Printalyzer Densitometer - A compact budget-friendly densitometer project

Since I've finally managed to write a sufficiently-detailed blog post on the project, I figured it was time to fork it off into its own discussion thread.



Here's the blog post:
http://hecgeek.blogspot.com/2021/07/the-printalyzer-densitometer-project.html
(More pictures, including ones of the latest prototype, are in the blog post.)

Here's where I've previously been discussing it:
https://www.photrio.com/forum/threa...mer-exposure-meter.180377/page-5#post-2409172

Here's where I'm posting the design artifacts (schematics, source code, 3D models, etc) for the project:
https://github.com/dkonigsberg/printalyzer-densitometer

For now I'm focusing on measuring in the "Visual" (B&W) spectrum. As much as I'd love to also do color, there are some technical limitations in getting accurate readings at the right wavelengths for that on a budget. (Though it'll continue to be a back-burner idea I may tinker with from time to time.)

Right now I'm thinking that this project may be something I attempt to bring to market a lot sooner than my main Printalyzer Timer/Meter project, because there are far fewer hurdles. Its also likely to be a good "test project" to get my feet wet with the whole process of actually doing something like that. I have no idea what my cost target will be yet, but I know what I'm competing against. (Heiland TRD-2 and used 80's/90's era devices on eBay) However, I may need to wait until the "great microchip shortage of 2021" starts to subside before I can build more than a handful of prototypes.

A very cool project! Thanks for the rather in-depth writeup. I am looking forward to seeing this as a product after the obscene lead times have passed

I figured it was time for another project update. So we'll begin with some quick photos of the latest prototype:



This prototype finally gets the whole hinge mechanism working correctly (with stronger springs), and can even detect when the device is fully-open or being closed on a target. Exactly how I use that information is still somewhat TBD, but right now its being used to turn the read lights on and off.

The buttons are still a bit on the tall side, but I'm going to be experimenting with some 3D printed alternatives to see if I can shrink them a bit. Most of my other options would require making the top cover thicker, which would have the undesired side effect of making the display appear more "sunken" into the device.

I'm also going to have a nice graphic overlay on the top cover, which will better protect the display and allow for nicer artwork. The first version of this (already ordered) is somewhat basic in its design, but I hope to eventually find a graphic designer to help make it better.

As far as the rest of the build, I'm going to likely end up doing a few iterations of commercial 3D printing options. What you see above was printed by me at home, but I need better for any sort of quantity. This is probably going to involve some trial-and-error, since everyone has long turnaround times and the material properties I care about aren't always listed on their websites (e.g. being completely opaque to IR).

I'm also still researching my options for the "stage plate" material (white concentric rings above). The stuff I'm using right now has a fairly glossy finish, and smudges easily. Meanwhile, what my X-Rite uses is a bit more matte. I'm limited to what I can easily order from Laser-cutting places (e.g. Ponoko), so that's going to take some trial-and-error as well.

Functionality-wise, reflection has been working well for a while. My results are generally consistent for the B&W paper I've tested, but handling colored targets (e.g. red or blue paper) is still a challenge. I'm probably never going to match the X-Rite units on those, but I'm still a lot closer than the Heiland TRD-2. Testing higher density (e.g. above D=2.05) targets is still proving to be a challenge as well, because calibrated reference materials don't go that high. The only thing I have that does go that high is the Dmax of an IT8 targer printed on color paper. I'm still not matching the X-Rite on this, but I might be matching or beating the Heiland.

For transmission measurements, I'm only just getting started with testing. I recently ordered a collection of materials from Stouffer to help fully test and compare transmission performance, and they should be arriving soon. The software for it is all written, and works in theory, but I still need to actually compare side-by-side graphs of step tablet measurements to be sure.

The software is coming together pretty well. Its already almost feature complete, as far as the important features go. There's still a lot of polish, of course, and I still need to come up with a way to make it easier for end-users to update the software. (I may switch to a microcontroller one more notch up the product line to help with some of this, pending availability.)

As always, I'm now trying to track the majority of my outstanding issues in the GitHub issue tracker:
https://github.com/dkonigsberg/printalyzer-densitometer/issues

The great chip shortage of 2020 is still a hurdle to making anything beyond a few prototypes, but if I'm really lucky I might be able to pull off a batch of evaluation units before 2021. (More than one component on this thing is affected by the shortage.)

Among your notes is the recommendation for anyone choosing to make and sell to buy non-calibrated scales. Stouffer calibrated scales are traceable to NIST so I would recommend calibrated to anyone making for sale to consider traceability to NIST in their decision.

Bill Burk said:

Among your notes is the recommendation for anyone choosing to make and sell to buy non-calibrated scales. Stouffer calibrated scales are traceable to NIST so I would recommend calibrated to anyone making for sale to consider traceability to NIST in their decision.

Click to expand...

What I was trying to say is this:
- If you're only building one or two of these, then go ahead and just buy the "calibrated" Stouffer materials.
- If you're building many of these, then you can save a lot of money by having your own properly calibrated reference densitometer, buying the "uncalibrated" Stouffer materials, and doing the calibration measurements yourself.

I like the tall buttons with their clear picture directions. Very 80's. Those were good years for design. From the 20's to the late 80's in America, people knew how to design things. One shouldn't have to read a manual to work something. Today, things are designed so poorly, especially industrial designs, that they even have to put PUSH and PULL on doors because they failed to design them in a manner where it would be obvious at first glance what you need to do. And it doesn't work, invariably someone will do just the opposite because it's not clear.

momus said:

I like the tall buttons with their clear picture directions. Very 80's. Those were good years for design. From the 20's to the late 80's in America, people knew how to design things. One shouldn't have to read a manual to work something.

Click to expand...

Actually, many complex electronic devices from the 80's are nearly impossible to figure out how to use without reading the manual. Or at least impossible to figure out how to configure, even if the basic functions are obvious. That's because such devices tend not to have a flexible enough display, or enough program memory, to actually provide useful on-device instructions.

If I can make this device easier to figure out than my X-Rite 810, I see it as a win. And honestly, that's not a very difficult thing to achieve. Simply being able to switch the display between "big clear numbers" and "multiple lines of text" depending on the situation, can go a long way towards achieving that.

Thanks to today's UPS delivery, I now have a complete collection of white(ish) stock from Ponoko to evaluate for the stage plate construction:


I'm currently using the one called "Opal Acrylic" (upper-middle) but its a bit too shiny on top. (Functionally this is probably fine, but I don't want it picking up smudges or fingerprints.
While I'll give these a full round of measurement on my X-Rite densitometer before making a final decision, I'm currently leaning towards switching to the "White (Matte) Acrylic" (upper-right) one for the next revision. Its slightly less transmissive, but probably still okay, and will give a better surface finish.

I also now have more Stouffer calibration references and wedge/tablet scales, so I should be able to to a complete workup of reflection and transmission performance soon.

So I spent some time today doing a full calibrated testing workup of my Printalyzer Densitometer (in its current state) versus my X-Rite 810 and my Heiland TRD-2.
To do these tests, I used various calibrated reflection and transmission materials from Stouffer:

• R550C 5-step reflection tablet (calibration)
  
• R2110C 21-step reflection tablet (comparison measurements)
  
• T5100C 5-step transmission wedge (calibration)
• T2120C 21-step transmission wedge (comparison measurements)

(I also have some X-Rite calibration materials, which were involved in some of this, which I mention again below.)

I made sure my X-Rite and Printalyzer Densitometer were calibrated the same. The Heiland is impossible for the end-user to properly calibrate, so I just zero'd it on whatever seemed most appropriate and offset any affected measurements accordingly.

Including graphs as thumbnails to avoid overwhelming the post:

(The "Limit" marked on the error graphs is based on the accuracy figures in the specifications for my X-Rite 810.)

From these tests, I've gathered the following takeaways:

• I'm pretty darn close on Reflection, maybe even good enough
• I'm fairly close on Transmission, but there is room for improvement. (The mechanics of keeping the device aligned need work, and the measurement/calculation routines could probably use some improvements as well.)
  
• The X-Rite Reflection Calibration plaque matches up with the Stouffer reflection tablets well enough that either can be used for calibration with comparable results.
• The X-Rite Transmission Calibration wedge (acquired via Acurad) I have appears to have been calibrated to a completely different standard than whatever Stouffer uses. I'm not sure which one is correct, but I can't interchange them. As long as I'm measuring Stouffer-calibrated step wedges, I have to calibrate the densitometer with a Stouffer-calibrated reference.
• The Heiland TRD-2 is not very good. I think its biggest failing is that there's no way for the end-user to properly recalibrate it, but it has other issues too. Regardless, I think the only "competition" actually worth comparing accuracy against is the X-Rite.

Yes, I completely agree with that point of view. And I agree with this article

Congratulations! And best wishes for success. A minor reservation: I looked up your blog and saw that you had settled upon the TSL25911 sensor. Briefly summarizing for the convenience of other readers, that sensor includes a visual sensor that also has some undesired IR sensitivity, and an IR-only sensor; the idea being to correct the reading of the first one by subtracting the IR-only output of the second one.
Generally, subtraction is not a good idea when aiming for high accuracy.
Specifically, the shape of the spectral response of the IR sensor does not match the IR part of the Vis+IR sensor, so the subtraction cannot be accurate, and a scaling factor (for the correction) designed for one parasitic IR illuminant will be fooled by a different illuminant. Unless, of course, you rely on the illuminant (a green LED?) to provide the spectral selectivity, in which case the IR correction becomes pointless.
Why not use a filtered sensor with an intrinsically visual (550nm more or less)? Like (among others) Vishay VEMD5510C or Osram BPW 21. Sure, you have to implement a zero-bias current-voltage amplifier and perform the A/D conversion, but that should not be a major obstacle for you.

bernard_L said:

Sure, you have to implement a zero-bias current-voltage amplifier and perform the A/D conversion, but that should not be a major obstacle for you.

Click to expand...

Don't underestimate this. In the other/previous thread @dkonigs and me briefly exchanged views on this and @dkonigs made it quite clear (at least to me) that his primary interest and competence is in the digital domain. Re-engineering an analog sensor front-end is a different area of expertise - it's certainly doable, but I wouldn't recommend it unless there is a compelling reason for it. As it is, I don't think such a reason exists in the context of this particular device. Please note that both photodiodes you linked to have significant IR responsiveness, so wouldn't make matters any better. I think your suggestion of filtering out IR hardware-wise, either by means of a filter or by means of selecting a light source with no significant IR component is much more sensible. The latter is more or less guaranteed when an LED light source is used in the first place, since LEDs tend to emit a negligible amount of IR unless the semiconductor used is specifically doped for it. All considered, I think you're offering a not entirely appropriate/effective solution for a problem that isn't convincingly there to begin with....
Just my 2 cts of course.

So there's something I think I need to state as clearly as possible, because its been the source of much confusion:

The Photopic Spectrum is NOT the same as the ISO 5-3 "Visual" spectrum.

Everyone seems to assume they're one and the same when making recommendations, even if I state the differences upfront (had some particularly frustrating exchanges in private message threads of this nature).

The Photopic Luminosity Function is centered at 555nm.
The ISO 5-3 Visual spectrum is centered at 570nm.

Allow me to illustrate with a chart:


Yes, I have a copy of all the ISO 5 standards, and yes I've read them.

If I could get a filter that will give me the proper sensitivity for the ISO 5-3 spectrum, I'd gladly use it. However, I'm not aware of any such thing as an off-the-shelf product nor how to actually get such a thing. (a 570nm bandpass filter, which I can get, is too narrow to do the job)

Regardless, in practical terms, the real test is how good are my numbers when I'm measuring a very blue or very red target. (Not a typical use case, but still something you can test.) The answer right now is: Not as good as the X-Rite, but a lot better than the Heiland.

koraks said:

All considered, I think you're offering a not entirely appropriate/effective solution for a problem that isn't convincingly there to begin with....

Click to expand...

I did spell out an alternate solution (narrow-band LED illuminant), not claiming the main proposal was unique.

Please note that both photodiodes you linked to have significant IR responsiveness, so wouldn't make matters any better

Click to expand...

The uncorrected "visual" chanel of the TSL2591 has equal response at 800nm as at 550nm. In the case of the Vishay photodiode, it's 5% (factor 20 down).

A reason for building from a discrete analogue detector is design flexibility: larger area -> more nA/lm, subject to how much spatial resolution designer wants; he can manage that compromise himself. From the datasheet of the TSL2591 (Fig.3, scaling to pin spacing 0.65mm), the detector area is ~0.4mm2, presumably shared between vis and IR channels. For the Vishay VEMD5510C, 7.5mm2, almost 20x larger area, still small enough (2.7mm on a side) and in the ballpark of the measuring spot of commercial densitometers.

dkonigs said:

So there's something I think I need to state as clearly as possible, because its been the source of much confusion:
The Photopic Spectrum is NOT the same as the ISO 5-3 "Visual" spectrum.

Click to expand...

Sure they are different, but I fail to understand why that matters in the context of your stated goal:

For now I'm focusing on measuring in the "Visual" (B&W) spectrum. As much as I'd love to also do color, there are some technical limitations in getting accurate readings at the right wavelengths for that on a budget.

Click to expand...

Do I understand correctly this means measuring B/W negatives and B/W prints? (Rather than the V channel of a color material, absent the R and b channels)
If so,

• either you are measuring silver negatives, and the difference in density between 550nm and 570nm is not practically relevant.
  
• or you are measuring something like a negative developed in, e.g. Pyrocat; an end-user printing on graded B/W, paper will need a B density rather than V, and someone printing on multigrade will need both...
  
• ditto reflection density of papers: what to do with toned papers (short of a spectrophometer, but I'm digressing)

Bottom line: possibly by insisting on a particular response among possible V-ish curves, you are raising unnecessary constraints. Or I missed something.

@bernard_L - I know about the design flexibility consideration. It's why I went the all-DIY route for my project, which is tangentially related to the project discussed here. From that experience, I also know first-hand the arguments NOT to go for a full-DIY route. I don't think it's at all necessary here, and then it will become a burden to the project.

Also, because its a neat picture, and for reference, here's a cutaway view of what the sensor head on this thing actually looks like:



As far as changing sensors or adding complexity (and cost) to the whole sensor mechanism... The only reason to do that is if it actually gives me better data than what I'm getting right now, with an acceptable cost/benefit. If I define my requirements as "ISO 5-3:2009 Visual" (as the B&W side of X-Rite hardware does), then I'm already doing pretty well. I'm not even sure I can get much better without a major increase in cost/feasibility.

If I want to also cover the ISO 5-3:2009 Status A/M ranges, there seem to be 2 available approaches:

1. Have a separate sensor for each "channel", tuned to the specific sensitivity of that channel (what X-Rite does)
2. Figure out how to correctly interpolate the data from a multi-spectral digital sensor (e.g. AS7341 or AS7265x) to get data that's functionally close enough to the first approach.

Some of these may have geometric considerations that may make it impossible to use the same sensor for both reflection and transmission.

Its probably time for another project update, so I'll start it off with a photo:


The top two are the initial prototypes have the "Rev A" circuit board, and the bottom one has the brand-new "Rev B" circuit board. The differences are minor, and mostly have to do with switching to a slightly more capable microcontroller (more memory, easier to update) and fixing some nitpicks from the first version. All the important components are unchanged. I've also been continuing to refine some elements of the enclosure, but that's also hard to see in the photo.

These prototypes also now have a nice "graphic overlay" with printing on top. This is currently more of a proof-of-concept to make sure the mechanical aspects are right, as I absolutely plan to change the colors and some of the artwork.

The bulk of my current project work in the past few weeks has been centered around evaluating commercial 3D printing technologies for the enclosure. This is a summary of what I've discovered:

• Fused Deposition Modeling (FDM) - This is what your home 3D printer does. Its great for its low-cost, and does a decent job with the internal components. Quality is generally good as long as your print doesn't need supports. The moment your design does need supports, there are a lot of tradeoffs to consider and its no longer the best option.
• Multi-Jet Fusion (MJF) - One of the more popular commercial options. Doesn't come in a variety of colors, but produces a nice result. Biggest problem, however, is that the "dimensional accuracy" of a print seems unusably bad. On paper its excellent, but all the prints I've ordered (from companies that I don't have a personal relationship with) are off by just enough that screw holes don't line up and parts don't fit together.
• Selective Laser Sintering (SLS) - Quality is good. Dimensional accuracy is good. Comes in a variety of colors. I've even found low-cost options for getting it done. Biggest problem is that the material isn't opaque enough for use around the critical parts of the sensor head.

So I think I've settled on what path I'm going to take. I'm probably going to have the bulk of the enclosure itself printed on an SLS machine, then use an FDM machine to make the light path components (sensor cone, etc) and some internal parts. This may end up with a cool two-tone design, with a blue body and a black cone. I'm probably going to make the first full-up version of this in the next few weeks. (Oh, and I also found a matte black PLA that I really like from 3DXTECH, so that's what all the FDM/black parts will probably be made from.)

The other thing I'm working on, now that the "Rev B" circuitry has been constructed, is getting back to work on the actual software development. First I'm making some architectural changes enabled by having more memory. Second, there's a lot of details I want to revisit about how I run the sensor and even how I calibrate/characterize the sensor. The hope is that this will make everything smoother and more accurate, once its all worked through. Probably not going to go into all the details of this just yet.

Regardless, at this point the whole project is just as much "developing a product" as it is "make a gadget that gives good numbers." As such, I've also been thinking through a variety of other issues from cost tracking through business concerns. Hopefully I'll have something more public-facing to share along those lines in the next few months.

I used to operate HP MJF machines among others, found dimensional accuracy fine to spec - sounds like poor build packing or fast cooling problems. The machines are good for doing entire build runs of a single part etc where these kind of issues are minimized or controlled. Once you have lots of different parts in a build together you start to get problems, warping and ripples due to poor part placement and orientation where there is an uneven build up of heat. If these are commercial printing companies running multiple builds 24/7 good chance they are just using basic auto-stacking for maximum throughput, possibly even fast cooling them. The parts can be dyed (another failure point) or HP has CMYK now but due due to the color process parts are inherently weaker (more for display).

Perhaps try find a university or research institution that has one of these machines for in-house prototyping for more controlled output. Another interesting feature of MJF is assemblies, may not work for your particular design but something to think about if wanting to reduce costs?

calebarchie said:

I used to operate HP MJF machines among others, found dimensional accuracy fine to spec - sounds like poor build packing or fast cooling problems.

Click to expand...

Yeah, I'm convinced it has to be something like this. For any given process, there are really two sets of specs worth considering:

1. How good is it when the operator gives a damn about you and your specific project
   
2. How good is it when the operator doesn't give a damn about you and your specific project

From everything I've read, MJF should be perfectly fine (and perhaps even better than SLS) in case #1. The problem is that it seems to be much worse than SLS in case #2. And if I'm not running my own machine, and don't want to pay a fortune for the prints, I probably have to live with case #2.

What makes matters worse here, is that the companies that provide these services can hide behind published specs/tolerances that let them think its okay to give you sloppy results. Shapeways, for example, thinks its perfectly okay for an MJF print to be +/- 0.7mm in any dimension. This is enough for screw holes to not line up, circular features to look oval'ish, and two halves of an enclosure to have an overhang >1mm when mated together.
(And Shapeways actually had better results than the other faster/cheaper place I tried in my research.)

calebarchie said:

Perhaps try find a university or research institution that has one of these machines for in-house prototyping for more controlled output. Another interesting feature of MJF is assemblies, may not work for your particular design but something to think about if wanting to reduce costs?

Click to expand...

Assemblies could be tempting as a way of making my hinge, except for two problems: First, I need to be able to fit a spring inside there. Second, design iteration is extremely frustrating when you have to wait 2 weeks between each "test print" and tolerance issues could break the whole concept.

Cost is also an extremely important factor here. Not so much for prototyping, but absolutely once I move on to small/medium quantity production. A place that charges prototyping rates and doesn't offer real quantity discounts is going to make the end result way too expensive. Meanwhile, I've found SLS options that give me enough of a quantity discount that the enclosure actually doesn't drive device cost through the roof.

A lot of these things you simply have to deal with through the prototyping stage but once the design is finalized I do not think it would be too difficult to find an AM willing to work with you. 3D Hubs can link you to a local print house to get a quote on entire builds just depends on the scale of production you want. To give you an idea the most popular MJF machines build volume is 380x284x380mm so that's quite a lot of small parts, one build could last you a long while.

For something like this, multiple sources of parts (just for enclosures etc) may be a logistical headache eventually I think something like that would only work if you were making the bulk of it yourself.

calebarchie said:

3D Hubs can link you to a local print house to get a quote on entire builds just depends on the scale of production you want.

Click to expand...

Thanks for the tip of that other broker. So far I've used Shapeways (a print house) and Craftcloud (a broker), so I'll give MJF one more try via whoever 3D Hubs decides to send my order to. They claim MJF and SLS have the same dimensional accuracy, so we'll see if reality matches up. They also have prices similar to what I was getting via Craftcloud, so quantity will be acceptable through them.

My only concern with 3D Hubs is that they seem to keep the print house anonymous at the quote/order stage, and there comes a point where I care about predictability.

Do I understand this is powered by and transmits data over usb-c? Cool.

Bill Burk said:

Do I understand this is powered by and transmits data over usb-c? Cool.

Click to expand...

Technically its just micro-USB, but otherwise that's correct.
(Not yet sure if it makes sense to change connectors, as with a device like this the only real difference between the two would be which cables you can use with it.)