This tool processes KMD piano touchweight data and generates analysis reports including:
The problem. The KMD doesn't report downweight (DW) and upweight (UW) at a single point. It averages the force over a window of key travel, which is 2–4 mm in every dataset I've seen (it can be changed on the device, and the report uses whatever window your file says). Balance weight (BW) and friction (F) are then worked out from those two numbers: BW = (DW + UW)/2 and F = (DW − UW)/2.
On a lot of pianos the damper starts to lift before the key gets to 4 mm. Once the damper is in play the key is suddenly carrying extra weight, and that gets averaged into the KMD's figures. The result is a downweight, balance weight and friction that are skewed by the damper rather than describing the key and action. On one piano I measured both with and without dampers, the KMD's own DW and BW values were 7–9 g heavier with the dampers on.
What this tool does about it. For each key, it finds where the damper lifts on the downstroke (that's the Damper Lift Timing Analysis). Then:
Why not just use a lower window for every key? Because the numbers change with key travel even when there's no damper at all. On every damper-free piano I've looked at, friction drops as the key goes down while balance weight stays about level or goes up a bit. I put that down to the leverage changing through the stroke. So a window down at, say, 0.5–2 mm would give numbers that aren't comparable with the KMD's or with each other. The tool stays as close to the KMD's window as the damper allows, key by key. For the same reason it ignores a custom window set on the device for its own calculation: a fixed window is a guess at where the dampers lift, and the per-key approach doesn't have to guess.
Does it work? On that with-and-without-dampers piano, the calculated values with the dampers on came within about 1 g of the KMD's values with the dampers off. That's one piano, so more paired datasets would be very welcome.
If you measured with the dampers raised (e.g. the damper lift rail up), say so in the “Tell us more” form on your report by setting Damper State to “Raised”. The report is regenerated without damper lift analysis and the KMD's values are used throughout. Otherwise the tool may mistake some other bump in the curve for a damper.
Problem keys. The high/low friction and downweight checks in the Problems list use F* and DW*, so a damper lifting early no longer gets a key flagged for heavy downweight or friction. Those entries are labelled “(F*)” or “(DW*)”. Because the damper skew is gone, the spread across the keyboard is smaller, so the same 2 standard deviation rule picks up smaller differences than before. It's also measuring against the average of the whole keyboard, so a bass that is meant to be heavier, or a treble that is meant to be lighter, can show up there too.
This is provided for informational purposes only. Chances are it's only going to function reasonably if your piano is in reasonable condition and regulation. If Renner change their data format from this device, this will break, and I may not be responsive in fixing it.
I have no relationship to Renner or KMD people. I was merely asked to help set up an "Innovation Centre" at the APTTA convention in Christchurch, New Zealand Oct 3-5, 2025. The suggestion was that the KMD might be an interesting addition. I borrowed one and promptly went down a silly thought experiment rabbit hole.
You are uploading data sets onto my server, so it becomes mine. Please don't put any identifying customer details or the piano's serial number in the name of the file. I don't want to know. After I strip all of that out I may use the dataset in a presentation sometime. I'll probably let you know before I do - particularly if you've given me your email address. Chances are I know you anyway.
You have the option to make your own datasets "public" if you have a piano that you think would be of interest. Please, please, make sure you don't have customer identifying details in it before you do that. If you do and I find it, I'll remove it. But particularly if you have a good dataset that you're proud of, please pass it along. For example, if we take out the known and common issue of the new dampers settling and lifting early, the "Sunken D" dataset that I have made public I think is an example of a very nicely regulated piano (just probably a tight bushing at F3).
There's a good chance that at some point I will send these data sets through some machine learning algorithms, but those will all be hosted in a closed system by something that I control, so data isn't going anywhere like to the Googles and Facebooks of the world.
Nov 2026 - Work following presentation to NZPTTG in Oct 2026. Address damper lift skewing of data points. Various improvements and documentation mods.
Sep 2025-ish - Initial release