For those who are trying to use an MQ135 module from e-bay or other sellers, on a small pcb with a chip, LED and potmeter:
It took me a while until I figured out that the little chip has a very big influence on the value that you read on the A-out pin.
Only after removing the connection between the A-out pin and the little chip, I got normal measurements.

I also replaced the small resistor marked "102" with a 10k resistor, to make measurements more compatible with code that uses a 10k pull down resistor. But that was already a known issue, found on many other places.

@epierre I striped-down that AirQuality-Multiple_Gas_Sensor1_4.ino to use only MQ135.
When used inside, the analogRead is 473, so a reasonble value.
But the MQCalibration function returns 0. Which means the MQResistanceCalculation also returns 0.

MQResistenceCalculation only does one calc:
return (long)((long)(1024 * 1000 * (long)rl_value)/raw_adc-(long)rl_value);
rl_value = float RL4 = 0.990
raw_adc = 473 (measured as said above)
(1024 * 1000 * 0.990) / 473 - 0.990 = 2142.266
I checked that raw_adc is filled with the correct value (to be sure it is not lost somewhere).

What looks to fix this, I changed everything to float . So:
return (float)((float)(1024 .* 1000. * (float)rl_value)/raw_adc-(float)rl_value);
Which can be simplyfied as:
return ((1024 * 1000*rl_value)/raw_adc-rl_value);
I went to float, because the function is created as float MQResistenceCalculation.

This now gives some ppm values that are within a reasonable range (about 2000 here).

Now I have to figure out how to do calibration on this method again...

Thanks, but when calibrating something strange happens when I put the module outside:

valr 270
Vrl   / Rs       / ratio  / ppm
61355 / 95673.00 / 0.00 / 399.00
valr 265
Vrl   / Rs       / ratio  / ppm
62928 / 98126.00 / 0.00 / 399.00
valr 260
Vrl   / Rs       / ratio  / ppm
64561 / 100673.00 / 0.00 / 399.00
valr 256
Vrl   / Rs       / ratio  / ppm
378 / 589.00 / 0.00 / 399.00
valr 252
Vrl   / Rs       / ratio  / ppm
1773 / 2764.00 / 0.00 / 399.00

As you can see, the valr decreases slowly, but suddenly Vlr and Rs jump over.
Vlr is the value of 'val', which is: uint16_t val = ((float)22000*(1023-valr)/valr);
uint16_t is 16bit, so max 65535. It looks like my environment needs more than that?
(in fact, your code defines the defaultro as 68550, which is also too much for a uint16?)

@epierre Thanks, I thought it would be something like that, but that didn't work.
I then got

valr 242
Vrl / Rs / ratio:5464 / 8520.00 / 0.00

Nothing like around 400ppm.

@epierre I am testing a stripped-down version of your code from your AirQuality-MQ135.ino. I stripped it to make it run just stand-alone.
What I don't understand is how to do the calibration. When I run it in outside environment (with 10k resistor), I get these values:

valr 223
Vrl / Rs / ratio:13387 / 20875.00 / 0.00

What would be the next step for calibration?