways. The standard process validates early on the engine and then iterates through the test cell; calG removes both, which is where the 21.25 days come from.
Early validation on the engine was removed, and so was the iteration loop that sent every failed result back through the test cell. The modelling and validation moved onto the calibrator’s machine, where an iteration costs minutes instead of days.
Figures are from one customer engine-out NOx programme and are not a guarantee of results on yours. We are happy to walk through the workings.
Where calG is different
Every calibration toolchain on the market expects your team to work the way the tool works. calG was built the other way round — it is a platform we extend against your process, not a product you conform to.
calG is shaped around how your team already calibrates, at whichever step of the programme you want to apply it — engine level, vehicle level, or variant optimisation.
The Excel macros, MATLAB and Simulink models and internal know-how your team has accumulated get rebuilt as proper calG features, with a real interface and visualisation, on an agile cycle.
If your programme needs a method the product does not have yet, we build it. That is a normal engagement, not an exception.
The Python bridge runs your script — neural network, custom regression, anything — inside the calG workflow, with the same validation graphs and control-variable grid.
calG is a Windows desktop application. Engine and vehicle level optimisation happen offline, and your test data never leaves your network.
Gannet runs calibration programmes, consulting and training as well as building the tool. Support comes from engineers who have done the work, in your timezone.
Eight licensable modules
Each module is licensed on its own, so a team that only needs zone identification does not buy an aftertreatment suite. Every screen below is the real application.
Find where the engine actually spends its time — and its emissions.
Classifies operating zones across ESC and ETC from engine performance curves alone, before detailed test data exists, or from uploaded steady-state and transient data with emissions and fuel consumption included. Shows the share of time, NOx, PM and fuel each zone accounts for.
OP_ZONE_IDCheck PTP points against the ESC control area for NTE compliance.
Determines which random Part Throttle Performance points fall inside the ESC control area between the A, B and C reference speeds, then verifies measured NOx against the interpolated ESC value within a configurable threshold — 10% by default — and optimises the points that fall outside it.
NTE_ZONE_IDChange a map in one zone, see the effect on emissions and fuel economy.
Takes engine test data across the full range of input variations, your input parameter maps and your targets, and estimates the emissions and fuel-economy effect of a calibration change before you run it. Three interchangeable model backends: the built-in Gannet regression, your own Python script, or a custom polynomial of order 2–6.
ENG_OUT_OPTBalance NOx conversion against ammonia slip across the envelope.
Regression modelling of tailpipe NOx against SCR control variables including urea dosing quantity, with per-zone optimisation and present-versus-optimised cycle evaluation for steady-state and transient cycles. Zones inside ±3% deviation are flagged high confidence.
SCR_OPTPrice a calibration change before you book the dyno.
Combines DoE PTP data with a NOx coefficient model to estimate emissions for any parameter combination inside the model’s valid range, with actual-versus-estimated validation on PTP data and time-resolved transient traces. This is the module behind the programme in the case study above.
NOX_ESTQuantify the other half of the trade-off, in FSN.
The same workflow for particulate matter measured as Filter Smoke Number. Because EGR and retarded timing cut NOx and raise soot, the two estimators together let you price the trade-off before committing dynamometer time.
SOOT_ESTTurn measured points into complete, smooth ECU maps.
Builds three-dimensional lookup tables — rail pressure, injection timing, EGR rate, boost — by interpolating between measured points, so the full operating envelope never has to be measured exhaustively.
3D_MAP_INTDerive mass air flow when the sensor data will not do.
Constructs volumetric efficiency maps from test data and calculates MAF across the speed and load envelope from displacement and manifold conditions, producing ECU-ready calibration data where sensor data is missing or too coarse. One of several virtual sensors calG supports.
MAF_ESTThe workflow
Four steps, in the order a calibration engineer actually works. Nothing here assumes a statistician on the team or a rig you do not already own.
PTP, steady-state or transient runs as Excel or CSV. calG maps your columns onto its parameters and tells you exactly which rows it could not read, rather than failing quietly.
Fit inside calG with no external tools: sweep polynomial orders, or on Expert compare Gaussian Process kernels, ranked by R², MSE, maximum percentage error and the number of points inside each error band. Coefficients from Python, R, MATLAB or Minitab still import as before.
Set emission targets for one operating point or a whole cycle, then run optimisation trials across the control variables. The trade-off stays visible at every step instead of disappearing into a solver log.
Interpolate the optimised points into full 3D calibration maps and export them as ECU-ready data, together with the validation evidence that justifies the change.
Model validation
calG shows actual against estimated and the deviation between them at every stage, for PTP data and for time-resolved transient traces. A model that cannot be checked is not worth calibrating against.
The estimated trace laid over the measured one across a full transient cycle. This is the chart a sceptical engineer scrolls to first, so it is the one we lead with.
The same validation for particulate matter in FSN, with range editing so you can pull apart a specific segment of the cycle rather than trusting an aggregate number.
calG against ASCMO and CAMEO
The two tools calG is most often measured against. All three are data-driven — ASCMO uses Gaussian Processes, CAMEO uses DoE with statistical and machine-learning models.
| Dimension | calG | ETAS ASCMO | AVL CAMEO |
|---|---|---|---|
| Cost of ownershipLicence, maintenance and procurement drag | Our edgeA fraction of a global-vendor seat, with a light desktop deployment and no maintenance shock. Quoted per programme — talk to us. | Global-vendor seat pricing plus annual maintenance | Global-vendor seat pricing, heavier again with the rig stack |
| Adapting to your processWho conforms to whom | Our edgeThe platform is extended against your process; in-house methods become calG features | You adapt your workflow to the tool | You adapt your workflow to the tool |
| Modelling approachWhat is actually under the hood | Polynomial regression and Gaussian Process | Gaussian Process | DoE with polynomial, GP and neural models |
| Domain scopeHow much workflow you build yourself | Our edgeEmissions-native: ESC/ETC/WHSC/WHTC, NTE and operating-zone ID, NOx, soot and MAF estimators built in | General modelling toolbox | General toolbox plus test-bed integration |
| Who operates itSkills the team needs on staff | Our edgeA calibration engineer, through guided screens | Assumes statistical fluency | Assumes statistical fluency |
| Uncertainty quantificationKnowing when the model should not be trusted | PartialGP confidence bands, broadening across features | Native to GP models | Yes |
| Test-bench automationDriving the rig itself | Not offeredNot offered — calG attaches to your existing rig | Partial | Yes |
ASCMO is a trademark of ETAS GmbH. CAMEO is a trademark of AVL List GmbH. Neither company is affiliated with Gannet Engineering. Figures reflect our own market observation and should be checked against a current quotation.
Licensing
Four tiers, and every module also sold on its own. Licences are managed by the Gannet License Manager and stored locally — there is no seat server to keep alive. Pricing is quoted per programme, because scope varies: which modules, how many seats, and whether we are building custom features against your process. Ask and we will answer quickly.
A first look at where the emissions are
Day-to-day engine-out calibration
Full engine-out and aftertreatment development
Teams bringing their own models and methods
calGAI
Technical evaluators buy a trajectory as much as a build. Here is ours, with no roadmap items dressed up as shipped features.
Who builds it
A controls and emissions engineering firm working with OEMs and Tier-1 suppliers, with calG built out of real BSVI calibration programmes rather than as a general modelling product looking for an application.
Direct contact with the people who build it, and customisation when your programme needs something the product does not do yet. Write to info@gannetsolutions.com.
Every module documented against the real screens and the real data requirements, for licensed customers and trial users. about documentation access.
Calibration fundamentals and product training through Gannet Academy, for teams building the capability in-house.
calG ships continuously, with new features built out of live calibration programmes rather than a marketing roadmap.
Watch
Two recorded walkthroughs of the platform and the calibration workflow behind it. Both predate the current interface — the screenshots above show where calG is today.
What the platform is and the problem it was built for.
The longer walkthrough of the calibration workflow end to end.
Questions we actually get asked
Licensing is quoted per programme, because scope varies a great deal — which modules, how many seats, and whether we are building custom features against your process. Get in touch and we will give you a straight answer quickly. A 30-day trial is available.
ESC, ETC, PTP, steady-state, transient, WHSC and WHTC.
Excel (.xls, .xlsx, .xlsm) and CSV. calG maps your column names onto its parameters and reports which rows it could not parse rather than failing silently.
No. calG attaches to the data your existing rig already produces. Our roadmap reduces the number of test points you need rather than automating the runs themselves.
Yes, and this is the point of the platform. The Python bridge runs your script inside the calG workflow with the same validation graphs and control-variable grid. You can supply your own polynomial coefficients of order 2 to 6, fit entirely inside calG with no external tools, or have us rebuild your existing Excel macros and MATLAB models as proper calG features.
Windows 10 or above, an i5 or better, 4 GB of RAM, roughly 100 MB of disk, and a 1280×720 display. It is a desktop application — nothing is hosted.
Feature by feature. Each module is licensed individually, so you can take one of the tiers above or only the modules you need. Licences are managed by the Gannet License Manager and stored locally on the machine.
No. calG is a desktop application and processes everything locally. Your test data never leaves your network.
Yes. Gannet Academy runs certificate programmes in engine calibration and performance, numerical methods for engine modelling and optimisation, and model-based automotive controls — instructor-led on site, instructor-led online, or self-paced through the Gannet LMS.
Next step
Not a sales qualification call. Tell us which cycles you are chasing and where the schedule is hurting, and we will tell you straight whether calG helps.
Request a trial or quoteEvery module and the data each one needs, documented in full. It goes to licensed customers and trial users.
About the documentationAn early preview: move injection timing, EGR and rail pressure and watch the NOx–soot trade-off respond in the browser.
Open the demo