Drivability calibration

A complaint arrives in words: "hesitates off idle", "shift feels harsh when warm", "surges in traffic". The engineering version of that sentence is a manoeuvre, a set of channels and a window in time. Once the manoeuvre is repeatable, two runs can be laid on the same axis and the difference becomes a number instead of an opinion.

The channels that usually decide it are pedal position and its gradient, engine speed and its derivative, indicated and effective torque, throttle and wastegate positions, lambda and fuel correction paths, gear and clutch state, and the transient compensation flags. Most are recorded at a raster fine enough that resampling them to a single rate would blur exactly the moment under investigation.

Two recordings of "the same" manoeuvre never start at the same point in their files. Unless one run is re-timed against the other, every downstream comparison is read at a slightly different moment, and small real differences hide behind a large artificial one. A time offset belongs to the file, not to a signal, so it has to move every channel of that run together.

Reads matter too. An ECU signal holds its value until the next update, so a cursor should report the last sample at or before it — never an interpolation between two. When a value is carried rather than measured, it should say so. That is why MDX marks held reads and reports each channel at its own raster instead of resampling everything to a common rate.

Drivability is not one map. It is a family of control functions, each with its own reason to exist, and a complaint usually belongs to one of them. Naming the function before touching a value is what separates a calibration change from a guess.

The transient path is where most of the work sits: wall-film compensation, which replaces the fuel that lands on the port wall during a fast load change; transient mixture control, which decides what lambda is targeted while the air path is still moving; and the torque filter that shapes how quickly a pedal change becomes a torque request at all. Around them sit the coasting functions — deceleration fuel cut-off, its entry threshold and the engine speed at which injection resumes — and the start and restart fuelling that governs the first seconds of a drive.

Once the behaviour is characterised, the change lands in a calibration object: a base map, a correction curve or a limit. CIE Pro is where that trade-off is modelled — a designed set of test points, a response surface fitted to them, and constrained optimisation that respects the boundaries the engine must keep. The optimum exports as A2L, CDF or HEX.

Hexx applies it to the calibration image itself, decoded through the A2L so you edit named maps in engineering units rather than bytes, with checksums recomputed on write. HELIX carries the change as a reviewable package through validation gates into a release you can trace back to the run that justified it.

One caution worth stating, because it costs teams whole campaigns: several drivability functions are not symmetric. A filter ramp that reduces the pedal work a driver spends on the acceleration side can increase it on the deceleration side, because the human being in the loop responds to the two directions differently. When a parameter behaves like that, the decision usually belongs at a transparent threshold rather than a felt slope — and you only see it at all if the two sides are accounted for separately, with their signs kept.