We have been running a DPA423 flyback design for more than ten years. The supply has been very robust. It has a nominal input of 18 - 36 Vdc and two outputs: 5V (+/-5%) @ .25A and 12V (10 - 18V) @ .16A. See attached schematic. In usage, they are fitting to a 24Vdc industrial supply.
In early December, a dozen units that were installed in cold climates reported that when cycling power (rare, since usually they are in an always powered on state), the 5VOUT/12VOUT power lights just blinked. After receiving a few back, we could only duplicate this condition by lowering the input voltage to 21Vdc at 25C and we found that the UV lockout had started up, but then quit and this cycle repeated: endlessly in some units OR eventually starting up after a few cycles in other units. With a 280K Ohm resistor connected to L, we expect a nominal UV of 16.4V. Instead, we found that on average, the UV voltage is 18.5 (60C) to 20.4 (-45C). Good, and old units are flat from 0C to 60C but do rise 2V at -45.
What changed? C3 in the PI filter. It was a Murata 1uF, 100V 1210 MLCC. Subs were added when that went end of life. The problem units had used AVX 1uF, 100V 1210 MLCC. The capacitances measured within 15% in all cases. ESR differences don't trend (see attached graph). On problem units, using hot air reflow on both ends of C3 solved the problem short term, but it returned after a few days. Also, lightly tack soldering a second capacitor (1uF, 100V) on top of C3 solved the problem.
One more oddity shows that the UV lockout from ON to OFF is stable and has no temperature coeficient: in all cases and temperatures, power goes off at 16.3V +/-0.6V.