BasicPI

One of the challenges I face is that I would like to start producing electronics in volumes, but assembling at an factory is expensive. It easily add 50-100.- USD to unit cost at low volumes and I don’t have the financial backing to go high volumes.

One alternative is to use an Asian factory, rather than the expensive Norwegian ones. They differ a lot on low volumes. The other alternative is to invest in equipment for low volumes yourself. Acceptable Pick & Place machines and Stencil Printer solutions exist from ca 4000.- USD. It is specially 2 machines that caught my attention: CHMT36VA and TVM802A. These are close to identical in functionality and price.

The pictures above is CHM-T36VA and CHM-T48VB. The difference is that the later have PC embedded (Linux) and 2 x as many feeders (58) and cost ca 5000.- USD compared to 3200.- for the smaller one. CHM-T36VA need an external PC. CHM-T48VB really is a good deal.

The key with these machines are that they have cameras capable to calibrate components giving them a much better accuracy that previous low end solutions. Their limit is that they only have 2 heads and 27-29 component feeders. So for more advanced electronics you will need to swap reels and nozle’s etc. They are also decently noisy and somewhat slow, but they will do the trick.

My reality is that I spend a huge amount of hours soldering proto-types and these machines can do that much faster and more reliable where I only solder on hole through and special components myself. And once the machine is set up I can easily produce a decent number of boards close to automatically. More important is that it will enable me to work with BGA style components.

Another cost is components. I basically need to invest in a decent stock of reels, meaning I will need a much larger stock of components. IC’s are different as the machines can pick them from trays and you can buy lower volumes. But, it is doable.

The design of a low cost Pick & Place machines consist of several main components you need to evaluate.

• The CNC frame with various axis and tools. This needs to be low noise as well as very accurate. The size/speed also decide prod volume.
• The feeders enabling the machinery to pick up components. You want as many of these as possible and a combination of Reel, Tray and Tube-feeders.
• Calibration cameras. Early machines did not have these and did decently well, but with camera’s you get automatic calibration on Components and avoid errors that was common in early machines.
• Placement heads. The smaller machines have 2 heads, but you really could need more as different nozzles are needed for different components.
• Control system on these are a combination of sensors, stepper motor drivers, camera inputs and usually a PC level computer to add Logic.

The challenge is that increasing number of feeders and heads also increase investment cost. I think 3.200,- USD is as far as I am able to go for now and I really like CHM-T36VA + I have plenty of PC’s and screens around.

I do at present have 8 PCB’s + 3 designs I have not ordered yet in my modular control system, so I have a far to large backlog of electronics that need testing and code. The difference with my last 12 x PWM Hat is however that it target a project – my upgraded 3D printer.

I purchased a Prusa i3 which basically is a clone of a generation 1 design and I want to use my modular control system on this one. I actually have 3 Arduino based control systems of which 2 used a fake FTDI chip that caused problems. But, I want to add features not available in the old control system.

Multi-material support means more than one extruder. This is mostly a mechanical re-design, but I need extra Stepper motor controller and logic to handle it.

Automatic calibration. Those of you who have used an early version of a 3D printer know how much work that goes into calibrating and ensuring that the printer is correctly configurated. This can be done automated by sensing the tilting of the plate using a “Z-Sensor”.

Torque based end-stops means that I do not use end-stop sensors, but sense current increase as the CNC machinery hit mechanical stoppers. This is goodbye to a lot of problems and wiring.

In the illustration above I assume I will use 5 modules to make a modular control system. One Wifi module that can be Raspberry PI or ESP32 based, a XPortHub to access USB, MMC and HMI. MMC can be used to spool jobs. A 12 X PWM used as 4xStepper to control X,Y and 2xZ Steppers. A 2nd 12 x PWM to control 2 x Extruder feeders and 3 x temperature heaters for heatbed and extruders, and finally a 32 x IO to get temperature sensors and a Z-distance sensor. This is just an early illustration.

The Z-distance sensor I want to use is a small laser that can sense the distance from Extruder to the plate. This can during calibration sense the tilting of the base plate and adjust G-Code parameters so it get correct.

This is an excellent test bed for my modular control system. It is so many excellent All-In-One systems for 3D Printers and my favorite is the new MK3 from Josef Prusa so we could have done this much easier. But, I need a test bed and this is an excellent start to drive development.

I decided to Change the 12 x PWM Hat to use High Side Current Sensors. Below is the schematics of the new sensors. This means I measure current out on each PWM signal.

Using high side sensors actually simplified the routing so it was a straight forward change. The reason I started with low side sensors is because that is straight forward for DRV8313 and more common for 3-phase. It would have been ok for Stepper and DC motors, but it made it difficult to measure current if you used a single PWM channel stand-alone.

The 2 blue signals crossing the bigger, red Power lane is asking for trouble. The blue lanes are 0-3V ADC signals, while the power lane is feeding PWM MOSFET’s. What will happen is that PWM noise from current spikes will jump over to ADC signals and create a false current signature.

But, I believe it is to my advantage that I have a fast MCU and no electronic filtering, meaning I can apply smart filtering in SW using the high sampling frequencies available on the ADC. Well, we will see. But, I have promissed myself that I will look into 4-layer PCB designers after this one.

I must admit that sourcing for STM32F405RG is an issue. I have bought batches from China and I am rejecting a lot of MCU’s. The worst is actually the time to solder them on before I can test, so I seriously need the tester shown in my last article.

This is worst as I like now deal with a new board. I have soldered on 3 different MCU’s that all show up with failing memory sectors, and it’s a lot of stress on the PCB so you can only solder on/off so many times. After 3 fails on a new PCB I also start questioning other possibilities – do I as an example have a bad SWD on this board? SWD is a bit sensitive so if you have a bad SWD connector it might show up in various forms.

I managed to find a good tester for 44.- USD, so I am looking forward to that one. As much as I would like to test new boards I don’t want to waste more time on MCU unsecurity, so I will wait for the tester. I want to know that my MCU is good before I put it on.

I do however still need to sort out sourcing for STM32F405RG because this is unacceptable. But, with a tester I can actually test the batches I receive and slam down on bad sellers instantly to get my money back. I don’t expect the MCU’s to be perfect, but I expect them to work. This is prototyping and you can’t use my sources in production. But, I am getting rather pissed of with fraud schemes of selling factory rejects through Aliexpress! STM32F405 is specially bad for some reason.

I do however see that Farnell/Element14 sell this MCU for ca 9.- USD each in quantity of 10 delivered “next day” inside Norway – it is 2x my current cost, but it is tempting because of the time I waste on this.

Finally – I just discovered test boards with sockets for STM32 LQFP64 and LQFP100. They are a bit pricy – ca 65.- USD yet, but it is a start. These are on the list of things I planned to make because I could not buy them.

I currently plan using INA194 that have a 50x gain. As our max voltage in is 3.3V this means the max voltage over the shunt need to be 3.3/50= 66mV. Given a max current of 2.5A that gives a Shunt resistor R = V/A = 0.066/2.5 = 0,0264 R or 25mOhm.

Now – 2.5A over 25mOhm is 0,156W which is fine since the 1206 resistor is 250mW.

If I use 50mA that gives 0,00125V (1,25mV) into INA194. And with a gain of 50x that is 62,5mV into the MCU. This is where we get into trouble because INA194 starts being inaccurate below 50mV in. I am not at all sure I actually can measure these low currents, but I do have one SW trick to help me.

We are NOT measuring a constant current, we are measuring a PWM pulse current, and the current we will see is 2.5A for 1mS and 0A for 999mS (just an example) due to the nature of the PWM. In software that becomes 2,5A / 1000 = 0,0025mA in average over a sec. So, if I take advantage of the 2,5Msps capability on the STM32 M4 I should be more than ok. I just need to make sure my sample frequency is at least 2 x the PWM frequency, but the higher sample frequency the better in this case.

Yet another trick is that I know then the Pulse is On/Off, and since we have raw data (no electronic filtering) I can eliminate noise then the pulse is off – as I simply know there should not be anything there. This is more difficult to do than I write here, but I will dig into it.

I used 2.5A as an example, but pulse current is V/R and can be higher that the rated max. This is why some MOSFET’s are rated to 160A with a pulse current to 400A etc. If we have motors with low inductive R we reduce PWM duty to maintain 2.5A in average, so I might actually need to cut down even further on the Shunt to avoid clipping on pulse currents, but I will not worry about that for now. I want to experiment with actual values because you can only cover so much in theory. Noise etc also add to the picture and is not easy to predict or model accurately.

The irony is that I removed passive filter components because of lack of space, and those filters would have prevented us to see the pulse currents. Now – since I have raw noise in I can actually use SW and see far more accurately than I would have done with analogue electronics. This is an excellent example where you need a good interaction between a HW and SW engineer.

Left side of the 3-Phase/Stepper/PWM Hat done. This is the hardest bit, but it went much easier this time because the previous attempts gave some hints and lessons learned. I am far from finished, but it is not hopeless. My main concern is the ground-plane getting to fragmented and unable to support the currents – but, I am not out of tricks yet.

This picture show the back-side and the ground input is actually the “2” in the left, bottom corner while DRV8313 are on the inside of that thick 36V power lane. While Ground is connected you can see that it is thin on some places due to the density on the left side. it will be equally dense on the right side, so what I am considering is to add a bridge on top lane (illustrated in light blue) to ensure that I have a direct path that will support the currents I need.

I must admit that doing PCB routing is a bit like a puzle. You need to like these mind games if your going to be good at this. I know many electronic engineers who seldom do PCB routing – only schematics.

I have suffered “Crash O’ Mighty” on this Board a lot. I use a free version of a Target 3001 EDA based on Java and well – it does crash a lot. I am also more and more feeling the pain and limitation of 2-layer designs, but I will need some time to get started with KiKad.

What worries me a little is the signal integrity of the current sensor signals. But, it is nothing I can do about that. The issue is that the ADC signals need to pass strong PWM signals, so that current on one PWM might give a false signature on a 2nd PWM’s current sensor. I do however have the option to filter and compensate for this in software, so lets see. I do expect noise on these signals regardless.