Almost three years ago I designed and fabricated some pill/caplet dispensers for the half-dozen or so prescription meds I have managed to accumulate over the last decade or so. A while ago, one of my prescriptions changed its tablet to a much smaller size, so I decided to update my design while fabricating a replacement dispenser.

Between the last project and this one I’ve been playing with OnShape, a web-based 3D CAD package, so I thought I would use it to see if I could do better than last time. I really like OnShape because it uses a 2D ‘sketch’ based design philosophy, which makes tweaks and/or modifications much easier – change a few 2D sketches, and the entire design changes along with it.

The previous design implemented a smooth collar that was a press-fit for the pill bottle cap which turned out to be kind of clunky. This time I thought I might try implementing internal threads on the collar so instead of a press fit it would simply screw on like the original cap, and I discovered that a ‘ThreadCreator’ extension existed for OnShape – neat!

So, I worked my way through the process, and came up with the following design, available to anyone with a free OnShape account

This design has internal threads for a 37mm cap with the standard 4.7mm thread pitch, so it will screw directly onto the pill bottle, ‘eliminating the middleman’. Here are some photos of the finished product:

And here is a short video showing the dispenser in action:

10 June 2024 Update:

Last night I attempted run two more prints of this model, as I have two additional pill bottles of the same diameter with older pill dispensers, but the prints failed catastrophically – bummer! I rounded up the usual suspects (bed temp, model arrangement, Z-axis tuning, etc, and finally managed to get another print going, at least through the raft and first few layers. After bitching and moaning about this for a while, it occurred to me that if I had documented the layout and settings more aggressively from the first print, I wouldn’t have wasted all those hours last night and today. So, once I’m sure I have a consistent print configuration, I will document it here.

I got a good print with the following settings:

• Flashforge Creator PRO II ID
• Left Extruder – Red PETG, 240C, 80C Bed
• Right Extruder AquaSys120 240C, 80C Bed
• 2-layer raft using support filament (AquaSys120)

See the following images for the full setup:

After a 3-hour side-trip into the guts of the Flashforge to clear an extruder jam, I was able to get the second print underway. As I write this it is about 6% finished, but all the way through all the support material parts (so it should finish OK).

17 June 2024 Update:

Not so fast! I realized that the threaded portion of the dispenser cap, while functional, was very poorly printed due to the lack of supports (and, as I found out later, also due to the resolution setting). In addition, the side walls of the V7 slide box were too thin and broke apart easily. After modifying the design, I attempted another print using the above settings, but the PVA dissolvable filament simply refused to stick to the print bed – arrrrrgggggghhhhh!

After going through the whole extruder & bed temperature search routine again yesterday, including replacing the heated bet PEI layer and even putting down blue painter’s tape with no success, I was perusing google-space for clues and kept running across reports where dehumidifying the PVA filament worked. I didn’t see how that would help me, as we control the relative humidity in our house to about 50% +/-, but hey, what did I have to lose at this point?

So, before going to bed I dug out my filament dehumidifier rig and left my filament in it overnight (and until about noon the next day for a little over 12 hours). Then I tried some prints and although not successful at first, the results were encouraging. I finally got two really good prints with the following setup:

• Slicer resolution: ‘0.15m OPTIMAL’ setting in Prusa Slicer
• Right (PVA) extruder: 220C
• Left (PETG) extruder: 240C (this was constant throughout)
• Bed: 40C
• Layer of blue painter’s tape on top of the PEI substrate

So, I think the big takeaway from this episode is: PVA must be explicitly dehumidified BEFORE each print session. Otherwise the PVA will not stick to the print bed, no matter what you do.

Stay tuned,

Frank

07 July 2024 Update:

After getting the threaded pill bottle dispenser cap working, I decided to try my luck with my two 57mm twist-lock pill bottles. The twist-lock cap geometry was considerably harder to design. Rather than trying to design and print everything as one piece, I decided to separate the dispenser piece from the cap mating piece, as shown below:

Posted 05 May 2024

My daily carry pistol is the M&P Bodyguard 380 in a ‘Sticky’ brand Holster, as shown below:

I carry this in my jeans front pocket, and it works great. I regularly practice smoothly drawing the pistol, activating the Crimson Trace laser, and getting the gun on target. Unfortunately after a year or so of use I started seeing occurrences where the laser wouldn’t activate, and investigation showed that the laser battery was dead. The first time this happened I just wrote it off to the normal battery life, but the second and third times were definitely too close together to be a battery life issue. I finally figured out that the laser was being inadvertently activated in the holster – clearly not a good solution. The good news is, it made me more determined than ever to not count on the laser. Now I practice with and without the laser (although I much prefer the ‘with’ scenario).

Thinking about the problem, I inferred that the ‘Sticky’ holster, when new, comparatively stiffer when new than after hundreds of cycles of inserting and removing it from my jeans pocket, and of course hundreds of dry-fire draw and shoot repetitions. Eventually the holster gets pliable enough so the normal inward pressure from my jeans pocket is enough to activate the laser at some point (and once is enough, as once activated it will probably stay ON until battery exhaustion). As you might imagine, replacing the battery in the laser module is a major PITA, as the pistol itself must be disassembled, and then the laser module removed to access the battery. Then the procedure must be run in reverse to re-assemble everything, and now the laser alignment must be checked and adjusted as necessary (another major PITA).

Thinking about solutions, I contemplated 3D-printing a holster insert that would replace the original holster stiffness (and I might still do this). However, I was struck by the idea that the real solution to the holster material pressing in on the laser activation button is to remove the holster material around the button; then the holster material thickness becomes an additional guard around the button. Instead of being the culprit, it now becomes the solution – cool! Here’s another photo showing the ‘Sticky’ holster with a (unfortunately crude) hold around the pistol’s laser button:

‘Sticky’ holster with material over laser activation button removed

After – once again – replacing the laser battery, I plan to run with this setup for a while and see how long the laser batteries last this time.

17 July 2024 Upate:

From May of this year until now I had no problem with laser battery life, but yesterday I found the batteries low/dead again. This indicated I was still getting inadvertent laser activation even with the laser button cutouts shown above. So, I decided to see if I could improve on the design a bit.

I went into OnShape, my 3D design tool of choice, and designed a hollow ‘holster bump’ as shown in the screenshot below:

The idea, as shown below, is to protect the laser activation button on each side of the gun, so it can’t be activated when in the holster

I then used hot glue to affix them temporarily to the gun to confirm their positions coincided with the holes I had previously mad in the holster. After this, I glued the bumps into the holes with superglue. We’ll see how this works.

Stay tuned,

Frank

Posted 05 May 2024

I and my wife are crossword puzzle addicts. To feed our habits, I signed up for the NY Times crossword puzzle archives and downloaded the Friday, Saturday and Sunday puzzles (the Mon-Thurs puzzles were too easy) for each week between January 2015 and December 2022.

Originally I would print out a puzzle as required by opening/printing the puzzle file using Across Lite, but this got old in a hurry. So, I decided I would create a program to automagically print an entire folder’s worth of puzzles in ‘batch mode’ utilizing two-sided printing – yay! Looking around for the best/easiest way to accomplish this, I ran across an application called ‘AutoIt’, specifically created as a shell-script generator to run Windows (or Mac) applications and system functions.

It took a while to work my way through the command reference and examples, but eventually I was able to create an AutoIt shell script to do what I wanted; The script prompts the user for a directory containing Across Lite *.puz files, and offers to print them all in equal N-puzzle batches, thereby allowing the user to print a batch and then move the printed puzzles from the output tray to the input tray for a double-sided result.

This worked great, with the only downside being that the user’s PC cannot be used for anything else while the script is running, as it grabs the mouse cursor to launch Across Lite and print the current .puz file.

Here’s the script, as it stands now in May of 2024 (saved on my system as C:\Users\Frank\Documents\Personal\Crosswords\Print Pending\240504 PUZ Print Script.au3):

And here is the output log for printing the contents of the ‘C:\Users\Frank\Documents\Personal\Crosswords\Print Pending\2015’ folder containing 156 puzzle files, printed in four batches of 39 files each:

After some time away from my autonomous wall-following robot, I have started spending time with it again. The first thing that happened was I tried a long-term run in my home, only to find that the ‘mirrored-surface’ feature I added some time ago caused the robot to enter an infinite loop, even when encountering a non-mirrored surface – oops! This eventuality was such a bummer that I stopped working with the robot for several months.

When I worked up the courage to address the problem, the first thing I did was to back out the ‘mirrored-surface’ code, reverting back to the state of affairs represented by the ‘WallE3_QuickSort_V5.ino’ Arduino project. This required quite a bit more work than I had anticipated; I had thought my process of incremental builds would shield me from that – NOT! Eventually I was able to use some ‘diff’ tools to work my way through the process.

After getting the code squared away without the ‘mirrored surface’ code, I decided to take my robot out for a walk – well actually it was the robot taking me for a walk ‘in’ for a walk around the house. Here’s a link to the (not-so-short) video showing the action. I have included this as a link to the video file on my Google Drive site, because it’s too long to fit on my WordPress blog page.

Posted 11 March 2023,

I have been wearing hearing aids for quite some time, compliments of a lifetime around airplanes, long before hearing protection became a thing. I recently got a set of Jabra ‘Enhance 200’ aids via direct order, and I like them very much, EXCEPT I have noticed that my perceived hearing acuity seems to vary quite distinctly over a period of a minute or two. I first noticed this when I would turn on a water tap while washing dishes or preparing to take a shower. I would turn on the tap, and then 20-30 sec later, the perceived sound of the water coming out of the tap would increase significantly – even though the water flow rate had not changed. Later, in a social setting (bridge club), I will experience significantly lower and higher perceived speech volumes – very frustrating! I hypothesized that the aids employ an AGC (Automatic Gain Control) of some sort that is getting triggered by an initial loud noise which reduces the gain (and my perceived noise/speech level), and then 10-50sec later the gain would go back up again.

Just recently though, another thought popped into my head – what if the perceived volume changes aren’t due to some property of the hearing aids, but instead are a physiological feature of my current hearing/understanding processes? Hmm, I know I have some issues with my Eustachian tubes blocking and unblocking, and I also know that on occasion the volume changes are correlated with a ‘blocked Eustachian tube’ feeling, so this isn’t a completely crazy hypothesis.

So, how to distinguish ‘meat space’ audio response from hearing aid responses? I decided I could set up an experiment where I could expose one or both of my hearing aids to a volume ‘step function’ and monitor the output for AGC-like responses (an initial rapid drop in output, followed by an eventual return to normal). Something like the following block diagram:

The idea is that the Teensy would produce an audio output in the human audible range that can be controlled for amplitude and duration. The audio would be presented to the hearing aid, and the amplified output of the hearing aid would then be captured with an external microphone connected back to the Teensy. A plot of captured amplitude vs time, with a ‘step function’ input should indicate if the hearing aid is employing an AGC-like response function.

After some fumbling around and searching through the posts on the Teensy forum, I ran across this post describing how to create a simple 440Hz sinewave output from the DAC pin. The original poster was having problems, but after Paul Stoffregen added the ‘magic sauce’ (adding the ‘AudioMemory(10);’) line, everything worked fine. When I copy/pasted the posters code and added the line, I got the nice 440Hz waveform shown below – yay!!

The next step is to hook the DAC output to a small speaker, so I can drive the hearing aid. When I tried hooking a speaker directly to the DAC output, it clipped badly – oops! Fortunately, I remembered that long ago I had purchased a ‘prop shield’ for the Teensy LC/3.1/3.2 controllers, and this contains a 2W audio amp whose input connects directly to the DAC output – nice! So I dug around and found the part, soldered on headers, and plugged it on top of the Teensy.

15 March 2024 Update:

So, yesterday Space X launched IFT3 (Integrated Flight Test #3), consisting of ‘Super-Heavy Booster 10’ and ‘Ship 28’. The combination was the largest, most powerful rocket ever launched, by a fair margin. The upper stage (Starship 28) flew from Boca Chica Texas to the Indian Ocean near Australia in about 49 minutes – wow! At that point, the upper stage was the largest object ever launched into space in all history – Wow Again!

OK, back to reality. I managed to get the prop shield working – after the usual number of false-starts and errors. One ‘gotcha’ was that I hadn’t realized the 2W audio amp was a Class-C type, which means it switches on and off at a very high frequency (100KHz or so) – way above human hearing range, and the audio AM modulates that signal. When it is connected to a speaker or other audio transducer, it acts like a low-pass filter and all that comes out is the audio; this is a really neat trick, but it means that the audio amp output signal is basically impossible to look at directly with a scope – oops. So, I got a pair of MEMS microphone breakout boards from Sparkfun and used the microphone to turn the speaker audio back into an electrical signal that I could view with the scope. Here’s the setup:

This worked great, and I was able to verify that the speaker audio output was a reasonable replica of what the T3.2 DAC was putting out.

The next step was to feed the MEMS mic output back into a Teensy 3.2 analog input so I could measure the signal amplitude (A4 in the above block diagram). Then I would modify the Teensy program to deliver a 1-2 sec ‘pulse’ of high amplitude audio to the hearing aid, followed by a constant low-amplitude signal. The measured amplitude of the hearing aid output (as received by the MEMS mic) would be monitored to see if the hearing aid exhibited AGC-like behavior.

However, as I started setting this up, I realized I would have to solder yet another flying lead to the top of the prop shield, as the Teensy 3.2 pins were no longer accessible directly. So I decided to fix this problem by adding female headers to the top side of the prop shield to allow access to all Teensy pins. The result is shown in the ‘before/after’ photos below:

19 March 2024 Update:

I soon discovered that my plan for routing the MEMS mic back to a Teensy analog input and just averaging the results over time wasn’t going to work, as a glance at the MEMS output to the Teensy (shown below) would make quite obvious:

The average value for this signal is just the DC offset, which will always be the same. The only thing that varies is the amplitude – not the average value – oops!

OK, so the obvious work-around to this problem would be to put a half-wave or full-wave rectifier circuit between the MEMS output and the Teensy so the analog input could measure the half- or full-wave amplitude instead of the average. But, I really didn’t want to add any more circuitry, and besides I have this entire 72MHz computer at my beck and call – surely I can get it to emulate a half- or full-wave rectifier?

So, after the usual number of screwups, I got this working reasonably well – at least enough for a ‘proof-of-concept. The basic idea is to take analog input readings as fast as possible, and use the resulting values to compute the average value (in A/D units – not voltage), and then take the absolute value of the difference between each measurement and the average value – this essentially implements a full wave rectifier circuit in software.

The following data and Excel plot shows the results for the waveform shown above:

The above data was collected by sampling the input at about 20KHz (50 Usec). As can be seen from the above, the average value is a constant 511.64 (out of a zero to 1023 scale), and the actual measured values varied from about 264 to about 755. Here are Excel plots of both the measured input and the calculated amplitude:

So it looks like this idea will work. For the intended application (determining if my hearing aids exhibit AGC-like behavior, I can perform a running average of the full-wave rectified signal using something like a 0.1 Sec interval (2000 samples). That should accurately capture the onset and release of the 1-sec HIGH tone, and have plenty of resolution to capture any AGC-like sensitivity increase over a longer time – say 30 Sec or so.

Here’s the code that produced the above outputs:

I made another run with the A/D resolution set to 12 bits to see if it made any appreciable difference. As can be seen in the following Excel plot – it didn’t:

Here’s another plot showing the microphone output, but this time with the DAC sinewave output amplitude reduced to the point where the microphone output isn’t large enough to clip.

In the reduced sinewave amplitude plot above, the ‘Meas’ plot is still centered about the halfway mark in the 12-bit range of values, while the average value of the ‘Amp’ plot has been reduced from about 1800 to about 200.

So, now that I know that the DAC-Speaker-Microphone-ADC loop works, I need to extend it to record amplitude values over an extended period – at least 30 sec, and more like a minute or more.

I modified my control program to create a 1-second ‘burst’ of a HIGH amplitude sinewave, followed by an infinitely long period of a LOW amplitude sinewave. The HIGH amplitude was chosen to fully clip the microphone output, and the LOW amplitude was chosen to be well above the noise floor, but still very small compared to the HIGH amplitude signal. Here are O’scope photos of both the HIGH & LOW signals:

Here is the output from the program (the LOW amplitude output was manually terminated after a few seconds):

31 March 2024 Update:

After getting all of the above working, I then installed one of my Jabra ‘Enhance 200’ aides in between the speaker and the microphone, as shown in the photos below:

With the aid installed, I got the following microphone output using my ‘burst + long-term low level audio’ setup.

Even though the Jabra aid did NOT exhibit anything like the AGC behavior I expected, there *was* a sort of cyclical response with the Jabra aid that wasn’t there without the aid in the middle. This cyclical behavior repeats about once every five seconds and *could* be some sort of AGC-like behavior – just not the one I was expecting.

Stay tuned,

Frank

Posted 23 March 2024

After a multi-year hiatus, I recently started flying contests again in the Condor Soaring Simulator. As sort of a side project, I have also been working with the XCSoar glider navigation program, to see if I could use XCSoar to help navigate AAT/TAT tasks in Condor (Condor doesn’t support AAT/TAT tasks natively with the in-sim PDA).

After using XCSoar for a while, I became frustrated with XCSoar’s inability to define ‘custom’ turnpoints based on LAT/LON coordinates, which are used quite frequently in Condor contest tasks. After a long-fought and ultimately unsuccessful battle with XCSoar’s source code to see if I could modify the program to facilitate this, I admitted defeat and decided to try another way to skin this cat. XCSoar will accept an ‘Additional Waypoints’ file, so I decided to see if I could create a program to convert the ‘new TP’ blocks in the Condor ‘Task Briefing’ description to XCSoar-compatile .CUP file waypoint lines, which could then be loaded into XCSoar for selection as task waypoints.

Here is the .CUP file format defintion page from the SeeYou (naviter) program website:

The above description is NOT very easy to read. It is full of errors, so some imagination is required to make sense of it. The ‘hardpoints’ in the description are as follows:

• Latitude strings are exactly 9 characters long. Longitude strings are exactly 10 characters long.
• In latitude strings, the decimal point is exactly the 5th character (Char4) . In longitude strings, the decimal point is exactly the 6th character (Char5).
• Both latitude and longitude strings apparently must be zero-padded as necessary to make the string character counts work out. For instance, in the longitude example the ‘degree’ value of ‘014’ must be exactly three characters.

Example Run:

Here’s a recent task briefing from Condor-Club:

As can be seen in the above screengrab, TP2-TP2 are all ‘custom’ turnpoints defined only by Lat/Lon coordinates. Manually adding these to a .CUP formatted file for use in XCSoar would be essentially impossible, given that the turnpoint coordinates only become visible 15 minutes before server start.

To start the process, the turnpoint blocks from the above briefing were copy/pasted one at a time into a text document (I use NotePad++), as follows (note that I manually changed the name of the last turnpoint from ‘Finish’ to TP 6, as my Python script currently only looks for ‘Start’ and ‘TP’ starting strings)

My CondorTPX_to_CupWP.py Python script opens a ‘FileOpen’ dialog where the input file (in this case ‘NewTPs_IN.txt’) can be selected by the user, and a ‘FileSave’ dialog where the output file (in this case ‘NewTPS_OUT.CUP’) can be selected, and then parses through the blocks in the input file, converting them to equivalent .CUP-formatted lines compatible with XCSoar. Here is the console printout from the ‘verbose’ (-v) version of the script:

At the very end of the above printout, the newly-written contents of the output file are read back out again as a verification that the conversion was successful. Here is the actual contents of the ‘NewTPS_OUT.CUP’ file:

This file now has to be transferred to the directory used by XCSoar for waypoints, and then selected in XCSoar (Config->System->Site Files) to load as the ‘More Waypoints’ selection. After this, all the above turnpoints will be available for task construction. Here’s a photo of my Android tablet with the above task turnpoints loaded:

It is clear from the above images that the Condor-Club ‘custom’ task turnpoints have been converted properly from text blocks to SeeYou .CUP format waypoint strings, so now I can use XCSoar to navigate Condor tasks with ‘custom’ turnpoints – Yay!

Here’s the Python script I created to do the conversion:

Enjoy!

Frank