A Chassis Dynamometer only delivers repeatable results when operators control physical setup, thermal state, and data interpretation, not just rely on annual calibration.
You ran the same vehicle through the line twice and got two different readings. Why?
Anyone who has watched a pass/fail margin drift across repeat runs on a Chassis Dynamometer knows the feeling. The vehicle didn't change. The test cycle didn't change. So why does the output wander?
Most published guidance stops at factory calibration, as if a once-a-year certificate is the whole story. On the floor, the repeatability question usually lives somewhere else: how the vehicle sits on the rollers, how warm things are when the run starts, how the data gets smoothed afterwards. Those three layers — physical setup, thermal state, and interpretation — are where most operators can tighten scatter without buying anything new.
Is the vehicle actually sitting the same way every cycle?
A Chassis Dynamometer only knows what the tires tell the roller. If the contact patch shifts between runs, the numbers shift. Start with the small things that nobody writes down.
Restraint force and tie-down geometry. On a motorcycle or light-vehicle dyno, the strap or hold-down arrangement changes how much vertical load the tire delivers to the roller. A strap overtensioned on the first run and slacked on the second will not only change slip; it will change the effective rolling radius at the contact point. Mark your strap positions. Use the same hook points. If your restraint system has a tension indicator, record it — not for the certificate, but because when results scatter you can go back and see whether geometry was constant.
Tire pressure and wheel alignment on the roller. Tire pressure changes contact patch area and, therefore, the relationship between vehicle speed, roller speed, and parasitic loss in the tire itself. On a two-wheel test line, even a small twist in the rear wheel alignment against the roller changes side-load losses that won't show up on any display but will flatten or inflate your reported power. Set pressures to the same number before each run. Visually check that the tire is tracking center, not skating toward a flange.
Roller surface condition. Glaze buildup, moisture, and embedded grit all alter friction without changing anything in the load cell reading. A quick wipe and a check for rubber buildup between shifts costs a minute and removes a variable that ambient correction factors can never compensate for.
What does "warm enough" actually mean for repeatable numbers?
The test procedure probably says "stabilised operating temperature" but rarely tells you which temperature matters. Three temperatures matter on a dyno, and they stabilise at different rates.
Powertrain temperature. Engine oil and coolant need to be in the real operating window. A vehicle that arrives cold and gets a thirty-second warm-up will read differently on inertia simulation than one that has been held at normal temperature for a few minutes. The difference is not dramatic on a steady-state point, but on a transient emission and power correlation run, it can swing the result by more than a typical operator expects.
Tire temperature. Tires heat through deformation every revolution. A cold tire has higher rolling resistance and a slightly smaller effective radius. On a Chassis Dynamometer that simulates road load through rolling resistance calibration, the calibration itself was done at a reference tire temperature. If you are running a vehicle on cold tires that haven't been pre-rolled, the parasitic loss absorbed in the contact patch is not what the calibration assumed. A short rolling warm-up — even idle speed on the rollers — brings the tire into a range where repeatability improves.
Environmental temperature and soak. Ambient air temperature and barometric pressure feed into the ambient correction factors that rescale raw outputs to reference conditions. That math is correct, but it corrects for density, not for the fact that a vehicle soaked overnight at 5 °C has different lubricant viscosity, battery internal resistance, and ECU fueling than one parked in a 25 °C bay. Where possible, record soak temperature alongside ambient at the time of the run. If your data review shows a systematic difference between morning-first and afternoon runs, soak is often the invisible variable.
Are your correction factors hiding a real drift?
Ambient correction factors exist to make results comparable across days and climates. They do not make a poorly prepared vehicle behave like a well-prepared one. This is a common confusion that turns correction math into camouflage.
Suppose a station runs the same reference vehicle every Monday and plots corrected power. The corrected values track flat for a month, then start dropping. The operator assumes seasonal change because the correction factor shifted. In reality, a cooling fan relay had begun sticking, and the engine was running hotter — a fault the correction factor isn't designed to catch.
Plot both raw and corrected values side by side. If only the corrected line is drifting, the error likely sits inside the correction input — temperature probe location, barometric sensor drift, or humidity reading error. If both raw and corrected drift in the same direction, something mechanical is changing, and no amount of math will fix it. This simple habit keeps operators from over-trusting the displayed number and looking for root causes in the wrong place.
Where does measurement drift creep in between calibrations?
Rolling resistance calibration, speed capture, and load-cell zero all drift at different speeds. Between formal calibrations, the following low-cost checks catch most of the degradation before it shows up in test results.
Load-cell zero and bridge. Run a zero verification with no vehicle on the rollers at the start of every shift. If zero drifts more than a few counts across the day, something in the wiring, junction box, or environmental seal is deteriorating — not your test technique.
Speed signal verification. Compare the dyno's reported roller speed against an independent tachometer or known-frequency pulse source once a week. Speed errors propagate into every calculation because vehicle speed anchors both inertia simulation and emission sampling timing.
Parasitic loss check. Roll the rollers at a set speed with no vehicle and observe the residual drag reading. Compare it to the value recorded during the last calibration. An increase signals bearing wear or brake drag inside the dyno, and it will absorb power that should be attributed to the vehicle.
Document these quick checks on a simple sheet. When a result later raises a question, the sheet tells you immediately whether the instrument was behaving the day it happened.
How much smoothing is too much in post-processing?
Operators often over-smooth scatter in the belief that a cleaner curve is more accurate. Sometimes it is. Sometimes it is just quieter. For a Chassis Dynamometer used in compliance or pass/fail work, aggressive filtering can mask a real overshoot during a tip-in that would have failed the vehicle.
The test standard usually specifies allowable filtering. Stay inside it. Produce a raw trace alongside the smoothed one and glance at the raw trace at known trouble spots — gear shifts, acceleration tips, brake application points. If the raw trace shows a spike that the smoothed version removes entirely, investigate the spike before trusting the smoothed result. It could be a real event. It could also be a momentary roller slip that your restraint setup didn't prevent. Either way, you want to know.
The small things that separate a defensible run from a debatable one
Repeatability on a Chassis Dynamometer is rarely lost in one big failure. It slips through multiple small variables that each move the reading a little and accumulate unnoticed. Consistency in vehicle positioning, purposeful thermal preparation, a healthy mistrust of correction math, and brief between-shift instrument checks are what keep those small variables from accumulating. The hardware doesn't change between the first run and the third. The operator's discipline does — and that's the lever most stations can move immediately.
Next step for your station. Pick your most frequent test — a pass/fail point, a reference vehicle run, or a disputed retest — and compare the last ten executions side by side. Look not at the headline number but at the raw trace, the vehicle preparation notes, and the ambient log. Where the spread is wide, circle whichever of the above four areas had least control that day. Fix that one thing first, before reconsidering equipment. In most stations, the repeatability gain is measurable within a single week.