trade you for this;-)

under 'benefits and features;

Tiny 5.6mm x 3.3mm x 1.55mm 14-Pin Optical Module
• Integrated Cover Glass for Optimal, Robust
Performance

https://www.analog.com/media/en/technic ... X30101.pdf

your reference points at what I think may be a different sensor;
Products
MAX86160
Integrated Heart-Rate Sensor for In-Ear Applications

MAXM86161
Single-Supply Integrated Optical Module for HR and SpO2 Measurement

I may however be completely off the mark

That's the same data sheet I linked.

Read page 29:



And the application note refers to OLGA packages in general.

Of course they are discussing manufacture of a PCB. But the fact that it is unsealed with a potential for dust ingress and that one should keep those vents open doesn't change after the chip is soldered onto the PCB.

you are right, from what I read it means that using a resin may be off...and that a gasket of some sort probably is required to seat the sensor, I do not read anything that makes the reference to the cover glass invalid..why would the cover glass have any advantage if the sensor still requires another glass cover?

The glass doesn't form an airtight seal but it's still protection for the sensor.

What's the disadvantage of having a second piece of glass?

jpender wrote: But don't you also see outliers? I found that I couldn't rely on a single reading because of that.

I resampled a number of coffees at different grind sizes and read each sample 8 times. For each reading I'd WDT the sample, apply, tamp. Then remove the sample, WDT again, apply, tamp, etc.. I did this as part of variance analysis testing to determine the optimal grind size. At 300-500 microns I got the least variation in grind size, and at that grind size I only saw 1-2 Agtron difference between readings of the same sample. YMMV.

jpender wrote: I think the idea was that you'd essentially be taking readings of multiple samples by adjusting the sensor to look at different places on the surface. But I agree with you that the original design has merit.

I can't find it, but Craig (baldheadracing) did a study on HB with the Roast Vision when it first came out where he took LOTS of samples to determine statistical significance. I guess the question is whether reading one 18gr sample from a roast X times is more statistically valid than reading X randomly selected 3gr samples from a roast one time each? Intuitively I'd think the latter is more valid, but I'm not a statistician.

With this new case, assuming the puck is prepped and tamped perfectly so the grinds are uniformly distributed, bed level, etc, I'd think you'd get pretty much the same reading anywhere on the puck assuming the sensor distance can be maintained. But all of this is just conjecture on my part. Maybe others know more. Hopefully Transparent Roaster will share more on this.

jpender wrote:The glass doesn't form an airtight seal but it's still protection for the sensor.

What's the disadvantage of having a second piece of glass?

No real big disadvantage, just that adding it can be a pain...what is not needed can be left out, and all that.
Not having a 3D printer and no stamina to try come up with a 3D model it'll stay theory anyway

I have started to learn to use OpenSCAD (I can't run Fusion360 on my computer). But for what I'm doing at the moment I'm not planning on using printed plastic. I've got off the shelf parts that I'm hand cutting/sanding/drilling. If what I'm doing works well then it can be made into a 3D model by somebody, if not by me.

This little chip isn't meant for what we're using it for. It's wrong in many respects. I have thought about using discrete parts (LEDs and photodiodes) instead. Maybe I could get that to work but nobody would use the design. The beauty of the DIY meter is that you click with your mouse to order a cheap 3D box and a few off-the-shelf boards and then more or less snap everything together like LEGOs.

I agree, when I first saw the design and looked up the parts I thought someone was having us on...an SaO2 probe, WTF...

Brewzologist wrote:I can't find it, but Craig (baldheadracing) did a study on HB with the Roast Vision when it first came out where he took LOTS of samples to determine statistical significance. I guess the question is whether reading one 18gr sample from a roast X times is more statistically valid than reading X randomly selected 3gr samples from a roast one time each? Intuitively I'd think the latter is more valid, but I'm not a statistician.

I think the variability is due to more than one thing. The coffee particles are heterogeneous, will not be perfectly mixed, and the surface presented to the sensor will vary with user prep.

The coffee color itself in a small sample may be lighter/darker than the coffee on average. Think about measuring a 30%/70% blend. With 7 beans what are the odds that you'll get exactly 30% of the first bean type? Even with an SO not all the beans are identical. You can minimize this source of error by looking at more coffee. Look at how big a dish the Agtron machines use. If the sensor you have can only look at a tiny area at a time what's the best way to cover that ground? That's the idea behind a device that lets you move the "eye" more quickly.

Brewzologist wrote:With this new case, assuming the puck is prepped and tamped perfectly so the grinds are uniformly distributed, bed level, etc, I'd think you'd get pretty much the same reading anywhere on the puck assuming the sensor distance can be maintained.

If every reading were the same then there would be no need to sample more than once.

I tested with the sensor 'naked' on the prepped puck in several places, and my DIY experiment showed variability around 1 'old' "Agtron' 'unit' (old because of the old software version, and 'Agtron' because I'm not convinced we're even close to that 'unit' which is no unit)

Obviously a sensor fixed in one place cannot reach every location in a puck as it is limited to one circle, however given that the sensor is a mere square mm or so and the radius of a 58mm basket is many times larger the uniformity of the coffee and the homogenity of it's blending (if appliccable) and grinding are likely more important but that may well be the same for any measurement device. Sample size and how sampling is done is probably important.