LM317 output voltage drop out of spec (maybe)

[wraper]

Try soldering just a single resistor between ADJ pin/GND instead of that switch/resistors and disconnect that switch connector. You may also try removing R3 and soldering it right to LM317 pins. In any case, long wires and a bunch of things dangling in feedback loop is not good.

[francesco_97]

Nothing new, I soldered the 120R resistor directly on the LM317 pin and I've got the same behavior. I'll post the new layout here in the next few days. Any idea before I definitely drop this one?

[David Hess]

Nothing new, I soldered the 120R resistor directly on the LM317 pin and I've got the same behavior. I'll post the new layout here in the next few days. Any idea before I definitely drop this one?

Then I do not know what you are doing wrong.  I have made current boosted regulators, that with the proper layout, handled 10 to 20 amps with excellent load regulation.  Voltage drop was completely explained by wire resistance.

Current boosted designs have the advantage of routing most of the current around the regulator, but I always ran the regulator at close to maximum current anyway for best stability and thermal protection.

Is the regulator genuine?  Maybe the regulator is defective.

If load regulation must be better, or at precision levels, then a different circuit like the single operational amplifier remote sense is required.  I like a modern implementation of the circuit shown below, which also replaces the integrated reference with a better one,

[francesco_97]

Now that i'm more relaxed i'm thinking about one thing that i've could missed. To draw the right amount of current i used two 4.7R resistors connected in series. They are rated to 17W but MAYBE when they turn hot their resistance changes going higher and lowering somehow the voltage. Below the photo of the resistor. But if this is the case how is possible that changing the resistance towards the high the voltage drops?

Moreover i've replaced the IC with a new one but now it acts stranger than before because at 60° it lower the current down to half an amp amd the voltage to 4.5 volts.

Edit1:
Could it be the mechanical connection between the heatsink and the IC? In order i used a screw, IC, thermal pad rubber, heatsink plastic insulator, metallic bolt. In this way the IC is insulated from the pcb but maybe it has a connection with the heatsink and heating up it changes somehow the voltage with it.

[PCB.Wiz]

Now that i'm more relaxed i'm thinking about one thing that i've could missed. To draw the right amount of current i used two 4.7R resistors connected in series. They are rated to 17W but MAYBE when they turn hot their resistance changes going higher and lowering somehow the voltage. Below the photo of the resistor. But if this is the case how is possible that changing the resistance towards the high the voltage drops?

You would see any load changes on your current meter.

Moreover i've replaced the IC with a new one but now it acts stranger than before because at 60° it lower the current down to half an amp amd the voltage to 4.5 volts.

Those values are nuts. It's either oscillating, or going into thermal overload.

Try a regulator only on a heatsink, (no PCB complexity) with minimal parts directly soldered, (two measured resistors, and two caps) and check it at 100mA 300mA 600mA 900mA.

Proper cooling is going to be a challenge, How large is that heatsink ?

This data sheet has some measured plots
https://www.lcsc.com/datasheet/lcsc_datasheet_2304120930_Shanghai-Siproin-Microelectronics-LM317T_C5444279.pdf

[David Hess]

Proper cooling is going to be a challenge, How large is that heatsink ?

A 20 watt finned passive heat sink is about the area of a small smartphone.

[dietert1]

In our lab i would find some fan to test whether it's about voltage regulator temperature.
I have seen many, many designs with overheated voltage regulators, e.g. vertical TO-220 regulators without heat sink and consuming 1 W or more, e.g. a 5V voltage drop * 200 mA. One could even consider the temperature fail safe voltage regulator a design trap. At least the OP noticed that something was wrong. The problem is easy to spot using a thermal camera.
A ground plane isn't possible on a one-sided PCB. Those fills may help for cooling, but not against oscillation. What you can do is add wire bridges to "shorten" the distances across some of the signal traces cutting through the plane, but why not a double sided PCB?

Regards, Dieter

[francesco_97]

Down you can find the photo of the board with a ruler to see the heatsinks. I'll try first the minimal configuration suggested with the components directly soldered and to eliminate any issue with the load resistors i ordered from amazon a cheap 60w dc load. If everything goes well i must go out from my laziness and make a double sided pcb but...
UFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF
IT'S SO HARD WITH THE CNC
But i'll do and this time star design with the ground everywhere.

[dietert1]

Just use double-sided material and CNC the bottom side as before. After drilling the holes you need to enlarge them on the Gnd plane side so you don't get unintended shorts when inserting the parts. Gnd connections get soldered on both sides - easy for resistors, transistors and diodes. For sockets and electrolytical caps just solder one side as before.
Of course an industrial PCB would be much better, with coating and thermals around the Gnd pins (easier to solder).

Regards, Dieter

[David Hess]

A two sided board is not required to make a good layout for a 3-terminal regulator.

[francesco_97]

Yes but here we have two regulators and all the resistances. I was thinking about replacing them with SMD 8085 to save a lot of space and to maintain the single layer of traces with the second layer for GND. Could it be a good idea?

[wraper]

Yes but here we have two regulators and all the resistances. I was thinking about replacing them with SMD 8085 to save a lot of space and to maintain the single layer of traces with the second layer for GND. Could it be a good idea?

The best idea is throwing out all those wires and using series resistors in voltage divider soldered directly on the switch or placing them on another PCB that is soldered to the switch, thus reducing wire count by 5x

[francesco_97]

Ok, first attempt of the day.
I used a minimum configuration with only a 120R, a 1K pot to set the voltage, the same IC and the same two 4.7K resistors used yesterday. Everything directly soldered each other.
Same results:
Open circuit 9.0V
Closed, 30°: 8.9V 0.95A
Closed, 80°: 8.0V 0.85A

You can find the setup below: the left red wire is the thermocouple going to Peaktech 2040, crocodile connectors on the right coming from the switching PSU set at 21V and reading the current and hooks in the middle going to the HP3478A to read the voltage.

Maybe with the full setup and all the bypass capacitors the situation gets better?

[David Hess]

Maybe with the full setup and all the bypass capacitors the situation gets better?

With those wire lengths, at a minimum the input ceramic/film capacitor is required to prevent oscillation.  A small output electrolytic capacitor should be used also.

[wraper]

You're not supposed to measure temperature like that. Actual temperature should be 1.5-2x of what you measured. Touching something with thermocouple junction does not magically make it heat to the same temperature. Also your LM317 is a fake ST imitation. Regular silicone pads are quite poor thermal conductors while being quite thick. Mounting directly on a heatsink with thermal paste can significantly lower the temperature when heat dissipation is high.

[PCB.Wiz]

Ok, first attempt of the day.
I used a minimum configuration with only a 120R, a 1K pot to set the voltage, the same IC and the same two 4.7K resistors used yesterday. Everything directly soldered each other.
Same results:
Open circuit 9.0V
Closed, 30°: 8.9V 0.95A
Closed, 80°: 8.0V 0.85A
..
Maybe with the full setup and all the bypass capacitors the situation gets better?

You should add caps for any test, but if that only changes with temperature, you have your answer.

I did find this in another thread on here - interesting comments re the venerable LM317, under 'extreme' stress conditions 

Next up is ambient temperature. This thing needs to pass a building fire test. That means board temperature in excess of 200 degC.
Good news is, the unit has to stay alive only for the first 5 minutes of said test. Yes this is a run to destruction application. Any thermal protections schemes kicking in at sub 200 degC does not stand a chance of passing the test.
The old LM317 series regulators does throttle the maximum output current available once they heat up, but they never shut down to zero -

You have many watts there, so it looks like it is simply doing the thermal reduction thing.
Another low tech test : Does it hiss with a wet finger ?

[GigaJoe]

I have good news   reeeally good news

that LM317 are fake ... 
That supposed to STM LM317 , are fake LM with unknown crystal inside ..

I have a lot experience with 317 from aliexpress ))  , how about LM317 , TO220 with 0.1A limiter    probably encapsulation cost more,  it has a thick pure copper sole btw

for a future look at engraving of original STM, that font  kinda unique ,  C  as line , not oval , "1"  tail  ... etc ...   STM never encapsulated TO220 317 in MAR ( morroco?) , only CHN
https://i.ebayimg.com/images/g/N9gAAOSwWkJY8nRQ/s-l1600.webparbage

your are garbage ...   good for magic smoke

[francesco_97]

Sorry guys for my silence, I'm busy with exams. Thank you all for the suggestion and for the effort. I did again the measures with the added caps and the IC now acts weirder raising the temperature almost in the first 3 seconds at 90° so FAKE CONFIRMED. I liked the comment about the thermal paste, is it really better than the thermal pad? If so I'll use the pc paste. Now I must find the original LM317 and this is a problem here in Rome. There are only 4 physical stores and the salesman's kindness in giving information about the components is astonishing, practically you have to get what they give you and SHUT UP. Excluding the option where I suffocate with my hands the salesman, how can I see at a glance that the IC is trustworthy?

[francesco_97]

Finally something near to the specs.

Today my new shitty dc load arrived from Amazon (60W dc load ZPB30A1). I'm still alive but I don't know why.
Using the LM shown below I obtained:
Open circuit 30°: 9.004V
80° 1A: 9.124V

Using the datasheet value of 1.5% load regulation at 9V we have a maximum ±0.135V error so we are in the specs range.
The only strange thing is the temperature because with the thermal pad now the temperature rises immediately so maybe it is cooked. Without anything as interface, I obtained the results listed so I probably have to switch to thermal paste.

In a few days the will arrive the results with the IC mounted on the PCB that you have seen to check if it's enough for this project or if the layout MUST be changed anyway.