I am using a blue LED from Rohm in our design.
I have gotten some suggestions from other sites, but I wanted to give you all a chance to take a look at the design.
We have been having premature LED failure within a month from the field. I am not sure what's going with the LED driver. I have attached the driver circuit/datasheet for your review.
what about Rds(on) for Q12? it's around 5-6 ohms would it be included in the current (LED)?
The continuous current will be too high for the LED but maybe you don't allow continuous current. Check that nothing can stall and allow such continuous current.
Also measure the 120Ω resistor network R59 to make sure they are not "120" (12 ohms) instead of 121 (120Ω). You may chortle, but I've seen it happen, with similar delayed field failures resulting.
Did you ever measure the LED current?
With 5V on LED_ADJ, the OP amp will drive Q10 until there is 5V drop over R59, which will happen at 41 mA. twice the rated current of the LED.
If the LM324 is powered from the 7.5 volt supply (not confirmed yet) the maximum typical output voltage could be at 6 volts but, it could be a shade higher (worst case) so maybe 6.5 volts. The volt-drop through the emitter follower might be 0.6 volts so, worst case (with some hand-waving) is that the LED and series resistor receive 5.9 volts.
Given that the LED might drop 3.3 volts when fed with 20 mA, the voltage remaining across the resistor is 2.6 volts and this implies a current of 21.5 mA so, immediately I can see how the current in the LED might be a little more than 20 mA.
However, the absolute maximum rating for the LED says 30 mA and, this should be OK for what I've calculated but...
If the LM324 is a different op-amp or, the LM324 positive power rail is greater than 7.5 volts, I'd begin to have stronger concerns.
Of course you can't immediately rule-out a bad batch of LEDs from the proper supplier or, even fake parts from a dodgy supplier.
I would imagine the steady-state operation is behaving exactly as it's designed. My take on this problem is oriented toward transients - specifically Q12 and possibly DAC changes. When Q12 is off, your integrator is going to saturate because the load is open circuit. When Q12 conducts again, it will do so, initially, with the full rail across the LED until the op-amp falls into regulation. The time the circuit takes to stabilize is proportional to the R57*C51 product. A similar issue could happen with setpoint changes from the DAC (only to a lesser degree especially if the setpoint delta is small).
What I would do, put a known 120 ohm resistor in place of the LED, and put a scope across it and monitor for spikes during Q12 transitions and DAC changes. If you know for sure the 120 ohm resistor in the circuit is truly 120 ohms, you can use that as a current pickoff while the LED, or the position of the LED, is shorted.
Other issues could be if you have long leads from the driver to the LED. The added inductance and capacitance from long leads make regulation performance worse. Also, the leads could act as an antennae "spiking" the LED by picking up radiated emission from nearby EMI generating devices. If these are field failures, you never know what these people we call customers are doing out there. In that case, you would need a reverse rectifier or TVS across the LED locally. This is a common protection technique used for diode lasers.
