I’ve been working on way more things lately than I ever expected, and as such updates have been slow. There should be an email about the Maelstrom controller Beta soon. As with all startups, even hedging bets with extra time, things seem to take longer than I’d ever believed. Though, I find that I’ve designed and developed rough versions of 4 products in a single year. Yes, 4. Maelstrom is only one of them.
Some of the work I’ve been doing had me looking at load cell designs and doing what is called a sensitivity analysis. In this case I was checking response to gage placement error. I was surprised by the impact. I had never explored it so deeply as this time. So surprised that it brought me around to powermeters—again! I can’t seem to escape these. What I found was that a single simulation could yield amazing results, but as soon as you simulate with the gage off center or turned on a little angle – Those 0.001% error results just turned into 6%. It is a lottery of if you buy the 0.0001% one, the 1% one or have terrible luck and get a 6% one.
One thing to understand is that these are variances with a load cell under design. I’ve talked about things in the past like pedal offset sensitivity, axial sensitivity, etc. These are different matters, but each impacts the other. As I’ve experienced, not all pieces of metal and carbon fiber are equal, manufacturing variances are just as much to blame as gage placement. It’s also why you can install a gage on one chuck of metal and it works great, but the next one it doesn’t. The inside shapes and bonds play a role in hollow structures just as much as the placement and orientation of a gage.
So this got me thinking and I started investigating the FCC filings of powermeters. Almost all have gotten simpler, a march to reduce the cost, and a few eschews the higher priced integrated components for discreet ones. ECK!
Now I’ll preface the next section, but it’s kind of the same as strain gages, these aren’t 100% rules. Experience plays to the perspective. While you absolutely can get good performance from discreet gain stages and cheaper SAR adc’s, be them discreet or integrated, generally speaking it’s like trying to install 20 strain gages. Sure you can do that, and it might save money, but you just added the positions and angles of every single one to the mix – or 40 potential problems if things go wrong. This is the same with replacing a fully integrated ADC with discreet components. My experience has been that I don’t believe it’s worth it. You trade of finding a random batch of 100 of something doesn’t work, weeks of engineering, to realize R26 is out of spec. I’d much rather spend 1 dollar a unit more (200 dollars), then deal with random 4 weeks of work caused by penny pinching. However, that’s just an opinion based on my experience. There are too many laser trimmed resistors and calibration tables that go into a modern good Sigma-Delta ADC. Heck the most expensive precision thin film resistors won’t even warranty them if you reflow them. They won’t even guarantee their thermal performance once installed really! But if you don’t care, and while 200 units might be bad you deal with it by replacing 3 under warranty of those who figure it out and claim it angrily, well it’s not the kind of thing I’d do.
So it got me thinking on the history and how trainers are now trying to put in cycling dynamics all the while everyone looking for more and more ways to reduce costs. It just left innovation on the floor back in the 2013 time. Every company has experimented with high speed data. I’m going to limit what high speed data is to more than 20 samples / rotation. So this discounts pioneers interesting, but flawed system (axial accuracy = HA, pedal offset). That leaves only Stages in a special BLE mode, which every single meter I’ve seen put in this mode has never ever come out. I saw many warranty returns due to this. In fact 100% of them I saw in 2015 put in this mode were returned.
So we’ve seen this price pressure, China’s started cloning Stages (with 4 individual gages that can have huge angle/position issues), and nobody ever got a widely adopted high speed sensor out. Sure there are some hilarious high cost and 2kg pedals for “Research”, but nobody has made anything accessible. And part of that might be what I’ve been learning about load cell gage placement. Unless you can over design it so it has redundancy built in and over calibrate it (way more data points than min) than you can’t know for certain that they will all be 0.5%. And that is the problem.
I don’t think any company could hit instantaneous measurement accuracy of their claims today — maybe Quarq. I mean, factually speaking we know anything on a R8000 or R9100 can’t met those specs, especially the right, but the ones that can do a full rotation and average out and be under 2% error, I don’t think any of them could be instantaneously 2% error for a pure force reading at a single sampled data point.
And price pressure might make someone break to try and gain profit from a lower end device spewing erroneous high speed data. Then what happens? Do brands try and rebuild the name of high speed, or will someone come out on the leading edge. Maybe whoever puts together the profile should also be the testing body?