Fixing Flywheel Power Readout Errors from Encoder Slip

You just wrapped a flywheel leg-curl session with an athlete who trains eccentric overload twice a week, and the app is showing an eccentric-to-concentric power ratio of 214%. On paper that means she braked the flywheel with more than double the power she used to spin it up - a number well outside anything published on isoinertial training, and not one you should trust just because a screen printed it with two decimal places. Vicens-Bordas et al. (2019), in a meta-analysis of 15 randomized controlled trials, reported eccentric overload effect sizes of 0.58 to 1.21 versus conventional resistance training, and the supervised studies underlying that pooled estimate generally show eccentric:concentric ratios clustering closer to 105-140%. A ratio above 180% in a real session is not a breakthrough; it is almost always the strap slipping on the shaft, a winding fault on the cone, or the encoder losing pulses at high angular velocity - three separate failure modes that all print the same symptom on the screen.

The encoder itself is rarely broken. It is faithfully reporting rotation it can detect, and after a slip or a winding fault, what it detects no longer matches what the flywheel is actually doing. This guide walks through where that mismatch originates, how to inspect for it in under five minutes, and a short verification sequence that tells you whether the number on the screen is trustworthy before you make a training decision off it.

Most isoinertial devices measure power indirectly. A rotary encoder mounted on the flywheel's axle counts pulses as the shaft turns; firmware converts pulse timing into angular velocity, multiplies by the flywheel's fixed moment of inertia (typically 0.025-0.150 kg·m² depending on which disc or cone is loaded) to get torque, and converts that into linear power at the strap using an assumed effective radius. That last step is where trouble usually starts. On a plain cylindrical shaft the radius never changes; on the conical shafts most commercial devices actually use, the effective radius shrinks as the strap winds toward the narrow end and grows again as it winds back out, and the power calculation depends entirely on the strap tracking that cone in one continuous, predictable spiral. If the strap crosses over itself, bunches, or slips against the shaft surface instead of gripping it, the real radius at any given instant stops matching what the firmware assumes - and torque, velocity, and power all inherit that error downstream.

Four distinct mechanical faults produce this symptom, and they do not announce themselves the same way. The table below separates them by what is physically happening, the direction and size of the error each tends to cause, and how to spot it during a normal session.

Vicens-Bordas et al. (2019), whose effect sizes anchor the ratio range cited above, pooled trials using flywheel devices with different moment-of-inertia settings and mounting geometries - a real limitation, since the typical range is itself averaged across studies with varying calibration rigor, not a single validated ceiling. Standardizing your own device's strap and winding, rather than chasing a published number, is the more reliable target. Any one fault above can push a ratio well past what is plausible; two together - common, since a glazed strap also slips into overlapping wraps more easily - can double it. None require replacing the device, and all four can be inspected by hand in the time it takes to rerack a set.

Start here, because a compromised strap invalidates whatever the encoder reports regardless of how carefully everything downstream is configured. Unclip the strap and run it slowly through your fingers along its full length, checking three things: a shiny or hardened patch where the surface should be uniformly matte, fraying or thinning at the edges (most common near the carabiner or D-ring end where load concentrates), and whether the strap width still matches the shaft's groove - a strap that has stretched even 2-3 mm wider than the groove will ride up and lose consistent contact under load. Manufacturer wear guidance typically calls for replacement somewhere between 150 and 300 heavy eccentric-overload sessions depending on strap material, but session counts drift - explosive, high-force work wears a strap faster per session than lighter loading, so a visual and tactile check beats a counter alone. Replace at the first sign of glazing rather than waiting for visible fraying; by the time fraying shows, the device has likely been reporting inflated numbers for several sessions already.

Even a new strap produces inconsistent readings if it winds onto the cone differently rep to rep. Mark a single reference line on the cone or shaft with a paint pen at the point where the strap should sit when fully wound, and use that same mark every session for that athlete and that exercise. Before each rep, rewind the strap by hand so it lies in one flat, continuous spiral with no crossover, checking that consecutive wraps sit edge to edge rather than overlapping or leaving a gap - a gap of even one strap-width lets the strap shift laterally under tension and land differently on the next unwind. Pull the cord fully taut at the end of the rewind instead of leaving a hand's width of slack; athletes rewinding quickly between reps to save time are the most common source of the slack-at-extension fault from the table above. Beato and Dello Iacono, in a 2020 Frontiers in Physiology review of flywheel training implementation, identify standardized strap winding and a fixed cone reference point as core reliability requirements for isoinertial devices - though their recommendations come from device engineering principles and practical consensus rather than a head-to-head reliability trial, so treat the tolerances here as a starting point to validate against your own device. On devices with multiple attachment points, log which one is used per exercise the way you would log which bar an athlete uses for barbell VBT work, since switching mid-block reintroduces the same inconsistency a strap swap would.

Once the strap and winding are standardized, confirm the encoder itself is counting correctly with a manual spin test that takes about ninety seconds. With the flywheel disconnected from any load, wind the strap fully as you would before a working set, mark the starting position, then rotate the shaft by hand for exactly 10 full revolutions at a slow, even pace while a second person counts and watches the app's live pulse or RPM readout. Ten hand-turned revolutions should register within about 3-5% of ten on the display; a reading off by more than that, especially if it is consistently low, points to pulse dropout or a partially disengaged encoder rather than anything related to strap winding. Repeat the same ten-revolution test at a faster hand-cranked pace, closer to what the eccentric braking phase actually produces on a heavy set. If the slow-speed test passes but the fast test drops noticeably below expected count, that is the signature of velocity-dependent pulse dropout described earlier, and the fix is a firmware update where the manufacturer offers one, or a hard load ceiling below the speed where dropout begins if not.

Rather than changing everything in one session, isolate each variable across four short sessions at a fixed reference load the athlete has used before, run for 6 reps.

If the ratio is still above roughly 160% once all three are controlled, stop trusting the device's power number for that exercise until the manufacturer has looked at the encoder itself - at that point the error is more likely a hardware fault than anything fixable through strap or technique changes.

An amateur handball player (bodyweight 78 kg, two years of flywheel leg-curl experience) produced the 214% ratio referenced at the start of this guide during a standard 0.075 kg·m² hamstring session. Applying the elimination sequence above: session 1 confirmed the baseline 214% ratio on a strap that showed a visible glazed patch near the D-ring and two overlapping wraps on the cone. Session 2, after replacing the strap, brought the ratio down to 178% - a meaningful drop, but still implausible. Session 3, after marking a winding reference point and rewinding by hand before every rep, brought it to 148%. Session 4's manual spin test came back within tolerance (10.2 counted revolutions against 10 hand-turned), confirming the encoder itself was not the remaining issue, and the ratio across three subsequent sessions settled between 118% and 131% - squarely inside the range Vicens-Bordas et al. (2019) associated with genuine eccentric overload in their pooled data. Nothing about this athlete's actual eccentric strength changed across the four sessions; the equipment simply stopped lying to the app about what he was doing.